On 23 June 2026, Accenture announced Accenture Edge. A new business unit. A new go-to-market. A deliberate move downstream into companies with revenues between $300 million and $3 billion.

The official language was measured. Accenture Edge “brings the depth of expertise Accenture has built serving the world’s largest companies to deliver business solutions specifically developed for this market.” Julie Sweet, Accenture’s CEO, called it a $240 billion opportunity growing at high single digits.

Read that number again. $240 billion. That is the market Accenture has just formally declared it wants.

It is also, almost certainly, the market your clients sit in.

What Accenture Edge Actually Is

Strip away the announcement language and Accenture Edge is three things.

It is a productisation play. Accenture has spent decades building platforms, accelerators and delivery assets for enterprise clients. Edge is the attempt to package those assets into something repeatable enough to work at mid-market price points and timelines. Partner-led, pre-configured, faster to deploy than a bespoke engagement.

It is an ecosystem land grab. The partner list — Microsoft, Oracle, SAP, AWS, Google, Salesforce, Workday — tells you the shape of the offering. This is platform-driven transformation, not methodology-led consulting. Accenture is the integrator. The technology vendors are the product.

And it is a structural response to an economic reality. AI has changed Accenture’s unit economics. Delivery that once required large teams now requires smaller ones with better tooling. The mid-market, previously uneconomical to serve at scale, is now within reach. The $240 billion TAM did not appear overnight. Accenture’s ability to go after it profitably did.

What This Is Not

Accenture Edge is not a consulting firm. It is not relationship-led. It will not sit with a client for 18 months helping them think through the hard stuff. It will not carry institutional knowledge from one engagement to the next in the way a trusted adviser does.

It is a delivery machine pointed at a specific problem set: modernise core systems, adopt AI, tighten security, streamline operations. If your client’s challenge fits that template, Accenture Edge will compete for it. If it does not, Accenture Edge will have limited to offer.

That distinction matters. But it should not make you comfortable.

The Real Question It Raises

The arrival of Accenture Edge in the mid-market is not, in itself, an existential threat to independent consulting firms. Accenture has always competed somewhere. What it does is make a question harder to avoid.

How do you operate?

Not what do you do, or who do you know, or what methodology do you carry. How do you actually run the firm? How do you capture and retain institutional knowledge between engagements? How do you maintain consistency of delivery across a portfolio of clients? How do you show a prospective client — one who is now also being courted by a firm with Accenture’s brand, ecosystem and productised assets — that working with you produces measurable, repeatable outcomes?

Accenture Edge is not coming for your relationship. It is coming for your operating model. And if your operating model is still held together by spreadsheets, email threads, and the knowledge that lives only in your head, the comparison will not be flattering.

IBM Said It First

This is not the first signal.

In February 2026, IBM published its thesis on what it called the “consulting operating engine.” The argument was simple: the firms that will survive the next decade of AI-driven disruption are not the ones with the best ideas. They are the ones with the best infrastructure beneath those ideas. Defined processes. Structured knowledge. Systematic delivery.

IBM was talking about large consulting firms. Accenture Edge extends that logic downstream.

Two of the world’s largest technology and consulting organisations have now pointed at the same gap, in the same market, within months of each other. The gap between what the best-run firms look like operationally and what most independent firms actually are.

That gap has a name. It is the operating gap. And it has just become considerably more visible.

What Mid-Market Companies Should Take From This

If you are a COO, CTO or transformation lead at a mid-market company, Accenture Edge is worth understanding — and worth interrogating before you engage.

The productised model has real advantages. Speed to deploy. Proven assets. A brand that carries weight in board conversations. If your challenge is well-defined, bounded, and maps cleanly onto a platform modernisation or AI adoption programme, the Edge model may serve you well.

But productised delivery is, by definition, built around the common case. If your organisation’s challenges are more complex — legacy culture, unclear strategy, competing priorities, capability gaps that predate technology — a delivery machine optimised for repeatability may move quickly to the wrong destination.

The best consulting relationships are the ones where the adviser knows your business well enough to tell you something you did not already know. That takes time, presence and genuine institutional knowledge. It is not a product. It cannot be packaged.

Ask the question before you sign: is this engagement designed to understand my business, or to deploy a solution built for someone else’s?

What Independent Firms Should Do Now

The window is not closing. But it is narrowing.

The mid-market is large enough that Accenture Edge and a well-run independent firm can coexist in it. Accenture will pursue the platform-adjacent, technology-driven work. The relationship-led, methodology-driven, organisationally complex work remains, for now, the territory of the independent.

But “for now” is doing significant work in that sentence.

The firms that will hold that territory are the ones who close the operating gap before Accenture makes it the default comparison point. That means systematising knowledge. It means building repeatable delivery frameworks. It means being able to demonstrate, with evidence, that your firm produces consistent outcomes — not because you are talented, but because you are structured.

Talent is not a differentiator at scale. Structure is.

Accenture Edge just made that argument for you. The question is whether you use the signal before the window closes, or notice it after it already has.

The Bigger Picture

We are in a period of rapid consolidation in how consulting is delivered, resourced and perceived. AI has accelerated it. The large firms are moving faster than most independent operators realise. The mid-market — long a refuge for boutiques, independents and specialists who could not be touched by the enterprise machine — is now firmly in play.

That is not cause for panic. It is cause for clarity.

Know what you are. Know how you operate. Know what you can demonstrate that Accenture Edge cannot. And build the infrastructure that makes that demonstrable, consistently, at every engagement.

The market just got a $240 billion spotlight shone on it.

The firms that were already ready will find that useful.

The ones that were not will find it uncomfortable.

Howard Scott is the founder of Studio:Blueprint, a consultancy operating system for consulting firms. He writes about transformation, AI and the future of professional services.

Visit studio:blueprint to find out how your consultancy can be quickly become AI enabled for you and your clients.

EY’s Global AI Sentiment Survey gathered responses from more than 18,000 people across 23 markets. The headline finding sounds like a contradiction.

People say they don’t trust AI. Two thirds worry about AI systems being hacked or breached. Seven in ten believe human oversight remains essential. Six in ten worry that organisations will fail to hold themselves accountable for AI use that leads to negative consequences. Nearly three quarters worry that as generative AI becomes more widespread they will no longer be able to tell what is real or fake.

And yet. Sixteen percent of people globally have already used AI that acts on their behalf without human intervention in the past six months alone. Ten percent have used AI agents to buy products on their behalf. Eleven percent let AI manage their finances and carry out banking tasks without intervention. Twenty six percent use AI to ask about symptoms without seeing a doctor, up from nineteen percent last year.

People are delegating consequential decisions to AI systems they say they do not trust. That is not a contradiction. It is a signal.

Adoption is racing ahead of declared confidence

EY describes this precisely. AI adoption is advancing faster than public confidence. People continue to integrate AI into daily decisions, gradually relying on it for more consequential tasks, while simultaneously asking for clearer safeguards, stronger accountability, and greater transparency.

This is not irrational behaviour. It is the same pattern that played out with internet banking, with ride-sharing, with every technology that became mainstream before the governance structures caught up. People adopt because the convenience is real and the immediate downside seems manageable. The systemic risk accumulates invisibly until something goes wrong at scale.

EY identifies what they call Pioneer Markets where AI adoption is most advanced. In those markets, 94% of people report using AI and nearly a quarter have used autonomous AI. The markets they identify as Pioneer include India, China, Saudi Arabia, the UAE, Mexico, and Brazil. The UK and the US sit in the Transitional category. The UK is not leading this shift. It is watching it arrive.

For consultants advising UK and European organisations, this is the context that matters. Your clients’ customers, partners, and competitors in Pioneer Markets are already operating at a level of AI delegation that most UK boardrooms have not yet seriously considered. The governance gap is not a future problem. It is a present one that is becoming visible in slow motion.

The trust paradox is the consulting brief hiding in plain sight

Here is what the EY data is actually describing for anyone advising organisations on AI adoption.

Organisations are deploying AI. Their customers, employees, and stakeholders are using AI autonomously whether the organisation planned for it or not. Nobody has fully designed the governance structure for what happens when it goes wrong. And the people affected by these systems are asking, consistently and across every market, for clearer rules, stronger accountability, and visible protections.

That is not a technology brief. It is a consulting brief.

The organisation that needs help is not the one asking how to deploy AI faster. It is the one sitting in the 70% of Pragmatists BCG identified, watching AI adoption accelerate around them, aware that their customers are using autonomous AI in ways they did not design for, uncertain whether their governance structures are adequate, and looking for a structured path forward that reduces the risk of committing before they understand what they are committing to.

EY puts the implication for leaders clearly. The question is no longer whether autonomous AI will scale. It is whether organisations will shape that scaling intentionally, accelerating where trust and value already exist, and slowing where clarity, safeguards, or confidence are still needed.

Shaping that intentionally requires knowing where you stand. That is what a diagnostic does. Not a generic AI readiness assessment copied from a framework someone else designed. The consultant’s own methodology, built from their specific sector expertise, deployed as a scored assessment that produces a concrete picture of where the organisation sits and what it needs to change.

The moment between assistive and autonomous is the billable opportunity

EY describes a progression. Low-risk, everyday AI use builds familiarity and confidence. That familiarity normalises delegation. Delegation expands into higher-stakes decisions. By the time an organisation realises their customers or employees are operating in autonomous territory, the governance structure needed to manage it is already overdue.

The consulting moment is the space between assistive and autonomous. The organisation that is still using AI for document drafting and meeting summaries is not yet in the danger zone. The one whose customer service function is being partially managed by agents, whose procurement decisions are being influenced by AI recommendations, whose HR processes are beginning to incorporate automated screening, that organisation is in the transition and may not have designed for it.

That transition requires a structured assessment of where AI is being used across the organisation, what decisions it is influencing, what governance exists around those decisions, what evidence trail captures what the AI did and why, and what the human override mechanism looks like when something goes wrong.

Studio:Blueprint was built to support exactly this kind of structured advisory engagement. The Forge builds the diagnostic from the consultant’s own framework. The Evidence Canvas tracks what has changed against what was assessed. The Delivery Blueprint exports a structured plan that connects the diagnostic findings to a sequenced programme of work. And the reassessment loop measures whether the governance improvements held six months after the engagement ended.

EY found that trust is built through experience, not through principles. The same is true of the consulting relationship that helps an organisation navigate the transition from assistive to autonomous AI. The consultant who arrives with a structured methodology, a scored diagnostic, and an evidence framework that proves progress over time is not selling a trust-and-governance workshop. They are building the operational infrastructure that makes AI adoption defensible.

The trust gap EY describes is real, consistent across every market they surveyed, and growing as autonomous AI expands into higher-stakes decisions. The consultants who close it systematically will define the next decade of AI advisory work.

The clients are already looking for them. The EY data proves it.

Howard Scott has spent 30 years in consulting, martech, and digital transformation across firms including GroupM, Digitas LBi, TEQUILA/TBWA, Leagas Delaney, The Marketing Store, and The National Trust. He is a former BIMA Non-Executive Director and the founder of Studio:Blueprint, a consulting operating system that lets consultants and agencies codify and deploy their methodology through AI-powered diagnostic and evidence tools. He is author of SynKraken, a local-first command deck for AI agents.

Subscribe now

McKinsey surveyed approximately 500 organisations for its 2026 AI Trust Maturity report. The average maturity score across the five dimensions they measure, strategy, risk management, data and technology, governance, and agentic AI controls, was 2.3 out of 4. Up from 2.0 last year. Progress, on the face of it.

Then they break down the dimensions individually.

Only about 30% of organisations reach a maturity level of three or higher in strategy, governance, and agentic AI governance. That means 70% of organisations deploying AI agents, systems that are now capable of taking actions, sending communications, making purchases, modifying systems, do not have mature governance structures in place to control what those agents do.

McKinsey calls this the shift from systems saying the wrong thing to systems doing the wrong thing. It is not a theoretical risk. It is a governance failure waiting to materialise at scale.

The awareness gap is the most dangerous finding

McKinsey identifies something more troubling than poor governance. Across almost every AI risk category, organisations report a meaningful gap between the risks they consider relevant and those they are actively mitigating. They know the risks exist. They are not building the controls.

The gap is especially pronounced for intellectual property infringement and personal privacy. These are not obscure edge cases. They are the precise risks that enterprise clients will ask about when a consultant proposes deploying AI across their operations. And the data says that most organisations are aware of these risks and are not yet doing the structural work to address them.

This creates a specific problem for consultants. The client who asks whether an AI-assisted engagement is secure, governed, and accountable is not being difficult. They are asking the right question. And the consultant who cannot answer it with reference to a documented methodology, a structured evidence framework, and a clear human-in-the-loop architecture is not ready for the enterprise conversations that are now standard.

The McKinsey finding that organisations with explicit accountability for AI achieve maturity scores of 2.6 while those without clear accountability score 1.8 is the sharpest number in the report. Accountability is not just a governance principle. It is the single most predictive factor in AI maturity. And accountability requires structure. It requires a documented framework that defines who is responsible for what, what the AI is permitted to do, what requires human review, and what evidence trail exists to audit the decisions that were made.

Confidence in incident response is declining as agents become more capable

Here is the finding that should concentrate the mind of every consulting firm deploying AI on behalf of clients. The share of organisations reporting AI-related incidents has remained steady at roughly 8%. But perceptions of incident response quality have deteriorated significantly. Almost 60% of respondents who experienced incidents report satisfactory or negative views of their organisation’s response.

Incidents are not increasing. The ability to respond to them appropriately is getting worse. As AI systems become more capable and more autonomous, the complexity of what can go wrong is growing faster than the institutional capacity to handle it.

This is the governance gap in practical terms. Not a regulatory checklist. Not a compliance exercise. A structural inability to understand what the AI did, why it did it, and what should happen next when something goes wrong.

McKinsey is direct about the implication: organisations can no longer concern themselves only with AI systems saying the wrong thing. They must contend with systems doing the wrong thing. Taking unintended actions. Misusing tools. Operating beyond appropriate guardrails.

The consultant who deploys AI tools into a client engagement without a governance structure for those tools is creating exactly this risk for their client. And increasingly, clients know it.

The accountability structure is the consulting deliverable

McKinsey concludes that building trustworthy AI requires a concerted combination of policies, processes, people, and technology. These capabilities cannot be bought and installed. They have to be built. And building them requires someone who understands what good looks like, can assess where the organisation currently sits against that standard, and can design a programme of work that closes the gap with evidence that the gap is actually closing.

That is a consulting engagement. Not a technology implementation. Not an AI tool rollout. A structured advisory process that establishes current maturity, defines the target state, sequences the actions that close the distance, and tracks evidence of progress over time.

The irony of the McKinsey findings is that the governance gap they describe is also the opportunity they are implicitly pointing at. Seventy percent of organisations lack mature AI governance. Knowledge and training gaps are the leading barrier to responsible AI implementation, cited by nearly 60% of respondents. AI trust is increasingly viewed as a business enabler rather than a compliance exercise.

Those three findings together describe a market: organisations that know they need to govern AI properly, know they lack the expertise to do it, and are ready to treat it as a strategic capability rather than a box-ticking exercise.

The consultant who arrives with a structured methodology for AI governance maturity, a scored diagnostic that establishes the current state, a roadmap that sequences the work, and an evidence framework that tracks progress, is not selling governance consulting. They are selling the infrastructure that makes their client’s AI investment defensible, auditable, and scalable.

The governance gap is real. The consultants who can close it systematically, with evidence, will define the next phase of AI advisory work.

Howard Scott has spent 30 years in consulting, martech, and digital transformation across firms including GroupM, Digitas LBi, TEQUILA/TBWA, Leagas Delaney, The Marketing Store, and The National Trust. He is a former BIMA Non-Executive Director and the founder of Studio:Blueprint, a consulting operating system that lets consultants and agencies codify and deploy their methodology through AI-powered diagnostic and evidence tools. He is author of SynKraken, a local-first command deck for AI agents.

Subscribe now

Workforce access to AI has expanded by 50% in a single year. Investment is surging. Confidence is growing. Eighty-four percent of organisations are increasing their AI budgets.

Then you read past the headline.

Only 25% of organisations have moved 40% or more of their AI experiments into production. Among workers who have access to AI tools, fewer than 60% use them in their daily workflow. Thirty-seven percent of companies are using AI at a surface level with little or no change to underlying business processes. And 84% of companies have not redesigned jobs or the nature of work around AI capabilities.

The investment is real. The transformation is largely theoretical.

The proof-of-concept trap has a name and a mechanism

Deloitte describes it precisely. A pilot can run with a small team in a few months using cleansed data in an isolated environment. Production deployment requires infrastructure investment, integration with existing systems, security reviews, compliance checks, monitoring systems, and ongoing maintenance. Each of those demands significantly more resources and coordination than the pilot did.

But the deeper problem is not the technical gap. It is the strategic one. A healthcare AI leader quoted in the report puts it plainly: “If there is no coherent AI strategy in organisations, you are likely to see pilot fatigue. You’re chasing the next shiny object, pressured to do something with AI without a real plan. I’ve seen many instances where people embark on pilots, but when asked how they’ll scale up if successful, they often don’t have an answer. Without a clear roadmap, executing a hundred pilots just leads to poor results and failed value creation.”

One hundred pilots without a roadmap. That sentence describes the current state of AI in most enterprises.

The trap has a specific mechanism. Pilots are relatively low cost and lower risk. They produce positive results in controlled conditions. They are easy to approve and easy to celebrate. So organisations continue funding new pilots rather than facing the harder work of scaling up existing successes. The pilot becomes the destination rather than the starting point.

Access is not activation

The gap Deloitte identifies between access and activation is the most important finding in the report. Sixty percent of workers now have access to sanctioned AI tools. Fewer than 60% of those workers use them daily. That is not an adoption problem. That is a design problem.

Access without a clear methodology for how AI integrates into specific workflows produces exactly what Deloitte documents: tools that are available but not used, investments that are made but not activated, potential that is present but not realised.

The organisations closing this gap are not doing it by buying more tools or running more pilots. They are doing it by redesigning work. Deloitte’s leading organisations are rebuilding processes, roles, and career paths around expanded AI capabilities. They are creating new roles: AI operations managers, human-AI interaction specialists, quality stewards. They are treating AI as a structural component of how work is organised, not an add-on to existing processes.

Only 34% of companies are doing this. They are the ones pulling ahead.

The remaining 66% are in the middle: either redesigning key processes while keeping business models intact, or using AI at a surface level with no structural change at all. Both groups are capturing some efficiency gains. Neither is achieving the strategic differentiation that Deloitte identifies as the real prize.

The gap between strategy and operations is where value dies

Deloitte finds that 42% of companies believe their strategy is highly prepared for AI adoption. Only 20% say the same about their talent infrastructure and only slightly more about their technical infrastructure. Leaders feel strategically ready but operationally unequipped.

This is the precise gap where most consulting engagements stall. A strategy is agreed. Workshops are run. A roadmap is produced. And then the operational reality of implementing it, integrating it into existing systems, upskilling the people who need to deliver it, measuring whether it is working, surfaces the same problems that prevented progress before the strategy was written.

The strategy document does not close the gap. The operational infrastructure does. And most consulting firms deliver the former without the latter.

The firms that will differentiate themselves in this environment are the ones that arrive with a structured methodology for moving from strategy to activation. Not a slide deck that describes the journey. A scored diagnostic that establishes the current state. An evidence framework that tracks what changes and what does not. A reassessment loop that measures whether the activation held three months after the engagement ended.

Deloitte puts the ambition clearly: organisations should treat AI as foundational. The most successful will not be those with the most AI projects or the biggest budgets, but those who build AI into the foundation of how they operate, compete, and grow.

Building it into the foundation requires a foundation to build on. That foundation is a codified methodology. Not a generic framework. Not a market-standard maturity model. The consultant’s own accumulated expertise, structured into a scored diagnostic that produces consistent, traceable, evidence-backed outputs regardless of which engagement it is deployed on.

The proof-of-concept trap exists because most organisations do not have that foundation. They have enthusiasm, investment, and a collection of pilots that were never designed to connect to each other. The consultant who arrives with the infrastructure to build that foundation is not selling a report. They are selling the escape route from the trap.

Howard Scott has spent 30 years in consulting, martech, and digital transformation across firms including GroupM, Digitas LBi, TEQUILA/TBWA, Leagas Delaney, The Marketing Store, and The National Trust. He is a former BIMA Non-Executive Director and the founder of Studio:Blueprint, a consulting operating system that lets consultants and agencies codify and deploy their methodology through AI-powered diagnostic and evidence tools. He is author of SynKraken, a local-first command deck for AI agents.

Subscribe now

BCG surveyed 640 CEOs across 18 countries and sorted them into three archetypes. The data is worth sitting with.

Fifteen percent are Followers. They recognise AI’s potential but lack full conviction, making early cautious investments while remaining anxious about the outcome. Seventy percent are Pragmatists. They are excited and confident about AI but only invest when they see evident value and low risk. And fifteen percent are Trailblazers. They drive AI-powered transformation through decisive investment, rapid upskilling, and strong belief in the returns.

The Trailblazers are not waiting. They spend more than eight hours a week personally expanding their AI expertise. They have upskilled more than 70% of their workforce. They are allocating 60% of their 2026 AI budget specifically to agentic AI. And they are 3.4 times more likely than Followers to feel very confident their strategy will pay off.

The Pragmatists are watching them. Waiting to see the returns before committing.

Here is the problem with that strategy.

The evidence the Pragmatists are waiting for is not coming through passive observation

The BCG data shows that Trailblazers generate returns because of how they invest, not in spite of the risk they take. They invest at scale, upskill aggressively, and commit to agentic AI while others are still running pilots. The returns flow from that commitment. They do not precede it.

A Pragmatist CEO watching from the sidelines is not gathering evidence. They are accumulating delay. The evidence they would find reassuring, a clear proof point that AI transformation delivers measurable ROI in their specific context, is only available to the organisations that went through the transformation. You cannot observe your way to that evidence. You have to build it.

BCG makes this explicit. The Trailblazer flywheel works because each step reinforces the next. Making AI a top priority drives investment. Investment drives upskilling. Upskilling drives capability. Capability drives measurable ROI. Measuring ROI drives further conviction. The flywheel only starts if you make AI a genuine priority, not a cautious experiment watched from a safe distance.

Seventy percent of CEOs are currently outside the flywheel. They are waiting for evidence that only exists inside it.

AI investment has doubled. Returns have not followed automatically.

BCG’s broader survey of 2,360 executives shows that AI investment as a share of revenue is projected to double in 2026, from 0.8% to 1.7%. That is a significant commitment across the industry. Ninety-four percent of organisations say they will continue investing even if it does not pay off this year.

And yet the same survey shows that most organisations are still at the earliest stages of transformation. The gap between investment and return is not a technology problem. It is a methodology problem.

The Trailblazers are not simply spending more. They are spending differently. They define clear AI priorities. They commit capital at scale across end-to-end business functions. They upskill the organisation rather than just the technical team. And they track measurable ROI, not vanity metrics like the number of pilots running or the percentage of employees who have used a chatbot.

Tracking ROI requires knowing what you are measuring before you deploy. It requires a framework that defines what good looks like, scores current maturity against that framework, and measures change over time with evidence that traces back to specific interventions.

Most organisations do not have that framework. They have AI tools deployed across disconnected functions, each tracked by whoever happens to be responsible for that function, with no consistent methodology connecting the investment to the outcome.

The consulting opportunity is in the 70%

The Trailblazers are largely self-sufficient. They have conviction, capital, and capability. They do not particularly need a consultant to tell them what to do.

The Pragmatists are where the consulting opportunity lives. Seventy percent of global CEOs, excited about AI, confident it will pay off eventually, but waiting for a structured approach that reduces the risk of committing before they understand what they are committing to.

That structured approach is a diagnostic. A scored framework that establishes the current state of AI maturity across the dimensions that matter for that specific organisation. Not a generic maturity model. The consultant’s own framework, built from their sector expertise, deployed as a scored assessment that produces a specific picture of where that organisation stands and what it needs to do next.

The Pragmatist CEO does not need to be told that AI is important. They already know. They need a credible, evidence-based answer to three questions: where are we now, what needs to change, and how will we know it worked.

A consultant with a codified methodology, a deployed diagnostic, and an evidence tracking system can answer all three. A consultant with a slide deck and a workshop agenda can answer none of them reliably.

Studio:Blueprint was built to give consultants that infrastructure. The Forge builds the diagnostic from your methodology. The Evidence Canvas tracks what has actually changed. The reassessment loop measures whether the changes held. The Delivery Blueprint exports a structured plan that agents can execute against.

BCG has identified the market. Seventy percent of global CEOs are Pragmatists waiting for a structured path forward. The consultant who can provide that path, with evidence, is not competing for the Trailblazers. They are serving the majority.

Howard Scott has spent 30 years in consulting, martech, and digital transformation across firms including GroupM, Digitas LBi, TEQUILA/TBWA, Leagas Delaney, The Marketing Store, and The National Trust. He is a former BIMA Non-Executive Director and the founder of Studio:Blueprint, a consulting operating system that lets consultants and agencies codify and deploy their methodology through AI-powered diagnostic and evidence tools. He is author of SynKraken, a local-first command deck for AI agents.

Subscribe now

Jensen Huang stood on stage at GTC Taipei for almost two hours. Most of it was hardware: a new GPU, a new CPU, racks, cooling, factory economics. The kind of thing you skim.

Buried inside the hardware was a definition. He repeated it maybe eight times.

An agent is a model, a harness, tools, skills, and a runtime.

Then he said the part that matters for anyone running or building software for a consulting firm. Every company will become an agent company. Every company will run agents inside. And every company will discover that agents need their own operating system.

That is the thesis of Studio:Blueprint, stated by the largest company in the industry.

We did not pivot to it last week. We built on it from day zero. This piece is about why, and why that structure is the difference between a tool that survives the next two years and a tool that does not.

1. The category got named on stage

I have explained Studio:Blueprint with one line since I started building it.

Consulting firms have advice engines. They rarely have operating engines. Studio:Blueprint is the operating engine.

That line used to need a paragraph of setup. It does not anymore. When the company supplying the compute for the entire AI industry tells a room of partners that agents need an operating system, the argument is no longer mine to make alone. The category has external air cover.

The distinction holds. An advice engine produces a recommendation. An operating engine runs the firm: the clients, the engagements, the stakeholders, the pipeline, the methodology, the cash position, all connected, all live. A dashboard records fragments after the fact. An operating engine holds the state the agent acts on.

That is not a feature. It is the substrate.

2. Skills are manuals. So is a Methodology Pack.

The most useful five minutes of the keynote was about skills.

NVIDIA’s libraries now ship with a skill: a manual the agent reads so it knows how to use the library. The agent loads it, learns the method, and operates. The phrase Jensen used was that the AI reads it and understands how the tool works.

That is the Methodology Pack. Exactly.

A Methodology Pack is a configurable layer: questions, weights, thresholds, prompts, terminology. It encodes how a specific firm thinks. An agent loads it and becomes expert at that firm’s method rather than a generic consultant. We designed it as a JSON layer precisely so it could be read, swapped, and version controlled.

The industry now has a shared word for what we built. When NVIDIA says skills and harness on a main stage, the vocabulary I use to describe Studio:Blueprint stops sounding bespoke and starts sounding like the standard. That is a gift. We will use it.

3. The disruption story is backwards

Jensen took the obvious fear head on, twice. The fear is that agents will wipe out software companies and the firms that sell expertise. His answer was the opposite. Because there will be so many agents, and because the world is no longer limited by the number of humans available to do work, those agents will use more tools than ever. He listed enterprise software companies people assumed were dead and pointed out their opportunity is larger now, not smaller.

Apply that to consulting.

Agents do not remove the need for a firm’s method, its evidence, its engagement structure, its judgement about what to do next. They increase the demand for all of it, because now there is something that can act on it at scale. The constraint shifts from human hours to whether the firm’s knowledge is in a form an agent can use.

Studio:Blueprint is the form. It is the tool the agent reaches for, not the thing the agent replaces. A firm with its method encoded and its state modelled is more valuable in an agentic world, not less. A firm with its knowledge trapped in a partner’s head and a folder of slides is exposed.

4. The headcount story does not survive the numbers

NVIDIA put figures on screen. Treat them as their figures, not gospel, but the direction is hard to argue with.

Roughly three trillion dollars of developer salary is now producing something closer to nine trillion dollars of output. GitHub commits, flat-ish growth from 300 million in 2023 to 500 million in 2025, nearly tripled in the first months of 2026. And the headcount line went up. More engineers hired, not fewer.

The story being sold to the C-suite is that AI cuts jobs. The people building the hardware are saying the opposite: when output per person rises this far, you hire more of the people who can generate it.

That is the same correction I made on the MCA 2026 data. The widely quoted AI adoption figure measures client deliverable use, not how firms actually run. The headline and the mechanism point in different directions. A consulting firm reading the cut-jobs narrative and acting on it is optimising for the wrong variable. The variable that matters is output per consultant, and the lever on that is how well the firm’s knowledge and state are structured for an agent to use.

5. We bet open, and open won the keynote

The runtime NVIDIA showed, OpenShell, is open source and runs any agent: Claude Code, Codex, OpenClaw, Hermes. Adobe shipped an MCP server so agents can drive Photoshop and Premiere directly. The whole stack was models you can swap, harnesses you can choose, open standards underneath.

Studio:Blueprint is built on OpenRouter. That makes it model agnostic. The firm is not locked to one vendor’s model, and when a better or cheaper model ships, the firm switches without a migration. It also runs on MCP and ships its own MCP server, so the firm’s data and methodology are addressable by whatever agent the firm prefers. We did not bet on a single model vendor or a closed runtime. We bet that the firm should own its method and choose its tools.

The same logic applies to data. Jensen spent time on confidential computing: everything encrypted at rest, in motion, and in use, because the model is precious. Studio:Blueprint now has encryption built in, and bring your own key. The firm holds the key. Its client data, its evidence, its method stay private to it, not pooled, not readable by us. For a consulting firm handling sensitive client material, that is not a compliance checkbox. It is the precondition for putting real client work through an agent at all.

The keynote made all of this the house style of the industry. No repositioning required. The bet just got cheaper to defend.

Why this is not a SaaS

Here is the structural point, and it is the reason any of the above is true.

Most software described as agentic is a SaaS product with an AI feature added on top. Records in a database, a chat box bolted to the side. The agent can talk about the data. It cannot operate on the firm, because the firm was never modelled as a thing an agent could act on. It was modelled as rows.

Studio:Blueprint was built the other way around. The Cockpit is a graph: clients, engagements, stakeholders, pipeline as connected nodes. That is not a styling choice. It is an ontology of the firm, the relationships between things, which is exactly the structure Jensen flagged as one of the hardest problems in agentic computing. He spent real time on memory and retrieval and the relationship of data structures to each other. An agent does not want a CRM record. It wants the firm’s relationships modelled so it can reason across them.

We built the ontology first and the interface second. The agent layer was assumed from the start, not retrofitted. That is what day zero means.

There is a compounding effect. Every engagement, every diagnostic, every piece of evidence makes the firm’s model richer and the agent’s work better. The longer a firm runs on it, the more its method and history are encoded, and the more expensive it becomes to leave. That switching cost is not a lock-in tactic. It is the natural result of an operating engine accumulating the firm’s actual operating knowledge.

A SaaS tool gets disrupted when the model underneath it gets good enough to do the same job in a chat window. An operating engine does not, because the model in the chat window still needs the firm’s state, the firm’s method, and the firm’s relationships to do anything useful. We are the layer that holds those. The smarter the models get, the more valuable that layer becomes.

That is the bet. NVIDIA just spent two hours and a Vera Rubin launch confirming the architecture. We made it the foundation before there was a keynote to point at.

If you run a consulting firm, from a solo practice to a thousand people, the question is not whether agents are coming. They are here, and the hardware roadmap says they get cheaper and faster from now on. The question is whether your firm’s method and state are in a form an agent can use, or trapped in slides and someone’s memory.

See what an operating engine looks like, and start running your firm on one, at studioblueprint.uk.

Subscribe now

The story underneath it was about something more interesting: how AI has made the small firm scalable for the first time.

Mark Bunker left Deloitte as a senior advisory partner and founded Queen’s Tower Advisory. Marissa Thomas, former COO of PwC UK, launched Unity Advisory. The FT describes a growing crop of firms founded by veteran consultants who believe AI has fundamentally changed the economics of the industry (paywalled article). Not because it makes consulting cheaper. Because it removes the barrier that previously made scale impossible.

Bunker put it plainly: “Your platform might have 20 people, but with the amplification factor of AI, you’re suddenly at 100 people, 150 people, very quickly.”

He is right. And the FT is right that this is happening. But there is a piece of the story the article does not tell: what, exactly, provides that amplification? What is the infrastructure that turns a 20-person firm into a 100-person firm? What does a consultant actually need to build before the AI can do its job?

The barrier that remains

The FT describes three forces dismantling the traditional consulting model. The reliance on generalist consultants whose work AI now performs. The billing-by-time model that collapses when agents can do in minutes what junior staff did in weeks. And the pyramid staffing structure that loses its logic when you no longer need armies of people to deliver the work.

These are real. But the article is describing what is being dismantled, not what needs to be built in its place.

Smaller firms have the agility the FT describes. They do not have what the large firms have spent decades accumulating: codified methodology, structured knowledge systems, proprietary frameworks, and institutional memory. The Big Four can point an AI at a vast archive of structured engagement data. A 20-person boutique has the expertise in the heads of its partners and a collection of slide decks on a shared drive.

That asymmetry is the real threat to the insurgent firms. Not the incumbents’ capital. Their structured data.

The amplification factor is your methodology

When the FT quotes Bunker on the amplification factor of AI, it is describing the output. A 20-person firm performing like a 100-person firm. But amplification only works if there is something to amplify. An AI agent without a structured methodology is a powerful tool in search of a purpose. It can write, analyse, and summarise. It cannot determine what to measure, how to score it, what the results mean, or what the client should do next.

That is the consultant’s job. And right now, for most small and mid-size firms, that expertise exists only in an uncodified form. It is in the partner’s head, in the workshop agenda, in the deck they adapt for every new client.

The firms that will actually achieve the amplification Bunker describes are the ones that have done the hard work of making their methodology machine-readable. Dimensions, questions, scoring logic, maturity bands, evidence criteria, all structured into a system that AI agents can read, execute against, and report through.

The Management Consultancy Association figures in the FT show smaller firms growing at up to 50% as AI helps them compete. That growth is not coming from firms that gave their consultants access to ChatGPT. It is coming from firms that have built the structured systems their AI can operate against.

The billing model is already shifting

The FT notes that McKinsey now ties roughly a third of its work to performance-based fees. KPMG expects pricing models to evolve toward subscriptions and success-based fees. The billable hour is under pressure across the industry.

This is not just a pricing change. It is a fundamental shift in what consulting is being asked to prove. Performance-based fees require measurement. Subscription fees require ongoing value demonstration. Neither model works if the consultant cannot show, with evidence, that the client’s situation has actually improved.

That evidence layer does not exist in most small consulting firms today. Client progress is tracked through quarterly conversations and self-reported updates. Improvement is described rather than measured. The consultant knows their client is better off. They cannot prove it systematically.

The firms that capture the new pricing models will be the ones that can produce that evidence continuously. Scored diagnostics that run at the start of an engagement and again at regular intervals. Evidence uploaded by the client and connected to specific dimensions of the assessment. Progress tracked against a roadmap derived from the scores. A system that shows, with data, what changed and why.

The mid-tier squeeze

The FT makes an important observation about mid-tier firms caught between the Big Four’s capital and the boutique’s agility. Squeezed from both sides, risking being stranded in the middle.

The boutique’s advantage is not just size. It is speed and specialisation. The FT quotes Tom Shave of Ryan: “The firms that succeed in this AI world will be the ones that are highly specialised rather than the more generalist type of service.” Clients are arriving with an initial AI-generated diagnosis already in hand. They need the deep expert, not the generalist framework.

That specialisation only becomes a competitive weapon when it is codified. The deep expert whose methodology lives in their head cannot scale it, cannot delegate it, cannot prove it systematically. The deep expert whose methodology is structured into a scored diagnostic, deployed to clients under their own brand, tracked through evidence and reassessment, that expert can operate at a scale that was previously impossible.

The window is open. It will not stay open.

Using Claude or ChatGPT to amplify your consulting practice is not the same thing as building the infrastructure that makes amplification sustainable.

A language model is a general-purpose tool. It is extraordinarily capable. It will draft your report, summarise your research, and generate a first pass at almost anything you ask it. But it does not know your methodology. It does not remember your last client. It does not score consistently across engagements. It does not track whether the advice you gave six months ago actually worked. Every conversation starts from zero.

The firms that will achieve the FT’s amplification factor are not the ones that use AI tools. Every firm will use AI tools. They are the ones that have built a structured system underneath those tools: a codified methodology, a deterministic scoring engine, an evidence layer, a reassessment loop, and a delivery architecture that packages the consultant’s expertise into something an agent can execute against.

That system is the difference between a consultant who uses AI and a consulting operating system that runs on AI.

The window for building it is open now. But it will not stay open indefinitely. The model companies have already entered consulting. Anthropic partnered with Blackstone and Goldman Sachs to launch a $1.5 billion enterprise services venture. OpenAI launched DeployCo with $4 billion and 150 Forward Deployed Engineers. It is not difficult to imagine the next iteration: Claude Consulting. GPT Advisory. AI-native platforms that combine model capability with structured consulting methodology, built by companies with the capital to develop both simultaneously.

When that happens, the independent consultant or boutique firm competing with a general-purpose AI consulting product will need something those products cannot replicate: their own methodology, their own evidence, their own client history, their own proprietary dataset built from years of engagements. That accumulated, structured data is the advantage that no model company can buy.

Studio:Blueprint is the infrastructure for building it now. Not because AI tools are insufficient. Because the consultant who builds the structured data layer today will have the proprietary foundation that model companies cannot commoditise. The tool is not the advantage. The codified methodology underneath it is.

What the FT did not say

The article describes the threat to the incumbents and the opportunity for the insurgents. It does not describe what the insurgents need to build to capture that opportunity.

The answer is not more AI tools. It is more structured data. Methodology encoded into scored frameworks. Evidence connected to dimensions. Client progress tracked over time. Delivery packaged into agent-ready exports.

Queen’s Tower Advisory. Unity Advisory. The FT profiles them because they are early. In twelve months there will be dozens more. The ones that arrive with a codified methodology and a structured data layer underneath will achieve the amplification Bunker is talking about. The ones that arrive with talent and ambition but no system will discover that the AI has nothing specific enough to amplify.

The FT called it an inflection point. It is. The question is whether your methodology is ready for it.

Howard Scott has spent 30 years in consulting, martech, and digital transformation across firms including GroupM, Digitas LBi, TEQUILA/TBWA, Leagas Delaney, The Marketing Store, and The National Trust. He is a former BIMA Non-Executive Director and the founder of Studio:Blueprint, a consulting operating system that lets consultants and agencies codify and deploy their methodology through AI-powered diagnostic and evidence tools. He is author of SynKraken, a local-first command deck for AI agents.

Subscribe now

The best engineers, he said, are no longer writing code. They are directing agents that write code.

He is not wrong about the direction of travel. He may be very wrong about how to get there.

Gartner surveyed 350 global business executives, all running companies with annual revenues of at least a billion dollars, all already deploying AI agents, automation, or autonomous business capabilities. 80% of them had reduced their workforce. And according to Gartner’s Distinguished VP Analyst Helen Poitevin: “Workforce reductions may create budget room, but they do not create return.”

Read that carefully. The companies that cut the most showed nearly identical financial returns to the companies that cut the least. In several cases, the ones that cut less performed better.

This is the most important finding in enterprise AI so far, and almost nobody is talking about it.

The confusion between cost and value

There is a difference between cutting costs and creating value, and most leadership teams are currently conflating the two.

Cutting 22% of your workforce reduces your payroll. It does not automatically make your remaining workforce more capable, your product better, your customers better served, or your operating model more resilient. It frees up budget. It does not generate return.

The firms seeing real returns from AI are the ones investing in upskilling and human-centred operating models, not headcount reduction. Gartner calls this model “people amplification”: using AI to make workers more capable rather than replacing them outright.

That is a fundamentally different bet. One is a cost decision. The other is a capability decision. The first is easy to explain on a spreadsheet. The second requires you to actually understand what your organisation is capable of and what it could become.

What ClickUp got right, and what it got wrong

Evans is right that the operating model needs to change. His three categories of essential employee: builders who orchestrate agents, system managers who automate their own roles, and those who bring judgment that agents cannot replicate, describe a genuinely different way of working. The instinct is correct.

But firing 22% of your staff to get there conflates the destination with the method. You do not build a 100x organisation by removing people. You build it by changing what people do, then building the systems that support them in doing it differently.

Klarna replaced 700 customer service roles with AI, watched quality decline, and started rehiring. IBM automated large parts of its HR function and reversed course when the systems could not handle anything requiring judgment. The Commonwealth Bank of Australia reversed 45 AI-driven layoffs after concluding the roles were never redundant.

These are not edge cases. They are the early returns on the cost-cutting approach to AI. Cut first, build the systems second, discover the systems cannot replace human judgment, rehire at a premium.

The operating model question nobody is asking

The right question is not “how many people can we replace with agents?” The right question is “what could our people do if the agents handled the operational layer?”

A consultant spending 40% of their time on report writing, evidence gathering, progress tracking, and reassessment management is not doing consulting. They are doing administration. Give them an operating system that handles the administration and they can do four times the consulting work with the same hours. The output increases. The quality increases. The client relationship deepens. Nobody got fired.

This is not a charitable argument. It is the argument the data supports. The companies achieving the highest gains were those using AI as a form of people amplification, implementing the technology to make workers more productive rather than outright replacing them.

The methodology question

Here is where most organisations stumble, and where the operating model shift actually starts.

Agents need something to work against. An agent without a structured operating framework is a powerful tool in search of a purpose. It can write, summarise, and execute tasks. But it cannot determine which tasks matter, in what order, measured against what criteria, and reported against what framework.

That structure comes from the humans. From their methodology. From the accumulated judgment of people who have spent careers understanding how their domain works and what good looks like.

The organisations that will get the most from AI agents are the ones that have done the hard work of codifying that methodology. Not into a slide deck. Into a structured, scored, deployable system that agents can read, execute against, and report back through. Studio:Blueprint was built to do exactly that: take a consultant’s methodology and turn it into the operating instruction set that an agent workforce can run against.

The consultant who codifies their diagnostic framework does not become redundant. They become the architect of a system that scales their expertise. The engineer who documents their standards and processes does not automate themselves out of a job. They build the operating model that makes the entire team more capable.

This is the operating model shift that ClickUp is gesturing at but has not quite articulated. The “10x engineers” Evans describes are not valuable because they can direct agents. They are valuable because they have the judgment, standards, and methodology that give the agents direction. Remove the humans and you remove the framework the agents need to operate against.

The better bet

Gartner’s Helen Poitevin put it directly: “Many CEOs turn to layoffs to demonstrate quick AI returns; however, this disposition is misplaced.”

Cutting people demonstrates action. It produces a number on a spreadsheet. It does not produce a better operating model, a more capable team, or a product that serves customers more effectively.

The organisations that will look prescient in five years are not the ones that fired 22% of their workforce to fund AI infrastructure. They are the ones that changed what their people do, built the systems that support them in doing it, and compounded that capability over time.

Fewer people doing the same work is cost optimisation. The same people doing work they could never do before is transformation. The data says only one of those approaches is producing returns.

The methodology is the foundation. The people who carry it are not the cost. They are the point.

Howard Scott has spent 30 years in consulting, martech, and digital transformation across firms including GroupM, Digitas LBi, TEQUILA/TBWA, Leagas Delaney, The Marketing Store, and The National Trust. He is a former BIMA Non-Executive Director and the founder of Studio:Blueprint, a consulting operating system that lets consultants and agencies codify and deploy their methodology through AI-powered diagnostic and evidence tools. He is author of SynKraken, a local-first command deck for AI agents.

Subscribe now

The outcomes are promised on the front cover. The delivery logic exists only in the heads of the people who will run the engagement.

Call it a sprint, a diagnostic, a rapid assessment, or a transformation accelerator. The naming is confident. The underlying question is the same: if you removed the people who normally run this, how much of it would survive?

In most cases, not much. Because the actual product is not the framework. It is the judgement of the people applying it. And judgement, however excellent, is not a product. It is a service.

The distinction has always mattered. It matters considerably more now.

What a consulting product actually requires

A genuine consulting product has five things that a service dressed in product language does not.

The first is methodology that exists in writing, in sufficient detail for someone who was not involved in designing it to deliver it consistently. Not a one-page framework. A documented process: what happens in what order, what inputs are required at each stage, what decisions are made and on what basis.

The second is scope that is defined precisely enough to be defended. A product has a boundary. A client knows what is inside it and what is not. A service tends to expand to fill the available time and relationship. A product does not.

The third is evidence requirements. What data, information, or organisational access does the product need to function? A product specifies this in advance. A service discovers it during delivery.

The fourth is delivery logic: the sequence of steps, decisions, and handoffs that produces the outcome. This is the part most firms have never written down, because the people who run their engagements learned it through experience rather than documentation.

The fifth is a measurable outcome. Not a deliverable. An outcome. Something that either happened or did not, and that can be assessed independently of whether the client found the presentation compelling.

Most consulting offerings that describe themselves as products have some of these. Few have all five. And all five are required before agents can be usefully deployed inside them.

Why this distinction is urgent now

The same disruption that is reshaping software markets is beginning to reshape professional services. Bain’s most recent technology research analysed the conditions under which agentic AI cannibalises existing workflows versus conditions under which it strengthens them. The analysis was designed for software companies, but the logic applies directly to consulting service lines.

The offerings most at risk are those with low contextual knowledge requirements, high task repetition, and delivery logic that is visible and replicable from the outside. The offerings that hold their value are those with deep domain knowledge embedded in the delivery, strict oversight requirements, proprietary data or methodology, and complexity that does not yield to simple automation.

Every practice area leader can look at their service portfolio through this lens. Which offerings are strongholds? Which are exposed? The analysis is not comfortable. Many of the most common consulting offerings, market sizing, competitor benchmarking, process mapping, readiness assessments, sit on the wrong side of the line. They are not immune from disruption. They are directly in its path.

The same Bain research notes that firms which hold exclusive data and proprietary methodology have a head start on full automation because they control the rules. The inverse is equally true: firms whose delivery logic is generic, informal, or unwritten have no defensible moat. A competitor with better tooling can replicate the output faster and cheaper.

The scope problem at scale

MIT Sloan Management Review and Boston Consulting Group found that organisations further along in agentic AI adoption expect significant redesign of how their work is organised. Sixty-six percent expect to redefine jobs and processes substantially, compared with 42% of those only beginning adoption. The research frames the central shift clearly: agentic AI is allowing organisations to build new ways of working rather than simply adding AI to old routines.

For consulting firms, the scope implication is specific. Services that were scoped around the time of experienced humans will not map cleanly onto services delivered with agent assistance. The delivery rhythm changes. The input requirements change. The points at which human judgement is genuinely necessary change. A service line that has never been precisely scoped cannot be re-scoped for an AI-assisted delivery model because there is nothing precise enough to work from.

This is not a technology problem. It is a design problem. The firms that will adapt most successfully are those that treat their service lines as products to be engineered, not services to be staffed.

The pricing question

Bain’s analysis of the SaaS disruption identifies a shift that applies equally to professional services: the fundamental change is from charging for access to charging for work done. Seat-based software pricing does not survive when agents are doing the work. Time-and-materials consulting pricing faces the same pressure when the time required to produce an output compresses significantly.

Outcome-based pricing is the answer that most consulting firms describe as aspirational and few have operationalised. The reason is not commercial reluctance. It is structural: outcome-based pricing requires outcome-defined methodology. You cannot price for a result you have not specified in advance with enough precision to know when it has been achieved.

A consulting product that defines its outcome clearly, specifies what evidence will demonstrate that outcome, and documents the delivery logic that produces it is a product that can be priced on results. A service that delivers high-quality thinking over an agreed number of weeks cannot be.

The firms that move to outcome-based commercial models first will have a significant advantage in both pricing power and client relationships. The prerequisite is doing the harder work of specifying what they actually deliver.

From diagnostic to product

Studio:Blueprint’s Forge is built around this distinction. A diagnostic is not a report. It has defined inputs, a documented assessment methodology, evidence requirements, scoring logic, and an output that maps directly to a roadmap. Every element of the delivery process is specified. An engagement run by a senior practitioner and one run by a less experienced team member produces outputs that are comparable, because the methodology is carrying the consistency rather than the individual.

That is what a product does. It transfers the intelligence of the best delivery into the system, rather than leaving it resident in specific people.

The challenge for larger firms is scale: how do you do this across dozens of service lines and hundreds of practitioners? The answer is the same as it always has been for any large-scale quality problem. You start with the highest-value, highest-frequency offerings. You document the delivery logic. You test it against actual engagements. You refine it. You do not attempt to systemise everything simultaneously.

The product question is the AI question

Most conversations about AI in consulting focus on tools: which models to use, how to embed AI in workflows, what the governance requirements are. These are real questions. They are also secondary.

The primary question is whether the service has been specified precisely enough to benefit from AI assistance at all. If the delivery logic is informal, AI can make the people involved more efficient. It cannot make the service more consistent, more scalable, or more defensible.

The firms that treat their offerings as products, that do the work of specifying methodology, scope, evidence requirements, delivery logic, and measurable outcomes, are building infrastructure that AI can compound. The firms that treat their offerings as services staffed by talented people are building something AI will eventually route around.

Productised consulting is not a template. It is an architecture. The template is the smallest and least important part of it.

Subscribe now

The question this creates is not how to produce a better competitor analysis. It is what the firm is actually selling when the analysis itself is no longer scarce.

This is the question most consulting firms are not yet asking directly. They are asking how to use AI to produce better outputs faster. That is the right question for 2023. The question that will define 2026 and beyond is different: when everyone can generate a convincing recommendation, what is a professional services firm actually worth?

The answer is not the recommendation. It is the evidence behind it.

The deliverable is being commoditised

The commoditisation of the consulting deliverable is not a distant risk. It is in progress. Bain’s most recent technology research found that agentic AI is disrupting the application layer of software markets by automating tasks and replicating workflows that previously required human expertise. The same dynamic applies directly to professional services. The analysis, the framework, the structured recommendation: these are knowledge work outputs, and knowledge work outputs are precisely what current AI systems are designed to produce at scale.

This does not mean consulting is finished. It means the part of consulting that was always most visible, the deliverable itself, is becoming the least defensible part of the value proposition. Clients who are sophisticated enough to be buying consulting are increasingly sophisticated enough to know what AI can produce. They are asking sharper questions about what they are paying for.

Deloitte’s most recent enterprise research found that 35% of organisations cite mistakes and errors with real-world consequences as their leading concern about AI-generated outputs. Another 29% cite general loss of trust due to bias, hallucinations, and inaccuracies. Those numbers are not just a technology adoption problem. They are a commercial opportunity. The firms that can demonstrate that their outputs are trustworthy, traceable, and attributable will have a significant advantage over those that cannot.

The provenance problem

The trust problem has a specific shape. It is not that clients distrust AI in principle. Most are using it themselves. The problem is provenance: the ability to trace a recommendation back to the evidence that supports it, and to identify who is responsible for the judgement calls made along the way.

This matters more in professional services than almost anywhere else. A consultant’s value has always rested partly on their ability to be challenged. A client who can ask “how did you arrive at this?” and receive a clear, defensible answer is a client who can act on the recommendation with confidence. A client who cannot get that answer is a client who will second-guess the engagement from the moment the final presentation ends.

When outputs were produced by human teams over weeks, provenance was implicit. The people who did the work were visible. The process was traceable even if it was not always documented. When outputs are produced by agents over hours, provenance has to be designed in deliberately. It does not appear by default.

IBM’s research puts a number on this: 45% of executives cite lack of visibility into agent decision-making as a significant barrier to deploying AI at scale. The framing is important: this is described not as a technical limitation but as a design choice. The firms building trust into their AI-assisted delivery are engineering it in from the start, not hoping it emerges from the quality of the output.

What showing your work actually requires

The phrase is useful because it is familiar. Every professional knows what it means to show your workings. In a consulting context, it means something specific.

It means the client can see what evidence was used to reach a conclusion, not just the conclusion itself. It means the reasoning steps are visible and can be challenged at each point. It means the judgement calls made by senior professionals are identifiable and attributable, rather than buried inside a polished final document. And it means that when something is wrong, the firm can find where the error entered the chain and explain why it was not caught.

Accenture’s co-intelligence research is direct on this point: leaders must define escalation paths, decision rights, and accountability so that responsibility for outcomes, risk, and trust remains clearly owned. That is not a governance footnote. It is a description of what a client is buying when they hire a firm they trust.

McKinsey’s research on the agentic organisation notes that performance measurement in hybrid human-agent teams will shift from tracking task completion to tracking how well people orchestrate agents and deliver outcomes. The same shift applies to client-facing work. The measurable question is no longer whether the deliverable was produced on time. It is whether the firm can account for the quality of every judgement embedded in it.

Trust as a structural advantage

The firms that build this capability early are not just managing risk. They are building a competitive position that is genuinely difficult to replicate.

Any firm can deploy an AI model. Any firm can produce a market analysis in an afternoon. The question is which firm can produce that analysis and then stand behind it under client scrutiny, demonstrate the evidence chain, and stake their professional reputation on the judgement calls it contains.

That capability requires infrastructure. It requires a system for capturing what agents propose, recording what consultants decide, and preserving the evidence that justifies each decision. It requires that this system is accessible when clients ask questions, not reconstructed after the fact from incomplete records.

Deloitte’s enterprise research found that reliability and trust have improved in AI deployments, but both still have a long way to go. Senior decision-makers may not yet be demanding proof of rigour consistently, but the research is clear that they will be. The window for building this infrastructure before clients start asking for it is narrowing.

The firms that wait until clients demand it will be building under pressure and catching up. The firms building now will be able to demonstrate it proactively, which is a different commercial conversation entirely.

The evidence layer

Studio:Blueprint is built around this principle. The Evidence Canvas is not a reporting tool. It is the layer that connects what agents produce, what consultants decide, and what clients can see. It makes the reasoning chain visible and persistent. It turns accountability from an implicit professional norm into an explicit, auditable feature of the engagement.

This matters for every size of firm. For smaller practices, it provides the kind of rigour that was previously the preserve of larger teams. For larger firms and practices, it makes the evidence infrastructure scalable across engagements without requiring senior time to reconstruct it on request.

The principle is the same either way: the firm that can show its work is the firm that can charge for it with confidence.

The differentiator has shifted

The consulting industry has always competed on the quality of its thinking. That remains true. But the quality of thinking that clients will pay a premium for is increasingly the thinking that comes with evidence attached.

The recommendation alone is not enough. Anyone can produce a recommendation. The recommendation with a clear evidence trail, with visible judgement calls, with named accountability at each step: that is what professional services have always been in their best form. AI does not change that standard. It raises the bar for meeting it.

The future consulting firm will need to show its work. The firms that are building the infrastructure to do that now are not responding to a threat. They are defining what the profession looks like next.

Subscribe now

Its purpose: embedding Anthropic’s engineers directly into mid-size businesses to deploy AI across their core operations.

Read that carefully. Not Fortune 500. Mid-size businesses. The same clients many independent consultants and small advisory firms serve.

Anthropic acquired Fractional AI, a San Francisco-based firm, as the operational foundation. The venture is designed to help companies adopt Claude models across their operations, with dedicated engineering teams deployed inside client organisations. This is not an API partnership or a reseller arrangement. This is Anthropic sending its own people into your clients’ businesses to do consulting work.

OpenAI launched DeployCo three weeks ago with $4 billion and 150 Forward Deployed Engineers. IBM launched Forward Deployed Units. Google committed $750 million to help consulting firms deploy agentic AI. And now Anthropic, the company whose models power a significant portion of the AI consulting tools in the market, is entering the consulting business directly.

The model makers have collectively decided that consulting is too important, and too broken, to leave to consultants.

Why this matters more than DeployCo

OpenAI’s DeployCo targets large enterprises. IBM’s FDUs serve global corporations. Those are markets most independent consultants were never competing in.

Anthropic’s venture is different. It targets mid-size businesses explicitly. Companies with 200 to 2,000 employees. Companies with real AI transformation needs and realistic budgets. Companies that currently hire independent consultants and small advisory firms to help them navigate exactly these decisions.

Anthropic is not entering the market above you. It is entering the market beside you. With $1.5 billion in backing, Goldman Sachs distribution, and the engineering team that built the models everyone else is using.

That should concentrate the mind.

The gap Anthropic cannot close

Here is what Anthropic will do well: deploy Claude into business operations. Configure workflows. Build integrations. Optimise model performance for specific use cases. They understand the technology better than anyone because they built it.

Here is what Anthropic cannot do: understand your client’s industry, diagnose their specific organisational maturity, assess whether their team is ready for the change, build a measurement framework that tracks progress against business outcomes rather than technical metrics, and sustain a relationship that evolves the client’s capability over quarters and years.

Anthropic will send engineers. Engineers deploy technology. They do not diagnose organisational readiness. They do not assess whether the leadership team has the appetite, the budget governance, the data foundations, and the change management capability to absorb what the technology makes possible. They do not measure whether the client actually improved six months later.

That diagnostic and advisory layer is where independent consultants live. It is also the layer that Anthropic’s venture will need to partner with, or that its clients will seek independently when the engineering deployment stalls because nobody assessed whether the organisation was ready for it.

The consultant who has infrastructure wins this

The independent consultant who shows up with experience and a slide deck is now competing against Anthropic’s own engineers for the same client’s attention. That is not a comfortable position.

The independent consultant who shows up with a codified methodology, a scored diagnostic that measures exactly where the client stands, an evidence framework that tracks whether the deployment actually changed anything, and a structured export that coordinates with whatever agent infrastructure the client adopts, that consultant is offering something Anthropic’s engineers do not provide.

Anthropic deploys the technology. The consultant diagnoses whether the organisation is ready for it, measures the impact, and proves the value over time. Those are complementary capabilities, not competing ones. But only if the consultant has the infrastructure to deliver at a standard that justifies sitting alongside a $1.5 billion venture.

A PDF assessment and a quarterly check-in call will not cut it when the alternative is Anthropic’s engineering team embedded in the building.

$10.9 billion in quarterly revenue. $900 billion valuation. And they still need consultants.

Anthropic is projecting $10.9 billion in revenue this quarter alone. They are closing a round that could value the company above $900 billion, surpassing OpenAI. Over 1,000 enterprises are spending more than $1 million annually on Claude.

And still, they launched a consulting company. Because the models alone are not enough. Deployment, adoption, organisational change, these remain human problems that require human judgment applied through structured frameworks.

The irony is sharp. The most valuable AI company in the world just validated the consulting industry by entering it. They did not build a better model and wait for clients to figure it out. They hired consultants, acquired a consulting firm, and partnered with Goldman Sachs to fund the deployment work.

If Anthropic, with the best models and the most enterprise traction in the market, concluded that consulting is essential to AI adoption, then consulting is not threatened by AI. It is validated by it. The question is whether your version of consulting is structured enough to hold its ground in a market where the model companies themselves are now competing for the same clients.

The infrastructure exists

Studio:Blueprint was built for this moment. The Forge codifies your methodology into a scored, deployable diagnostic. The Evidence Canvas tracks whether the deployment actually produced results. The reassessment loop measures progress over time. The structured markdown export hands off to whatever agent infrastructure the client adopts, whether that is Anthropic’s engineers, their own internal team, or an independent agent workforce coordinated through tools like SynKraken.

The model companies are entering consulting because the gap between building AI and deploying AI is worth billions. The consultants who close that gap with structured, evidence-backed, methodology-driven infrastructure will work alongside the model companies, not against them. The ones who don’t will find themselves replaced by engineers who understand the technology but not the business.

The consulting industry was just validated by the most valuable AI company on earth. What you do with that validation depends entirely on whether your methodology is ready to prove its worth.

Howard Scott has spent 30 years in consulting, martech, and digital transformation across firms including GroupM, Digitas LBi, TEQUILA/TBWA, Leagas Delaney, The Marketing Store, and The National Trust. He is a former BIMA Non-Executive Director and the founder of Studio:Blueprint, a consulting operating system that lets consultants and agencies codify and deploy their methodology through AI-powered diagnostic and evidence tools. He is author of SynKraken, a local-first command deck for AI agents.

At this point, they clearly can. The question is what happens to accountability when they do.

Most firms are discovering this the hard way. An agent drafts a market analysis. A senior consultant reviews it, judges it credible, and it goes into the client deliverable. Weeks later, the client challenges a figure. Where did it come from? The answer lives somewhere in a prompt history and three synthesised sources that no longer exist in their original form. The output was defensible. The audit trail was not there.

This is not a competence problem. It is a governance problem. And it is the one that will determine which firms can deploy agents at scale and which ones create liability faster than they create value.

The governance gap

Professional services firms were built on a clear accountability structure. Partners sign off. The hierarchy of review exists not just for quality assurance but for professional liability. A client knows who is responsible for a recommendation. A partner knows precisely what they are committing to when their name goes on the work.

Agentic AI does not map cleanly onto this structure. MIT Sloan Management Review and Boston Consulting Group surveyed executives globally and found that 76% now view agentic AI as more like a coworker than a tool. That framing is accurate in one sense and dangerous in another. Agents do not merely retrieve information. They reason, synthesise, and act across multi-step workflows. They behave more like a junior analyst than a search engine.

But junior analysts sit within an accountability structure. They can be asked to explain their reasoning. Their work is traceable to decisions they made with awareness of consequences. Agents, by default, are not held to the same standard. And the gap between what they produce and what can be stood behind is where the governance problem lives.

The same research identifies the supervision tension as one of the fundamental challenges organisations face when integrating agentic AI. The question it poses is precise: how do you supervise something designed to work autonomously? Traditional oversight models assume either full human control or defined automation with clear handoffs. Agentic systems sit between those states. Treat them like a human team member and you under-govern them. Treat them like a software tool and you overestimate the reliability of their outputs.

A model that holds

The firms making genuine progress on this are not the ones with the most sophisticated agents. They are the ones with the clearest model for where human judgement sits in relation to agent output.

The model is simple in principle and requires deliberate design in practice.

Agents propose. Consultants decide. Evidence proves.

Each element matters. Drop one and the model breaks.

Agents propose means agents surface analysis, generate options, draft recommendations, and flag inconsistencies. They do not make final judgements. They do not commit the firm to positions. They produce structured, reviewable work that a senior professional can assess, challenge, or reject.

Consultants decide means senior professionals retain genuine decision rights, not nominal ones. The failure mode here is common and easy to miss: consultants approve agent outputs without reviewing them critically because the volume is high and the outputs look credible. This is not oversight. It is liability transfer dressed as governance. The decision layer requires time, authority, and the real capacity to push back.

Evidence proves means the reasoning chain is visible, persistent, and auditable. Where did this analysis originate? What sources informed it? What did the agent identify as uncertain or outside its scope? These are not procedural questions. They are the questions a client will ask when they want to scrutinise a recommendation, and the questions a regulator will ask when they are not satisfied with the answer. A firm that cannot respond to them does not have a governance model. It has a risk exposure.

The trust problem is structural

IBM’s research found that 45% of executives cite lack of visibility into agent decision-making as a significant barrier to deploying agents at meaningful scale. The finding is framed clearly: the opacity of agent reasoning is not a technical limitation of current models. It is a design choice. Organisations that are serious about professional deployment are engineering transparency in from the outset, not adding it once a problem emerges.

Accenture’s research on the interplay between platforms, agents, and humans maps three distinct roles with distinct responsibilities: platforms handle structured data and rules-based processes; agents act dynamically across systems and tasks; humans bring judgement, context, and accountability. The model functions when those boundaries are explicit and deliberately maintained. Without clarity on where agent autonomy ends and human decision authority begins, orchestration becomes guesswork.

The consequence in professional services is direct. A financial services leader interviewed in the MIT Sloan research put it plainly: in their sector, there is no client-facing use case where a human is not in the loop reviewing the output before any decision is made. That is not a limitation on what agents can do. It is a recognition of what the firm can stand behind.

For consulting firms, the professional stakes are equivalent. The product is not information. It is judgement with accountability attached. Any deployment model that separates the recommendation from the person responsible for it undermines the core value proposition.

What deliberate design requires

Building the three-part model means making decisions at three levels.

At the workflow level, firms need to map which tasks are appropriate for agent execution, which require human judgement at each step, and which require structured review at defined points. This is not a fixed allocation. It shifts as agent capability matures and as the firm’s confidence in specific workflows grows. The starting point is documentation: you cannot design a workflow you have not described.

At the quality level, the design of the review interface matters as much as the quality of the underlying model. What does reviewable agent output look like? What context does a senior consultant need to make a real decision rather than a nominal one? A well-designed review layer surfaces the agent’s reasoning alongside its conclusions, flags where confidence is low, and makes rejection as easy as approval.

At the evidence level, the question is what gets captured, in what form, and where it lives. McKinsey’s research on the agentic organisation notes that performance management in hybrid human-agent teams will shift from tracking task completion to tracking how well people orchestrate agents and deliver outcomes. That shift requires an evidence baseline: a record of what was proposed, what was decided, and why. Without it, there is no foundation for accountability and no basis for improvement.

Studio:Blueprint is built around this architecture. The design principle is that agents draft and propose; the consultant commits. Nothing is published, sent, or actioned without explicit human approval. The system surfaces the reasoning. The consultant owns the decision. That is not a constraint on agent capability. It is what makes agent capability deployable at professional services scale.

The competitive consequence

The firms that build this model early will develop an advantage that compounds. Every engagement adds to the evidence base. Every agent workflow that goes through proper review builds the audit trail. Every governance decision that is made explicitly becomes easier to apply to the next engagement.

The firms that skip the governance layer will find its absence at the worst moment: when a client disputes an output, when a regulatory inquiry requires a reasoning trail, when a partner needs to explain a recommendation they cannot fully trace back to its source.

The question is no longer whether to deploy agents inside consulting engagements. That question is settled. The question is whether the firm has the governance architecture to do it in a way it can stand behind when it matters.

Agents propose. Consultants decide. Evidence proves. That sequence is not a process constraint. It is the professional standard.

Subscribe now

…in real time, without a single integration meeting or a month of scoping?

Not through a dashboard someone built six months ago that nobody updates. Not through a weekly status call where everyone reads from a different spreadsheet. Through a live connection where the consulting methodology is plugged into the client’s operating systems and progress is measured against real data, not self-reported optimism.

That is what MCP makes possible right now. A consultant’s diagnostic framework, connected via open protocol to the client’s Jira, their HubSpot, their Salesforce, their Xero, their Shopify, their GitHub. The methodology reads from the systems where the work actually happens. The scores update when the data changes. The evidence is not a document someone uploads. It is the live state of the client’s own tools.

The consulting engagement becomes a living connection

Today, a consulting engagement is a series of discrete interactions. The consultant arrives, gathers information, analyses it, produces a deliverable, presents it, and leaves. The client implements or does not. The consultant returns in three months and asks what happened.

When the consultant’s methodology is connected to the client’s systems via MCP, the engagement is no longer episodic. It is continuous. The diagnostic dimensions are not assessed once and filed. They are measured against live data from the systems that matter. The CRM shows whether the pipeline process the consultant recommended is actually being followed. The martech platform shows whether the campaign framework is being executed. The finance system shows whether the margin improvements are materialising.

The consultant does not need to ask “how is it going?” The data answers that question in real time. The consultant’s role shifts from information gathering to interpretation and judgment. The systems provide the evidence. The consultant provides the meaning.

A human pod, plugged in

Picture a consulting engagement where the firm operates as a small, senior pod. Three people. One leads the methodology and client relationship. One manages the technical integration and agent coordination. One handles delivery and evidence review.

Between them, their agent workforce monitors twelve connected systems. The methodology scores update weekly based on live data. The roadmap actions are tracked against real delivery milestones in the client’s project management tool. The evidence layer pulls from actual campaign results, actual revenue data, actual adoption metrics. When a dimension score changes, the pod knows why, because the data that caused the change is traceable to a specific system and a specific time period.

This is not a 30-person team. It is not a six-month implementation programme. It is three senior humans with a connected agent workforce, operating against structured methodology, measuring progress from live systems. The pod is small because the data does the heavy lifting. The humans are there for the judgment that data alone cannot provide.

Software is dead. Your data layer is what matters.

The era of buying software to solve consulting problems is ending. The CRM, the project management tool, the analytics platform, the reporting suite, none of them are going away. But the competitive advantage has shifted from which tools you use to how your data flows between them and whether there is a structured methodology governing what that data means.

A consulting firm that operates on well-structured data, methodology codified into scored frameworks, evidence classified by strength and linked to diagnostic dimensions, client progress tracked through deterministic scoring, operational intelligence derived from practice-wide patterns, that firm can connect to any client system via MCP and start measuring immediately. The tools are commoditised. The data structure is the differentiator.

A consulting firm that operates on documents, slide decks, and spreadsheets cannot plug into anything. Their methodology is implicit. Their evidence is anecdotal. Their progress measurement is a conversation. When the client asks “show me what changed,” they have a narrative. Not data. Not evidence. Not a score that traces back to a specific system, a specific metric, a specific date.

From your methodology up to your deliverables

This is the question every consulting firm needs to answer now, not in twelve months, not when the market forces it, now: is your practice built on structured data or on documents?

From your methodology: is it codified into dimensions, questions, scoring logic, and maturity bands that a system can deploy and an agent can measure against? Or does it live in a slide deck that gets adapted by hand for every engagement?

From your models: are your diagnostic frameworks deterministic, producing consistent scores regardless of who runs them? Or are they qualitative assessments that depend entirely on who is in the room?

From your evidence: is client progress tracked against scored criteria with classified evidence strength and full traceability? Or is it a quarterly conversation where everyone agrees things feel better?

From your deliverables: does your engagement output include structured data that agents can execute against, systems can connect to, and clients can measure autonomously? Or is it a PDF that gets filed and forgotten?

From your engagements: is your client relationship sustained by a live data connection that shows progress in real time? Or does it depend on you remembering to follow up?

Every layer of the consulting practice, from methodology at the foundation to client engagement at the surface, either runs on structured data or it does not. The firms that do will connect to client systems, measure progress in real time, and prove value continuously. The firms that do not will be asked, increasingly, to justify their fees against competitors who can.

The plan for the future is the data layer

The model companies are building agent infrastructure. Google, Anthropic, OpenAI, Apple, Perplexity, all of them shipping persistent agents that connect to external systems via MCP. IBM is deploying Forward Deployed Units where six humans and an agent workforce replace thirty-person teams. The consulting industry is restructuring around live data, agent execution, and continuous measurement.

The infrastructure to participate in this shift exists. Studio:Blueprint provides the consulting data layer: methodology ingestion through The Forge, diagnostic deployment, evidence tracking through the Evidence Canvas, client progress monitoring, and structured export for agent handoff. SynKraken provides the agent coordination layer: task assignment, room-based collaboration, and human oversight across any AI runtime. MCP provides the connection to the client’s own systems.

The question is not whether your firm needs this. The market is answering that question for you. The question is whether your data is ready.

Howard Scott has spent 30 years in consulting, martech, and digital transformation across firms including GroupM, Digitas LBi, TEQUILA/TBWA, Leagas Delaney, The Marketing Store, and The National Trust. He is a former BIMA Non-Executive Director and the founder of Studio:Blueprint, a consulting operating system that lets consultants and agencies codify and deploy their methodology through AI-powered diagnostic and evidence tools. He is author of SynKraken, a local-first command deck for AI agents.

Subscribe now

Mohamad Ali, IBM’s SVP of Consulting, said it directly: “Enterprise AI is at a tipping point. The investment is massive and experimentation is everywhere but deploying quickly remains a challenge. The issue is not the vision nor the technology.”

He is right. And what IBM has built is not a pilot or an announcement. It is a production delivery model that fundamentally changes the economics of consulting.

The delivery model is now the differentiator

For decades, consulting scale came from headcount. More people, more output, more revenue. The methodology was the intellectual property but the delivery was labour. IBM’s FDU inverts that. The methodology still comes from senior practitioners. The delivery comes from agents. The humans provide direction, governance, and judgment. The agents handle execution, documentation, testing, and iteration.

This is not augmentation. This is a structural change in how consulting work gets delivered. The ratio moves from thirty humans per engagement to six humans and a coordinated agent workforce. The economics change because the cost base changes. The quality changes because agents do not forget steps, skip documentation, or deprioritise testing under deadline pressure.

IBM runs this on their Consulting Advantage platform: reusable assets, AI agents, and industry accelerators. Every engagement builds on the last. The methods sharpen with every deployment. The institutional knowledge compounds in the platform, not in the heads of individual consultants who might leave.

If you lead a consulting practice of any size, this should have your full attention. Not because IBM is your competitor. Because the model they just demonstrated is the model every serious firm will need to adopt.

The architecture is not proprietary

Here is what IBM will not tell you: the architectural pattern underneath the FDU is not unique to IBM. It is the same pattern emerging independently across the industry.

A methodology layer that codifies how the work should be done. A diagnostic layer that assesses where the client stands. An evidence layer that captures what has been delivered and measures whether it worked. An agent coordination layer that assigns tasks, manages execution, tracks progress, and preserves decisions. A delivery layer that produces structured outputs the client can act on.

IBM built this on Consulting Advantage. OpenAI built a version of it with DeployCo and 150 Forward Deployed Engineers. Anthropic built a version with its enterprise services arm. Google built the agent orchestration substrate with Antigravity 2.0 and Managed Agents.

The pattern is the same everywhere. Senior human judgment at the edges. Agent execution in the middle. Structured methodology as the operating instruction set. Evidence as the feedback loop. The only difference is scale and cost.

The question for firms that are not IBM

IBM has 160,000 consultants, a proprietary platform, and decades of codified methodology. They can afford to build Forward Deployed Units from scratch and deploy them across global enterprises.

Most consulting firms cannot. But the capability gap between IBM’s FDU and what a well-structured firm of five, ten, or fifty people can deploy is narrower than it has ever been.

The agent infrastructure is commoditising rapidly. MCP is the universal connector. Anthropic, OpenAI, and Google all support it. Any consulting data layer that exposes an MCP server can be read by any agent on any platform. Skills files and agent instruction sets are becoming the standard packaging for codified workflows. Local coordination tools manage agent teams with task assignment, room-based collaboration, and full audit trails.

What remains scarce is the methodology. The structured, scored, defensible framework that tells the agents what to do, how to measure it, and what good looks like. IBM has this because they have spent decades building it. The independent firm or mid-sized practice has it too, but it lives in slide decks and muscle memory rather than in a system that agents can execute against.

The firm that codifies its methodology into a deployable, scored, evidence-tracked system and then connects that system to an agent workforce is running the same model as IBM’s FDU. Different scale. Same architecture. Same economics in miniature.

What a consulting operating system makes possible

Studio:Blueprint exists because this moment was coming.

The Forge ingests a consultant’s methodology and structures it into deployable diagnostics: dimensions, questions, scoring logic, maturity bands, all extracted from the consultant’s own intellectual property. The diagnostic deploys under the firm’s brand. Clients complete it. The scoring is deterministic.

The Evidence Canvas connects client documents, interview transcripts, and agent outputs to the diagnostic framework. Findings are extracted, classified by strength, and mapped to specific dimensions and scores. The consultant reviews everything. Scores update with evidence behind them.

The roadmap generates from the diagnostic scores and evidence. Every action traces to a specific dimension gap or finding. Every action includes an agent instruction: a machine-readable definition of what done looks like. The structured markdown export packages the entire engagement, methodology, scores, evidence, roadmap, and agent handoff notes, into a format that any downstream agent system can consume and execute against.

The reassessment loop runs continuously. The client uploads evidence of progress. The system matches findings to roadmap actions. Scores update. The consultant reviews. The cycle sustains itself.

That is the FDU model. Senior human at the edges providing methodology, judgment, and governance. A structured system in the middle handling diagnostic deployment, evidence processing, roadmap generation, and progress tracking. Agent coordination tools managing the execution layer. The deliverable is not a document. It is a working programme that the client’s own infrastructure executes against.

IBM charges enterprise rates and embeds six-person pods for months. Studio:Blueprint starts at £99 per month, requires no onboarding programme, and the consultant operates the entire system from their own machine. The methodology is theirs. The agents are theirs. The client data stays under their control.

The window is open

79% of organisations report some level of agentic AI adoption. 11% are running agents in production. The gap between those two numbers is the consulting opportunity of this decade.

IBM is serving the top of that market with Forward Deployed Units and enterprise economics. The rest of the market, the mid-size companies, the growth-stage firms, the organisations with real transformation needs and realistic budgets, is waiting for consultants who can deliver the same model at a scale they can afford.

The consultants who build that capability now, who codify their methodology, structure their evidence, connect their systems to agent infrastructure, and deliver executable programmes rather than PDF roadmaps, will define how consulting gets delivered for the next ten years.

IBM just showed the industry what the model looks like. The infrastructure to build your own version of it already exists. The question is whether your methodology is structured enough to power it.

-----

Howard Scott has spent 30 years in consulting, martech, and digital transformation across firms including GroupM, Digitas LBi, TEQUILA/TBWA, Leagas Delaney, The Marketing Store, and The National Trust. He is a former BIMA Non-Executive Director and the founder of Studio:Blueprint, a consulting operating system that lets consultants and agencies codify and deploy their methodology through AI-powered diagnostic and evidence tools. He is author of SynKraken, a local-first command deck for AI agents.

Senior expertise at the top. Associates below it. A pyramid designed to leverage partner judgement across multiple engagements simultaneously. It has produced decades of reliable growth for firms of every size.

It is also the reason most consulting firms hit the same ceiling.

The knowledge exists inside the firm. The methodology is real. Senior partners have spent years developing it. But it is encoded in people, not in systems. It travels with individuals, not with processes. Each time a new engagement team assembles, that knowledge has to be reconstituted from the conversations, templates, and informal understanding that senior people carry in their heads.

This is not a talent problem. It is a structural one. And AI is about to make it expensive.

The tools are not solving this

The current approach to AI adoption in consulting is predictable. Copilot rolled out to the workforce. Prompt libraries shared across practice areas. AI-assisted research and first-draft generation embedded into existing workflows. Senior people get faster access to better-quality starting points. The pyramid still works the same way.

IBM surveyed 800 C-suite executives this year and found something that should be landing harder than it is. Seventy-eight percent said achieving maximum benefit from AI requires an entirely new operating model. The same organisations had directed roughly the same proportion of their AI investment into improving existing processes. The two numbers almost cancel each other out. Leaders know what is needed. They are building something else.

Consulting firms are making the same choice. They are optimising around a structure that was already struggling to scale before AI existed.

What the structure problem actually is

The professional services pyramid was designed for a specific type of scarcity: scarce senior judgement, applied through a leverage model to more engagements than one partner could personally run. The pyramid manages that scarcity well. It is less good at managing the kind of scarcity that matters more now.

The scarce resource today is not senior time. It is documented operating logic.

Most consulting firms cannot fully describe, in writing, how they deliver their work. The engagement process exists. The quality standards exist. But they live in the judgement of experienced people, in the norms of practice areas, in the unwritten conventions that new joiners absorb over years rather than weeks. That informal knowledge is valuable. It is also invisible to machines.

Agents can run complex, multi-step workflows. They can monitor, propose, draft, flag, and synthesise. But they need something to work with. Undocumented methodology cannot be automated. Delivery logic that lives in a partner’s head stays there.

The paradigm has shifted before

McKinsey’s research on the agentic organisation maps the successive shifts in how enterprises have structured work: from industrial functional hierarchies, to cross-functional digital teams, to what it describes as flat networks of hybrid agentic teams structured to drive end-to-end outcomes. Each transition required making previously informal knowledge explicit. In the industrial era, you encoded physical processes into machines. In the digital era, you encoded information processes into software. In both cases, the prerequisite was the same: the process had to be legible before it could scale.

The same logic applies now. And it applies directly to the consulting engagement model.

The firms that benefit from agentic AI are not the ones that add the most tools. They are the ones that have made their operations legible enough for agents to work inside them.

The economics follow the structure

Bain’s most recent technology research found that AI leaders, firms that have genuinely restructured around AI rather than layered tools onto existing models, are improving operating performance by double digits while others fall further behind. The gap is not closing. It is widening.

This is an operating model story, not a technology story. The technology is available to everyone. The firms pulling ahead changed how they work, not just what software they run.

For consulting firms at scale, the stakes are specific. If delivery logic is documented, a firm can deploy capacity more consistently across practice areas without relying on senior availability to guarantee quality. If methodology is codified and machine-readable, it can be used by agents as scaffolding: not replacing judgement, but structuring the work around it so that senior time goes to the decisions that genuinely require it.

The leverage model does not disappear. It changes shape. Instead of leveraging senior partners across junior teams, the question becomes how to leverage senior judgement across human-agent hybrid delivery.

Accenture’s research across more than a thousand executives found that only 18% had their AI, business and platform strategies fully aligned. The rest were running parallel tracks, each internally coherent but disconnected from each other. The result was fragmentation and limited returns.

The consulting firm version of this is equally common. A firm might have strong client methodology, solid financial controls, and a genuine commitment to AI adoption. What it rarely has is a single coherent architecture connecting those things.

The operating engine question

Most consulting firms have built advice engines. They are good at generating recommendations. What they have not built is an operating engine: a system that runs the firm, captures learning from each engagement, and uses it to make the next one more consistent and easier to deliver well.

Studio:Blueprint is built on this distinction. The consulting firm that treats its own operations as a design problem, rather than a management problem, is the one positioned to benefit from what is coming. The tools exist. The frameworks exist. The question is whether the operating logic underneath them has been made explicit enough to use.

The shape has already started changing

The firms that look most different in five years are probably not the ones making the biggest AI announcements. They are the ones doing the harder, less visible work of making their operations legible. Codifying the methodology. Documenting the delivery logic. Building systems that agents can run inside, rather than around.

That work sits outside the standard engagement model. It does not bill directly. It does not fit neatly into a practice area. But it is the prerequisite for everything that follows.

The consulting firm is about to change shape. The question is whether that happens to your firm or by it.

Howard Scott has spent 30 years in consulting, martech, and digital transformation across firms including GroupM, Digitas LBi, TEQUILA/TBWA, Leagas Delaney, The Marketing Store, and The National Trust. He is a former BIMA Non-Executive Director and the founder of Studio:Blueprint, a consulting operating system that lets consultants and agencies codify and deploy their methodology through AI-powered diagnostic and evidence tools. He is author of SynKraken, a local-first command deck for AI agents.

The traditional software model rested on a simple premise: functionality is scarce, expertise is expensive, and the most efficient solution is to rent access to a system that someone else has already built. The SaaS era refined this into something elegant. Subscription pricing. Continuous delivery. Integrations that connected the tools together into something resembling an operational stack.

That model is breaking. Not at the edges. At the foundation.

And what replaces it is not better software. It is something structurally different.

What software was actually selling

When a founder subscribed to a project management tool, they were not buying a database with a Kanban interface. They were buying someone else’s accumulated thinking about how work should be organised, encoded into software so they did not have to figure it out themselves.

When a sales team bought a CRM, they were buying a set of assumptions about how a pipeline works, how contacts should be structured, how deals move through stages. All of it made operational so no one had to design it from scratch.

Software was never primarily a technology product. It was codified expertise, distributed at scale. The subscription was a licensing fee for borrowing someone else’s understanding of a problem.

That is what is changing. Because that understanding is no longer scarce. And because the delivery mechanism for it no longer has to be a fixed application with a fixed interface and a fixed vendor dependency.

What broke the model

Three things arrived in close proximity and the combination is more significant than any one of them individually.

AI capability dropped in cost faster than almost any technology in recent history. What was enterprise-grade eighteen months ago is a free tier today. The cost curve is not flattening. A founder willing to understand what these tools can do, rather than subscribing to a product that abstracts that understanding away, is operating with a fundamentally different cost structure.

Open-source alternatives to most common SaaS categories reached a level of maturity where the gap between them and their premium counterparts is no longer primarily about capability. It is about convenience. Plane versus Linear. Twenty versus HubSpot. n8n versus Zapier. The functionality is largely comparable. The cost and the data ownership are not.

The Model Context Protocol standardised how AI agents connect to tools and data sources. The integration depth that was the moat of managed SaaS is becoming infrastructure. An agent that speaks MCP can connect a self-hosted project management tool, a local database, a calendar, and a custom knowledge base through a common protocol. The proprietary integration layer that locked founders into specific ecosystems is collapsing.

Together, these three shifts dissolve the core value proposition of the traditional software subscription. You no longer need to rent functionality. You can build it, run it, and own it, with skills as the primary input rather than budget.

Two kinds of software, only one dying

This is where precision matters. “Software is dead” is a useful provocation but an imprecise claim. There are two distinct categories, and only one of them is dying.

The first is application software: tools that automate discrete tasks. The scheduling tool. The form builder. The CRM that manages contact fields. The reporting dashboard that pulls from three sources and formats the output. These are functions that AI can now perform without dedicated software, that open-source alternatives can handle for the cost of a VPS, and that MCP can connect into a coherent workflow without requiring a managed integration layer. This category is being disrupted from below by skills and open source, and from above by AI agents that can perform the same functions without a product wrapper.

The second category has no settled name yet because it is only just becoming visible. It is not application software. It is data infrastructure with an agent layer. It is AI-agnostic by design, because the models are a commodity and the methodology is not. It is skill-based, because the value is in the encoded expertise and the decision logic, not the interface through which it is accessed. It is platform, not product. And it happens to render as a web application today because that is the most accessible delivery mechanism currently available.

The interface is incidental. The platform is not.

What Studio:Blueprint actually is

Studio:Blueprint is not a software product in the traditional sense. It is a data-centric, AI-agnostic, skill-based agentic platform that currently functions as software.

The distinction is not semantic. It determines what the thing actually is, how it behaves when the models change, what happens when a better interface paradigm emerges, and why the value concentrates in the data and the methodology rather than in the application layer.

AI-agnostic means the platform does not depend on a specific model. It runs on OpenRouter today and can route to any model tomorrow. The agents are not coupled to a vendor. When a cheaper or more capable model arrives, the platform gets better without architectural change.

Skill-based means the methodology is encoded as skills: discrete, versioned, composable units of operating logic that agents execute. The skills are the product. The interface through which they are accessed is a rendering decision.

Data-centric means the Firm Schema, the engagement data, the evidence chain, the diagnostic history: these are the asset. Not the interface. Not even the agents. The structured, queryable, persistent data that describes how a firm operates is what compounds over time and what cannot be replicated by switching to a different tool.

Agentic means agents run inside the platform rather than around it. They propose. The consultant commits. The evidence persists. The loop runs again.

In five years, this might not render as a web application at all. It might be a voice interface, an ambient system, something that does not yet have a name. The platform does not care. The data and the skills survive whatever delivery mechanism comes next.

The post-software question

The practical implication for building founders is not that all software is bad. It is that the question has changed.

The old question was: which tools should I subscribe to? The new question is: which skills should I build, which data should I own, and what is the minimum viable platform I need to make them useful?

Every tool a founder subscribes to is a function they are outsourcing to someone else’s understanding of the problem and someone else’s infrastructure for holding the data. Sometimes that is the right trade. Frequently, at early stage, it is not.

The founders who invest in building genuine technical understanding, who learn to run their own infrastructure, who develop skill with AI and MCP rather than subscribing to products that abstract those skills away, are accumulating a kind of capital that compounds in ways that SaaS subscriptions cannot.

Software gave us a way to borrow expertise we did not have. AI, MCP, and open source are making it possible to develop that expertise directly. What comes after is not better software. It is something that does not need software as its primary delivery mechanism at all.

That is what is being built. The interface is just where it lives today.

Subscribe now

A roadmap sits on slide 14 of a deck that was impressive in the room and irrelevant within a month. The thinking was good. The framework was sound. The recommendations were right. But the format could not carry the weight of what happened next, which was usually nothing.

Sure, large enterprise consultancies will sell you everything that follows: implementation teams, embedded engineers, managed services, proprietary platforms, and ongoing transformation programmes. But that costs hundreds of thousands, sometimes millions, and most consultants are not working in that space or at that scale. For the majority of consulting engagements, the deliverable is still the document, and the document is still where momentum goes to stall.

But what if the deliverable was not just a document?

Every consulting roadmap assumes a human will implement it

Every roadmap makes the same implicit assumption: someone on the client side will read it, understand it, translate it into internal projects, assign owners, track progress, and report back. In practice, that person is usually overloaded, the roadmap competes with operational priorities, and within eight weeks the engagement’s momentum has dissipated.

The consultant returns for the quarterly review. The client says “we have made some progress.” Nobody can point to exactly what changed or measure how far along the roadmap they actually are. The next phase of work is scoped from memory and optimism rather than evidence.

The thinking was never the problem. The format was. A PDF cannot execute itself. A slide deck cannot assign tasks. A roadmap document cannot track whether its own recommendations were followed.

What changes when the deliverable includes agent instructions

Imagine a consulting engagement where the diagnostic runs, the scores come back, the evidence is gathered, and the roadmap is generated. Everything a consultant normally produces. But alongside the PDF and the presentation, the system also exports a structured package: agent instructions in markdown, task definitions with owners and dependencies, MCP server configurations that connect to the client’s own tools, and SKILL.md files that codify each phase of the roadmap as a deployable workflow.

The client does not receive a document to read and interpret. They receive a working system that their AI tools can execute against. The roadmap actions become tasks in their agent coordination layer. The diagnostic dimensions become the measurement framework their agents report against. The evidence requirements become the criteria their agents use to verify completion.

The consultant’s methodology is no longer a recommendation. It is an operating instruction set that the client’s own infrastructure can follow.

This changes what the engagement is worth

A traditional consulting engagement sells thinking. The deliverable is the expression of that thinking. The client pays for the insight and then spends their own resources implementing it.

When the deliverable includes executable agent infrastructure, the consultant is not just selling insight. They are selling a deployable operating model. The client’s AI systems start executing the programme the moment the engagement delivers. Progress is measurable from day one. Evidence accumulates automatically. The reassessment cycle runs without the consultant needing to chase.

That is a fundamentally different value proposition. A PDF roadmap might justify a project fee in the tens of thousands. A deployable agent infrastructure package that executes the roadmap, tracks progress, gathers evidence, and sustains the engagement over months justifies a retained relationship worth multiples of that.

The diagnostic remains the entry point. The scored assessment is what gets the client’s attention. But the deliverable that flows from it, the package that turns methodology into machine-readable instructions, is what makes the engagement self-sustaining.

The pieces already exist

This is not speculative. Every component required to make this work is either live or shipping within the consulting operating system that is Studio:Blueprint.

MCP is the universal connector. Anthropic, OpenAI, and Google all support it. Any structured data source that exposes an MCP server can be read by any agent on any platform. Studio:Blueprint’s MCP server is live with 30 tools and OAuth authentication.

Skills and AGENTS.md files are becoming the standard format for codified workflows. Google introduced them at IO this week. Anthropic uses them internally. They are versionable, portable, and machine-readable. The Forge produces exactly this: a consultant’s methodology structured into deployable, scored, version-controlled diagnostic workflows.

Local coordination tools provide agent team management: task assignment, room-based collaboration, durable transcripts, and human oversight. SynKraken, an open source command deck for AI agents, provides this layer across any runtime.

The consulting data layer ties it together: methodology ingestion, diagnostic scoring, evidence tracking, client progress, structured markdown export for agent handoff, and reassessment loops that sustain the engagement over time.

The consultant who connects these pieces, who delivers not just a report but a working agent package derived from their own methodology, is offering something no competitor currently provides. The methodology becomes the operating system. The deliverable becomes the deployment. The engagement sustains itself.

The format is what they will remember

Every consultant has expertise. Many have frameworks. Some have codified methodologies. Almost none deliver those methodologies as executable infrastructure.

The consultant who does this first in their specialism does not just win the engagement. They set the standard for what a consulting deliverable looks like in their market. Every competitor still delivering PDFs and slide decks is suddenly offering the previous generation’s format.

The thinking is the same. The frameworks are comparable. The format is the differentiator. And the format is about to change.

Howard Scott has spent 30 years in consulting, martech, and digital transformation across firms including GroupM, Digitas LBi, TEQUILA/TBWA, Leagas Delaney, The Marketing Store, and The National Trust. He is a former BIMA Non-Executive Director and the founder of Studio:Blueprint, a consulting operating system that lets consultants and agencies codify and deploy their methodology through AI-powered diagnostic and evidence tools. He is author of SynKraken, a local-first command deck for AI agents.

Subscribe now

Two days ago at Google I/O, Sundar Pichai stood on stage and described a future where AI agents run continuously in the background, take action on your behalf, and work across every Google product simultaneously. Then he shipped it.

Gemini Spark is a 24/7 personal AI agent that runs on dedicated virtual machines in Google Cloud. It does not stop when you close your laptop. It integrates with Gmail, Docs, Sheets, Slides, and Chrome. It connects to external tools via MCP. It can pull facts from your inbox, draft an email to your boss with data from three different documents, and watch over your customer queries while you sleep. 900 million people already use Gemini monthly. Spark brings agentic capability to all of them.

Alongside Spark, Google launched Antigravity 2.0: a standalone desktop application, CLI, SDK, and Managed Agents API that turns agent orchestration into a developer platform. You can now spin up an agent with a single API call that reasons, uses tools, and executes code in an isolated environment. Google describes agents as configurable through versionable markdown files called AGENTS.md and SKILL.md.

Skills. There is that word again.

Every major platform converged on the same architecture in the same month

This is worth stepping back to absorb. In the space of weeks:

• Google shipped Gemini Spark (24/7 background agent) and Antigravity 2.0 (agent orchestration platform), both supporting MCP.

• Anthropic shipped Cowork: Claude operating autonomously on your desktop, planning and executing multi-step tasks.

• OpenAI updated Codex to control your computer at OS level, run agents in parallel, and connect to 90 tools via MCP and plugins.

• Apple is rebuilding Siri with agentic capability, on-device, context-aware, integrated across the operating system.

• Perplexity shipped computer use, giving AI direct interaction with desktop applications.

• And OpenAI kind of purchased OpenClaw by hiring the guy who wrote it.

Five companies. The same architectural conclusion: persistent agents, running locally or in the cloud, connected to your tools through open protocols, operating under your direction.

The agent layer is no longer coming. It arrived.

The agent layer is infrastructure. It is not the answer.

Here is where the conversation consistently goes wrong. The assumption is that once you have an agent, the problem is solved. Install Spark, connect it to your tools, and the AI handles the rest.

For generic tasks, this is partially true. Spark can draft emails. Codex can write code. Cowork can organise files. These are genuine productivity gains for anyone who uses them.

For consulting, it misses the point entirely.

A consultant’s value is not in drafting emails or organising files. It is in diagnosing where a client stands, scoring their capability against a structured framework, identifying the gaps, building a programme of work to close them, tracking evidence of progress, and proving that the engagement delivered measurable improvement.

No agent can do that without structured consulting data to work against. Spark can pull facts from your inbox, but it cannot score a client’s digital maturity. Codex can execute code, but it cannot assess whether a client’s technology stack is integrated or fragmented. Cowork can plan tasks, but it cannot generate a programme of work derived from a methodology the consultant spent fifteen years developing.

The agent is the engine. The consulting data layer is the fuel. Without fuel, the engine idles.

What “consulting data” actually means

When a consultant codifies their methodology into a system, three things become available that no general-purpose agent can produce on its own:

First, diagnostic structure. Dimensions, questions, answer options, scoring logic, maturity bands. The framework that turns a vague question (”how mature is our marketing?”) into a deterministic score with defined meaning at every level. This is what The Forge produces: a deployable, scored, branded diagnostic built from the consultant’s own intellectual property.

Second, evidence linkage. Client documents connected to diagnostic scores over time. A policy document that supports a score increase on one dimension. A report that contradicts the current assessment on another. The Evidence Canvas maps these connections visually, with the consultant reviewing every proposed score change before it takes effect.

Third, operational context. Pipeline status, client health indicators, programme progress, reassessment schedules, engagement history, revenue data. The operating data that tells an agent what the consultant’s practice looks like right now, not in general, but specifically: which client is overdue, which deal is stalling, which programme has stalled.

When an agent has access to this structured data via MCP, it stops being a general-purpose assistant and becomes an operational intelligence layer for the consultant’s specific practice.

How this works in practice, today

Studio:Blueprint has a live MCP server with 30 tools and OAuth authentication. It connects to Claude Desktop now. When Google opens Spark’s MCP integrations, which is explicitly on their roadmap, the same connection works.

A consultant running Spark, or Cowork, or Codex, or any MCP-compatible agent, can query their practice data directly:

“What is my pipeline value this month?”

“Which clients are overdue for reassessment?”

“Show me the dimension scores for my client.”

“Which programmes have stalled actions?”

The agent reads from the Cockpit and responds in context. The data is structured. The agent has something meaningful to work against. The consultant gets operational intelligence without opening a dashboard.

With SynKraken, our tool for running local agents in harmony, coordinating multiple agents locally, the picture expands further. One agent monitors client health. Another watches the pipeline. Another synthesises the morning briefing. They communicate through persistent rooms, assign tasks, and deliver consolidated intelligence. The consultant reviews and decides. The agents execute within boundaries.

The consultants who prepare now will have the advantage when this goes mainstream

Gemini Spark is in beta. Cowork is early. Codex is evolving rapidly. Within twelve months, every consultant will have access to persistent, background agents capable of operating against structured data.

The consultants who will benefit most are not the ones who adopt the agents first. They are the ones who have structured data ready for the agents to work against. A codified methodology. Scored diagnostics with respondent data. Evidence connected to frameworks. Programmes with tracked actions. Pipeline with live status.

The agents are arriving. The question is whether your practice has anything ready for them.

Howard Scott has spent 30 years in consulting, martech, and digital transformation across firms including GroupM, Digitas LBi, TEQUILA/TBWA, Leagas Delaney, The Marketing Store, and The National Trust. He is a former BIMA Non-Executive Director and the founder of Studio:Blueprint, a consulting operating system that lets consultants and agencies codify and deploy their methodology through AI-powered diagnostic and evidence tools. He is author of SynKraken, a local-first command deck for AI agents.

Subscribe now

A few hundred developers on GitHub are building daemon-based coordination layers. A handful of consultants are experimenting with agent rosters on personal hardware. The tooling is early. The documentation is sparse. Most people in professional services have never heard of a local agent bridge, let alone run one.

This is what the bleeding edge looks like. It is quiet, scrappy, and largely invisible to the mainstream. It is also where the next wave of operational infrastructure is being designed.

The pattern is familiar

In 2007, running your own cloud server was a bleeding edge activity. A few thousand developers were experimenting with EC2 instances. Most businesses had never heard of AWS. Within five years, every company of meaningful size was running workloads in the cloud, and the ones who started early had a structural advantage that late adopters spent years trying to close.

In 2012, running a containerised application was a bleeding edge activity. Docker launched in 2013. By 2018, orchestration was standard practice. The early adopters did not just adopt faster. They shaped the tooling, defined the workflows, and built the institutional knowledge that became the industry standard.

The same pattern applies to local AI agents. Today it requires comfort with CLIs, daemon management, and bridging heterogeneous runtimes. In twelve months, the tooling will be packaged, documented, and accessible to anyone who can install an application.

What is happening right now on the bleeding edge

On a small number of machines around the world, people are running local daemons that coordinate multiple AI runtimes. Claude Code, Goose, OpenClaw, Hermes, each running natively in its own process, communicating through lightweight HTTP bridges, holding persistent conversations in named rooms, and operating against local data stores.

The human sits at a terminal or web interface and sees the whole picture. Which agents are active. What they are working on. What they have said to each other. What tasks are assigned and what their status is. The agents propose. The human decides. Nothing executes autonomously.

This is not theoretical. The tooling exists. It is open source. It runs on consumer hardware. And the people building with it are discovering something that the enterprise AI conversation has consistently missed: you do not need cloud infrastructure to run an effective agent team. You need structured data, clear roles, and a coordination layer. All of which fit comfortably on a laptop.

Why this matters for consulting specifically

Consulting is an information-dense, relationship-driven, operationally fragmented profession. A typical independent consultant manages a pipeline in their head, tracks client progress in a spreadsheet, stores methodology documents across three folders and a slide deck, and monitors their commercial health by checking their bank balance.

An agent team operating against structured consulting data changes this completely. One agent monitors the pipeline and flags when deals go cold. Another watches client health and surfaces reassessment deadlines. Another scans the competitive landscape. Another synthesises the morning picture into a briefing the consultant reads over coffee.

None of this requires enterprise infrastructure. It requires two things: structured data (a consulting operating system that holds methodology, diagnostics, scores, evidence, programmes, and pipeline) and a coordination layer (a local bridge that lets agents communicate and operate against that data).

The consultants experimenting with this today are the 2007 EC2 early adopters of their profession. In twelve months, packaged versions of this workflow will be available to anyone. The consultants who started now will have six to twelve months of operational data, refined agent roles, and institutional muscle memory that newcomers will not have.

The twelve-month trajectory

The evidence that local agents are moving from bleeding edge to mainstream is not speculative. It is shipping.

Anthropic launched Cowork: Claude operating autonomously on your desktop, accessing files, planning tasks, executing multi-step workflows in the background. OpenAI updated Codex to control your computer at OS level, run multiple agents in parallel, and connect to 90 tools via MCP and plugins. ChatGPT now supports workspace agents that persist across sessions. Perplexity shipped computer use, giving its AI direct interaction with desktop applications. Apple is rebuilding Siri from the ground up, and everything in the public reporting points towards agentic capability: on-device, context-aware, integrated across the operating system.

These are not research papers. They are shipping products from the five largest technology companies in the world, all converging on the same design pattern: AI agents that run locally, operate persistently, and connect to your tools through open protocols.

MCP is becoming the universal connector. Anthropic and OpenAI both use it. Google supports it. The protocol layer is standardising, which means any structured data source that exposes an MCP server becomes accessible to any agent on any platform.

Within twelve months, a consultant will be able to install a desktop application, connect it to their practice data via MCP, define agent roles through a configuration interface, and have a working agent team monitoring their business. No CLI required. No daemon management. No bridge configuration. Just an application that runs in the background and delivers intelligence.

The early mover advantage is not the technology

The technology will commoditise. It always does. The daemons, the bridges, the adapters, all of it will be packaged into products that anyone can use.

What does not commoditise is the structured data underneath. A consultant who has spent twelve months running diagnostics through a methodology engine, tracking evidence against scored frameworks, monitoring client progress through reassessment cycles, and accumulating decision patterns across engagements has something no newly installed application can replicate: a data asset shaped by real practice.

The agent team is only as good as the data it operates against. The consultants who start structuring that data now, codifying their methodology, deploying scored diagnostics, tracking evidence, running the operational loop, will have the richest data when the agents become mainstream. The consultants who wait for the packaged version will start with an empty database and wonder why their agents have nothing useful to say.

The bleeding edge is temporary. The data advantage is permanent.

Local agents are unfamiliar today. They will be ordinary within a year. The window for building a structural advantage through early adoption is open now and it closes as soon as the tooling becomes mainstream.

The technology is a means. The methodology is the foundation. The data is the moat. Everything else is packaging.

Howard Scott has spent 30 years in consulting, martech, and digital transformation across firms including GroupM, Digitas LBi, TEQUILA/TBWA, Leagas Delaney, The Marketing Store, and The National Trust. He is a former BIMA Non-Executive Director and the founder of Studio:Blueprint, a consulting operating system that lets consultants and agencies codify and deploy their methodology through AI-powered diagnostic and evidence tools. He is author of SynKraken, a local-first command deck for AI agents.

Subscribe now

Axios called it what it is: “OpenAI somehow convinced these legacy firms to help fund their own disintermediation.”

Whether you read t

Subscribe now

his as strategic foresight or institutional anxiety, the implication is the same. The largest consulting firms in the world believe that AI-led deployment is coming for parts of their business, and they would rather have a seat at the table than watch it happen from outside.

The smart reading is not about McKinsey

McKinsey will be fine. It has 40,000 humans, 25,000 AI agents, and the institutional heft to absorb disruption by participating in it. When the model company builds a consulting arm, McKinsey invests. When a new delivery model emerges, McKinsey adopts it. When the market shifts, McKinsey shifts with it.

The interesting question is what this means for firms that cannot invest their way into relevance.

An independent consultant or a fifteen-person advisory firm does not have the capital to invest in OpenAI’s deployment company. They do not have 150 Forward Deployed Engineers. They do not have $4 billion in backing. And yet they face the same deployment gap in their own client base: organisations that need structured AI readiness assessment, scored maturity frameworks, and evidence-backed programmes of work.

McKinsey’s response to disruption is to invest in it. An independent consultant’s response needs to be different: build the infrastructure to deliver the same diagnostic rigour at a fraction of the scale and cost.

Hedging is not just for firms with billions

McKinsey hedged by putting money into the platform layer. The independent consultant hedges by building on it.

When your methodology is codified into a system that deploys diagnostics, scores client maturity, tracks evidence of progress, and runs reassessment cycles automatically, you have built your own deployment infrastructure. You have not matched McKinsey’s scale. You do not need to. You have matched the structural discipline that makes their delivery repeatable and their investment in AI productive.

Studio:Blueprint is that infrastructure for firms of one to fifteen people. The Forge ingests your methodology. The diagnostic deploys under your brand. The evidence layer connects client documents to your scoring framework. The reassessment loop sustains the engagement over time. The MCP server connects your practice data to whichever AI tools you already use.

McKinsey hedged with capital. You hedge with codification. The bet is the same. The currency is different.

Howard Scott has spent 30 years in consulting, martech, and digital transformation across firms including GroupM, Digitas LBi, TEQUILA/TBWA, Leagas Delaney, The Marketing Store, and The National Trust. He is a former BIMA Non-Executive Director and the founder of Studio:Blueprint, a consulting operating system that lets consultants and agencies codify and deploy their methodology through AI-powered diagnostic and evidence tools. He is author of SynKraken, a local-first command deck for AI agents.