On Sunday, April 26, Raphaël Dabadie published an essay on X that names what the industry has been circling for months without saying out loud. Traditional consulting, McKinsey and the four other big firms included, is not built to deliver AI transformation. The opening line of his piece is the one that should make a partner committee uncomfortable: “consulting was an optimization of the capabilities we had before AI.”

The argument is direct. Consulting is built on sampling. A small team interviews part of the organization, synthesizes what it heard, and produces a roadmap. When the problem is narrow (redesign a business unit, optimize one function, integrate two areas after a merger), that model works. AI transformation is not narrow. It touches every workflow, every handoff, every approval loop, every tool, and almost every role. It cannot be mapped by sampling. It can only be mapped by throughput, and humans do not scale on interview throughput.

The other half of the data point comes from McKinsey itself

The funniest part of Dabadie’s argument is that the central data point holding it up was published by McKinsey itself. The April 2026 Agentic Organization paper reports that 78% of companies say they use AI, and more than 80% report no measurable impact on operating profit yet. The paper’s explanation is honest. Most companies treat AI as a tool plugged into the existing flow with the button pressed. That flow was designed five or ten years ago for humans doing each step. AI accelerates one step and the bottleneck shifts to the next. Adoption is not transformation.

It is the same data point I covered five days ago when the Anthropic runtime turned commodity at $0.08 an hour. Anthropic took away the excuse that infrastructure was expensive. McKinsey took away the excuse that buying AI was the answer. But there is a third layer that until Sunday no one was naming. The consulting industry that sold itself as the help-you-redesign-your-workflow layer also does not scale to the problem. Not for lack of talent. For lack of model architecture.

The factory analogy

Dabadie cites a colleague with an analogy worth pinning to the wall. In the 1890s, factories replaced their steam engines with electric motors. Productivity barely moved. Factories were built around the central transmission shaft the steam engine required: pulleys, belts, machinery laid out radially around the shaft. An electric motor worked just as well (better, in fact), but the factory was still steam-shaped. Real productivity gains came thirty years later, when factories were redesigned with one small motor at every station and the spatial layout followed material flow, not the cable.

The exact line: “we have swapped the motor; we have not yet redesigned the factory.” That is the precise position most companies hold with AI right now. They bought Claude. They bought ChatGPT Enterprise. They bought Copilot. They glued each one onto the existing flow. The P&L needle did not move because the existing flow was built for humans doing each step.

Why traditional consulting cannot fix this

This is the structural part. Three constraints no PowerPoint can resolve.

One. Consulting bills by the project, ideally short and well-scoped. A three-month engagement with three consultants runs $600K to $1.5M depending on the firm. AI transformation does not fit in three months and does not have a clean scope. When the quote moves from $600K to $8M for a twelve-month transformation, the client’s spend committee chops it back into pieces, and you are back at square one.

Two. The roadmap the project produces ages faster than it executes. Every time a new Claude or a new GPT ships, parts of the roadmap stop being valid. The traditional consulting firm has already billed for the roadmap. It has neither the incentive nor the business model to keep it alive. The client ends up with an 80-page PDF that is useful for two quarters.

Three. Implementation gets separated from discovery. The consultant delivers the roadmap and leaves. The integrator starts from zero because they were not in the interviews. What the client ends up paying for twice is context, which is exactly the scarce resource.

The new model: software plus service

Dabadie’s proposal reorganizes both layers into a single operation. The software layer is agents doing interviews at scale. 500 people, 1,000, 10,000 if the organization is large, in parallel. The output is not a deck; it is an operational graph of how the work actually gets done. Not the wiki version. The real one. The one that lives in handoffs, workarounds, informal dependencies, email approvals, and context that only sits in the heads of three people on each team.

The service layer keeps what was actually valuable in traditional consulting. Specialized humans who bring judgment. With two important differences. First, fewer of them with more context, because the graph gets built by software. Second, they do not leave when the roadmap is done. They stay inside the operation, keeping the graph current as the company changes and as models change.

Dabadie names two roles inside the service layer. Forward Deployed Advisor (FDA) is the role the senior partner used to play, but now embedded inside the client with continuous visibility into the graph and authority to prioritize. Forward Deployed Engineer (FDE) is the person who puts hands on implementation with the FDA’s context next to them, never having to relearn the client’s business.

The model has a name in Dabadie’s piece: software plus service. Not consulting plus implementation, not agency plus SaaS. It is one operation where software does discovery at continuous scale and service contributes judgment where judgment is the bottleneck.

Why this lands now

Three things changed in April that align the conversation.

The first is runtime price. Anthropic put Managed Agents at $0.08 an hour. Before that, standing up an agent that could interview at scale inside a company required three senior engineers and two months of setup. Now it is an API. That removes the technical barrier that kept sampling as the only practical option.

The second is the maturity of judgment companies are now ready to buy. McKinsey’s own paper says 80% see no impact. The average client’s executive committee already ran the first adoption cycle and figured out that buying tools was not the answer. They are open to a second conversation, a more operational one, where someone tells them which workflow to redesign first and why. That opening did not exist eighteen months ago.

The third is opportunity cost. The piece I published yesterday covered the other dimension of the same problem: AI productivity is a codebase property, not a model property. The company that stays with traditional consulting loses twice. Once on the obsolete roadmap. And once on the codebase and workflow that Anthropic already redesigned, which beats them on iteration speed.

Five cycles of consulting watched from the inside

I have been in this 36 years. I started in 1990, at 15. I have watched the consulting industry try to claim ownership of a transformation five times. The result was similar each time.

Late-1990s digital transformation. The big firms sold decks on digital strategy while Amazon, eBay, and Yahoo built actual operations. The clients who won shipped operations. The clients who paid for the deck lost.

Mid-2000s web transformation, this time wearing a SaaS and SOA logo. Salesforce won. The consulting firms that sold Service-Oriented Architecture in PowerPoint lost their clients to Salesforce, who shipped operation.

2010-2015 mobile transformation. Apple and Google delivered stores, SDKs, revenue models. The big firms sold mobile-first as a concept. Clients who learned to operate in mobile hired internal teams. Clients who paid for the deck shipped late.

2015-2020 cloud transformation. AWS, Azure, GCP delivered operation with the price calculator on the front page. The consultancies sold cloud strategy. The companies that won shipped real migrations. The companies that lost paid for two years of roadmap and then hired an integrator to do what the roadmap described.

Now the AI transformation. The fifth cycle. Same pattern, more compressed. If the previous four gave the client five to seven years to course-correct, this one gives nine to twelve months. The compression is the only new part. The structure is the same.

What we do at IQ Source about this

AI Maestro is the software-plus-service model already in operation. The software layer is the discovery audit that produces the operational graph of how work actually happens across the organization. It is not a deck. It is a living map. When a decision changes or a new model lands, the graph updates; it does not get rewritten from scratch.

The service layer comes from the IQ Source team operating inside the company. Sustained presence is the FDA’s job: working with the executive committee, prioritizing, keeping the graph current, and handling the trade-offs an agent does not decide. When implementation needs code, the FDE comes in with the FDA’s context next to them. They do not relearn the business. They were already there.

The operational difference from traditional consulting is the same difference Salesforce had with a 2005 SOA deck. We do not sell roadmaps; we sell operation. We do not leave when the first map is delivered; we stay maintaining it. And the engagement is not a three-month scope with six months of setup. It is a monthly operation where the client pays for continuous presence and live discovery.

For software companies whose product lives in the codebase from day one, AI Maestro pairs with Tech Partner, the line that takes over when discovery identifies that the critical piece is not the operational workflow but the codebase itself. The two lines share methodology, share team, and share the discipline of not delivering PDFs but delivering live systems.

The diagnostic question for the next AI transformation invoice

The question worth running before signing the next check to a consultancy is short. If your AI program is shaped like a project with a delivery date and PDF deliverables, you bought adoption, not transformation. If it is shaped like a continuous operation with a live graph, a person owning the graph, and a team that comes in to implement using the graph as context, you bought the new model.

The difference shows up inside the first 30 days. In the old model, the first 30 days are interviews, workshops, and kickoffs. In the new model, the first 30 days are agents running interviews in parallel, a graph growing every night, and a first executive review where the conversation is already on data, not on hypotheses.

If your next internal conversation about AI transformation still has someone proposing a six-week discovery kickoff with consultants flying in, that is the conversation worth pausing. Two hours with your team, written map at the end, clear separation between what Dabadie calls the motor and what he calls the factory. No quote attached. The address is info@iqsource.ai.

Dabadie named the model on Sunday. The line that closes his piece, in his own words: “agents do everything they can do better, and humans step in only where they are truly needed. That is how this model can go deeper, move faster, and serve more companies than traditional consulting ever could.” It is not a prediction. It is a description of something already running. The question for your company is not whether the new model wins. It is how many more checks you will sign to the old model before switching.