Two days ago, this column ended with a question: what produces the senior consultants of 2035, when the apprenticeship work that builds judgment is being absorbed by AI?
On Monday, the AI labs answered. Or half-answered.
Anthropic — together with Blackstone, Hellman & Friedman, and Goldman Sachs — announced a $1.5 billion joint venture to embed forward-deployed AI engineers directly into mid-sized companies. OpenAI announced the structurally identical move the same day, with Bain Capital, TPG, Advent, Brookfield, and SoftBank. Both ventures bypass the consulting intermediary entirely. Engineers from the labs, augmented by acquired AI consultancies, are inserted directly into client operations. The model is Palantir's, scaled and capitalized.
This is not disruption rhetoric. It is a specific, channel-locked structural answer to Tuesday's question. The labs are not waiting for consulting firms to figure out how to rebuild their production function. They are acquiring the senior judgment they need from elsewhere — their own engineering ranks and the AI-consultancy acquisition market — and deploying it directly inside the mid-market companies their PE partners already own.
It is the acquisition model at industrial scale.
This was the development I argued Tuesday was structurally available but uncertain in timing. The timing turned out to be shorter than I thought.
But the announcement also sharpens the original argument in a way that matters more than the urgency does.
Engineering infrastructure and domain decision infrastructure
The labs have built engineering infrastructure for AI deployment. Models, embedded engineers, integration capability, distribution. What they have not built — and have no structural reason to build — is domain decision infrastructure. The two are not the same thing, and the consulting industry's pipeline problem is mostly the second.
A forward-deployed AI engineer, embedded inside a regional health system, can build Claude into the workflows the clinical staff already uses. That is real, valuable work. There is a genuine shortage of people who can do it well — Goldman Sachs' Marc Nachmann said exactly that when announcing the venture. The shortage is binding, the new firms will be busy, and the model will work.
What the embedded engineer cannot do is tell that same health system whether its current bid posture toward a major payer is leaving five points of margin on the table because a comparable system in a comparable market negotiated a better contract three years ago. That is domain decision work. It comes from the kind of pattern recognition Tuesday's piece described — built by working ten or fifteen comparable situations under the supervision of someone who has worked a hundred.
The engineering problem and the domain problem share a root cause. AI is compressing the apprenticeship work that produces senior people across both. But the two problems have different solutions, and only one of them is being solved this week.
The engineering problem is being solved by capital. The labs have an engineering bench. They are acquiring AI consultancies to scale headcount. PE-backed mid-market companies have the urgency to pay for the integration. The market is clearing.
The sharpest objection to this distinction is that the labs will hire the rare individuals who carry both — former operating executives who can also build production AI systems. Those people exist. The labs will compete aggressively for them, and they should.
The objection is correct on the cases but wrong on the structure. The hybrid profile is rare for a specific reason: it requires roughly twenty years of operating judgment plus modern AI engineering capability. The first part is built through the apprenticeship pipelines this column has argued are breaking down — including, in many cases, the consulting firm apprenticeships themselves. The labs hiring from this pool are not solving the apprenticeship problem. They are running on the stock of senior judgment that other systems paid to produce. For the next decade this works. For the decade after, the question is what produces the next cohort of hybrid hires when the pipelines that produced this one are no longer running.
The domain problem is not being solved. It is, if anything, being made worse, because the most accessible AI deployments — workflow integration, document synthesis, RPA replacement — do not require domain pricing, procurement, or commercial decision capability. The engineering deployments will appear to work. The domain decisions inside those same companies will continue to be made by people whose own apprenticeship pipeline is eroding for the reasons Tuesday's piece laid out.
The fourth response that consulting firms cannot pursue
Tuesday's piece offered three options for firms that recognized the apprenticeship problem and had the structural ability to act: slow the substitution deliberately, externalize senior judgment into teachable systems, or buy and partner for the judgment they could no longer grow.
Monday's announcement introduces a fourth response that consulting firms cannot pursue: bypass the apprenticeship production function entirely by recruiting senior engineering judgment from somewhere consulting firms cannot recruit from.
The labs can do this because they have an engineering bench and an acquisition market. Consulting firms have neither. There is no equivalent senior pricing-judgment market they can recruit from at industrial scale. There is no $300 million Sequoia round funding a "senior commercial-strategy partner acquisition vehicle" they can bid against.
It is also worth noting where the capital is coming from. The Anthropic venture is funded by the same private equity industry whose ownership of large parts of consulting makes apprenticeship investment structurally hardest, as Tuesday's piece argued. PE has now placed a $1.5 billion bet that the answer to its own consulting problem is to bypass consulting rather than reform it.
For the engineering layer, the production-versus-acquisition choice has been made for the labs. They chose acquisition, capitalized it heavily, and locked in a distribution channel that gives them a multi-year head start. Consulting firms competing in that layer are competing against a structurally better-funded model.
For the domain layer — pricing, procurement, commercial strategy, the work consulting firms have actually been built to do — the choice is still open. But it just got narrower.
It got narrower because the labs' embedded engineers will be inside mid-market clients within months. While they are there, those clients will start asking them adjacent commercial questions. The engineers will not be qualified to answer them, but they will be present, and presence is most of distribution. The consulting relationship that used to start with a commercial diagnosis and expand into adjacent work now has a competitor in the next office.
The window to stand up an answer to Tuesday's question — what produces the senior consultants of 2035 — just shortened by the time it takes the labs' embedded engineers to start asking pricing questions on behalf of their clients.
Three options, sharpened
Slowing the substitution deliberately becomes harder. Partnership firms are now competing against ventures that have explicitly chosen to substitute, at $1.5 billion of scale, in their core mid-market territory. A firm that absorbs the productivity hit to preserve apprenticeship is making that choice in a market where its largest mid-market competitor is doing the opposite. The choice was hard before. It got harder this week.
Externalizing senior judgment into teachable systems becomes more important, not less. If the apprenticeship production function cannot be defended at full strength, the answer is to extract the senior pattern recognition that exists today and embed it in deployable form before that knowledge leaves with retiring partners. This is harder than it sounds. Senior pricing or procurement judgment is tacit, built through hundreds of decisions where the partner saw what mattered and what was noise, often without being able to fully articulate why. Decomposing that into operational decision architecture is real work. Most attempts produce checklists. The work to do it well exists, but not at scale. The labs' embedded engineers will be inside client operations doing real work. The question is whether they are surfacing the commercial decisions back to a firm's actual senior judgment — embodied in some structured, deployable form — or making those decisions themselves with whatever ad hoc reasoning the model produces.
Buying or partnering for judgment that cannot be grown internally becomes the most defensible move on a short clock. Acquiring a boutique whose senior partners carry the pattern recognition the firm is losing solves a five-year pipeline problem in eighteen months. The integration challenge is real, but it is a smaller problem than rebuilding apprenticeship from scratch in a market where the labs are eating the bottom of the pyramid. The firms that move first on this will pay less. The firms that wait will be acquiring against the labs themselves, who will increasingly want domain assets to round out their engineering teams.
The Princeton senior, again
Tuesday's piece opened with a Princeton senior who turned down a top consulting offer because they didn't see how to be integral as an analyst whose work AI was already doing. Two days later, with the labs deploying $1.5 billion against direct competition with consulting on the engineering layer, that detail looks different. The Princeton senior was reading the market more accurately than the firms were. They saw the integral work was moving — not to consulting, but to the venues where engineers and operators were building AI integrations directly into the businesses that needed them.
The engineering layer has found its answer. Embed the engineers, capitalize the bet at $1.5 billion, lock in the distribution channel.
The domain layer still has none.
The firms that build the domain decision infrastructure first will define what consulting becomes after AI. The firms that wait will inherit whatever shape the labs' embedded engineers leave for them — and the Princeton seniors of the next decade will not be choosing them either.