A Princeton senior, asked recently whether they'd accepted an offer from a top consulting firm, said no. The reasoning, in their own words: "I want to be integral, and right now it's not very clear that you will be integral as an analyst anymore."
The senior consultants I know would dismiss this as one student's opinion. I think it's more right than they want to acknowledge.
McKinsey now operates with roughly 40,000 human employees and 25,000 AI agents. Bob Sternfels, the firm's global managing partner, expects those numbers to be roughly equal by year-end. He has said publicly that consultants at the firm are "moving up the stack," taking on more complex problems while AI agents handle the synthesis, charting, and document preparation that used to be the work of junior analysts. The firm saved 1.5 million hours of human labor on this kind of work last year alone.
I take Sternfels at his word on the productivity story. The AI tools work, McKinsey is genuinely doing more with smaller teams, and there's no point arguing with the gains. What I want to ask is the question Sternfels has not yet answered.
What produces the senior consultants of 2035?
The production function for judgment
Senior consultants are not born. They are made, slowly, by doing the work that develops pattern recognition. The pricing partner who walks into a complicated commercial situation and immediately sees which factors matter and which are noise — that person did not arrive with that capability. They built it by doing dozens of pricing analyses as a junior, then more of them as a manager, then watching hundreds of client situations unfold from a partner's seat. Each one added a few cells to a tissue of judgment that, by the end, looks like instinct but is actually compressed experience.
That tissue of judgment is what clients pay senior consultants for. Not the deck, not the framework, but the ability to look at the deck and the framework and know which 15% of it is wrong in ways that matter.
The training that builds that ability comes from doing the work AI is now doing.
This isn't fundamentally a labor substitution story. What's being eroded is not consulting's delivery economics. It's the mechanism by which the industry has historically reproduced its own expertise. The pyramid was never just a staffing model. It was a production function for judgment, and AI is now compressing the bottom of it in ways that change what the top of the pyramid is made of.
McKinsey's own description of how its internal AI platform Lilli has changed the work makes this concrete. Scoping decks that once took two days of junior analyst effort now emerge in under three hours. That work is gone, at one firm, in one year, to the tune of 1.5 million hours.
A reasonable objection: weren't most of those hours useless toil anyway? Some of them, yes. Formatting and copy-pasting that built no judgment and should have been automated decades ago. The disappearance of that work is a gain. But not all of those hours were toil. Some were the analyst staring at a competitor's pricing data trying to figure out why it didn't add up, and the analyst who got the answer right developed a sliver of pattern recognition that the analyst who didn't, didn't. That cognitive struggle was where the learning actually happened. It's what's now being removed alongside the toil, and the firms have not separated the two.
The new model describes junior consultants as "supervising AI outputs" and "guiding agent workflows." The unspoken assumption is that supervising AI output develops the same judgment as doing the work originally did. There's no evidence for that assumption. The pattern recognition that lets a senior consultant catch a wrong answer was built by making wrong answers and being corrected by partners who could see the mistake. Reviewing an AI's output develops a different and lesser capability: spotting obvious errors in machine-generated work, which is not the same thing as constructing a defensible answer from scratch.
This won't be visible at the firms for years. The current senior bench is fully formed; clients are getting the analyses they expect. The problem will surface only when today's partners begin to retire and the firms reach for the people who would have replaced them, only to find those people were never produced. By then the output will look the same on the surface, but the senior judgment that used to interrogate it before it reached the client will be thinner and increasingly absent. The firms will keep operating. They may keep growing. What they sell will be quietly different from what they sold in 2020, and only a few clients will notice until much later.
Ownership structure decides what's possible
This isn't happening uniformly, though, and the pattern is worth being honest about. The structural pressure to under-invest in capability reproduction varies sharply by who actually owns the firm.
Partnership-owned firms — McKinsey, Bain, the senior-led specialist boutiques — face a real but resolvable trade-off. The partners are the owners. Investing in apprenticeship is investing in the future value of their own equity. The partner who funds pipeline development today is funding the firm her successor will inherit. The decision to under-invest is a decision to take cash out of her own future. It's a hard decision but a rational one, and partnerships can, and sometimes do, make it the other way.
Public and PE-owned firms are a different story. Accenture, the Big Four operating subsidiaries, the parts of the consulting market that have rolled up under outside ownership. The owners aren't the partners. The pressure is quarterly margin and EPS. AI substitution drops to the bottom line in the current period. Pipeline erosion shows up in 2032, on someone else's watch. The structure makes apprenticeship under-investment not a temptation but close to inevitable. The CEO who refuses to cut juniors to fund pipeline development will be replaced by one who will.
This isn't a moral observation. It's a structural one. Public-firm consulting may be unable to solve this problem regardless of what individual leaders prefer. Partnership-firm consulting can solve it, but only if partners accept lower current cash distributions to fund the investment. Both choices are hard. Only one is structurally available.
Three options for firms that can act
For the firms that recognize the problem and have the structural ability to act, three options exist. Each is a different way to redesign how judgment gets built.
Slow the substitution deliberately. Keep some apprenticeship-quality work in-house even when AI could do it faster. Accept the productivity hit on the engagement to preserve the pipeline investment. This requires partners to absorb the cost in their own distributions and to defend the choice when peers are not making it. It's the most expensive option in the short run and the most reliable in the long run, and it's realistically only available to partnership firms.
Externalize judgment into teachable systems. If apprenticeship can no longer scale through accumulated case experience, the alternative is to extract senior pattern recognition from the people who hold it and embed it in deployable form: decision architectures, structured methodologies, diagnostic frameworks, supervised AI reasoning systems. Done well, this preserves the methodology even if the apprenticeship pipeline weakens, and it gives junior consultants something concrete to interrogate rather than just a stream of AI outputs to ratify. Done badly, it ossifies into checklists that produce confident-but-shallow work. The category is hard. It's also the only one of the three options that's structurally available regardless of ownership model.
Buy or partner for the judgment you cannot grow internally. Acquire boutiques whose senior partners carry the pattern recognition you're losing. Build alumni networks that retain access to retired senior judgment. Partner with academic institutions or specialist advisors. This isn't a complete solution. The integration of acquired firms typically destroys some of what was bought, but it's a way to bridge the gap while the longer-term options play out.
These options aren't mutually exclusive. The firms best positioned five years from now will be doing all three, deliberately and at the same time, while their competitors are doing none of them.
The question for senior consultants
The Princeton senior who turned down the consulting offer wasn't making a statement about AI. They were making a statement about what consulting firms now appear to offer. "I want to be integral." They didn't see how to be integral as an analyst whose work AI was already doing.
That's the leading edge of a recruiting problem the firms haven't yet acknowledged. The smart undergraduates and MBAs who used to choose consulting because it offered the steepest learning curve in business are increasingly choosing other paths. The firms can replace them in the short run. They cannot replace, in the long run, the judgment those people would have developed.
The question for senior consultants reading this isn't abstract.
Five years from now, when the analyst who would have been your trusted senior associate doesn't exist — because they were never hired, or never given the work that would have trained them — what is your firm's actual mechanism for producing the judgment your clients are paying you for? Not your stated mechanism. Your actual one.
The firms that have an answer will be fine. The firms that don't will continue to operate, and no one will notice what's missing until the people who would have caught the mistakes are gone.