This week the Financial Times ran a Big Read on a shift the consulting industry has spent two years politely refusing to name: artificial intelligence is eroding the structural advantages that let a handful of giant firms dominate professional services for the better part of a century. The detail that captures it best came from Mark Bunker, a former Deloitte senior advisory partner now building his own boutique, who said he is deliberately designing teams that are “twenty per cent humans, eighty per cent AI agents.” Read that ratio again. It is not a productivity tweak. It is a different species of firm.
The supporting evidence has been arriving all month. Private equity has poured hundreds of millions into AI-native advisory challengers, several founded by ex–Big Four partners. Smaller firms report growth of up to fifty per cent while the giants trim graduate intake — one shed more than five thousand roles last year. McKinsey now ties roughly a third of its work to performance-based fees and is rewiring partner pay; rivals are racing to hire “forward-deployed engineers” to close the gap. The AI labs themselves have begun selling directly into the consulting market the incumbents thought they owned. Even the symbolism is unkind: a recent report from one of the Big Four was flagged as carrying the hallmarks of AI “slop” — fabricated citations and invented figures — from a firm that sells governance advice for a living.
What is actually breaking
It is tempting to read this as “AI replaces consultants,” but that misses the mechanism. What is breaking is one specific, beautiful, ninety-year-old machine: leverage. The pyramid model works because armies of junior analysts generate billable hours, and those hours fund partner profits. The whole structure rests on two assumptions — that synthesis work is scarce and slow, and that you bill for time. AI detonates both. The synthesis a generalist analyst produces — read everything, summarise it, build the deck — is exactly what a capable model does in minutes for the price of a coffee. Remove the scarcity at the base and the pyramid loses its foundation; remove the billable hour and you remove the meter the whole business runs on.
The pyramid didn’t sell expertise. It sold leverage. AI is the first thing that ever attacked the base instead of the apex.
Three consequences follow, and they are already visible. Pricing migrates from inputs to outcomes — clients increasingly refuse to pay for hours and insist on results. The moat moves from scale to its opposite: deep specialisation, genuine judgement, and the trust that only relationships carry. And the middle gets crushed. Boutiques are agile and cheap to run; the giants still own sector depth, restructuring scar tissue and the reassurance a nervous board pays a premium for. It is the broadly-capable mid-tier — big enough to be expensive, generic enough to be automated — that has nowhere to stand.
The cost nobody is pricing in
There is one danger the cheerleaders are ignoring. The junior analyst was never only cheap labour; the pyramid was also the apprenticeship. Grinding through the unglamorous work is how a twenty-four-year-old becomes someone whose judgement is worth a fortune at fifty. Replace the base with agents and you optimise this quarter while quietly defunding the next generation of senior judgement. An industry that automates its own training ground may find, a decade out, that it has plenty of agents and no one left who learned the hard way what the agents are getting wrong.
The long view
For anyone working independently, this is not a threat — it is the most favourable structural shift in a generation. The same agents that hollow out the pyramid hand a solo advisor the delivery capacity that once required a floor of analysts. The constraint was never talent or insight; it was leverage, and leverage has just been democratised. What cannot be democratised is the scarce thing the giants were always actually selling underneath the head-count: earned judgement, real relationships, and the willingness to put your name on an outcome and be accountable for it. AI commoditises the synthesis and inflates the value of taste and trust. The winners of the next decade in advisory will not be the largest firms or the cheapest tools. They will be the smallest viable units of genuine judgement — and they have never had better weapons. For those who take time seriously, the instruction is simple: own the part of the work that no agent can sign its name to.
