The Rope Sellers

HHere's a fun thing that happened in October 2025. Bloomberg reported that close to 150 former consultants from McKinsey, Bain, and BCG had been quietly contracted — through a project code-named Argentum, run by a data-labeling startup, reportedly for OpenAI — to train AI models on how to do entry-level consulting work. Market sizing. First-draft decks. The grunt stuff. The pay was $110 an hour, for up to 19 hours a week.

Read that again slowly. The people who used to bill six figures a year to build slide 14 of a 60-slide deck are now moonlighting at $110 an hour to teach a machine how to build slide 14, so the machine can do it instead of the next batch of them. Lenin supposedly said the capitalists would sell you the rope you hang them with. He never imagined the capitalists would show up, hourly, to teach the rope how to tie the knot.

That's the whole story, really. But let's do the numbers, because the numbers are where it gets interesting — and where the comfortable narrative (“AI is just a tool, we'll all be fine, it frees us up for higher-value work”) falls apart.

The pyramid was always the product

The Big 4, the strategy shops, and the Indian IT giants are wildly different businesses that all secretly run the exact same machine. Call it the billable pyramid. You hire an enormous base of cheap juniors, you bill them out by the hour at a markup that would make a loan shark blush, and a thin layer of partners at the top pockets the spread. The junior does the work; the partner sells the relationship and signs the invoice. Leverage — the ratio of cheap bodies to expensive names — is the entire business model.

Agentic AI is dangerous to this model in a way that spreadsheets and offshoring never were, because it attacks both inputs of the pyramid at once. It doesn't just replace the juniors (the bodies). It compresses the hours (the unit you sell). Research, synthesis, first drafts, document review, reconciliation, journal-entry testing — that's 60 to 70% of what a junior professional does, and it's precisely the stuff a decent model now eats for breakfast. When the work gets faster and needs fewer people, you're not looking at a productivity boost. You're looking at the collapse of the thing you were charging for.

And clients have noticed. This is the part firms hoped nobody would say out loud. PwC's Chief AI Officer Dan Priest admitted to Bloomberg that clients “would hear us talking about using AI and say, ‘We want our fair share of those efficiencies.’” There's now an actual phenomenon of enterprise buyers asking the Big 4 for an “AI discount.” My favorite instance: KPMG — a Big 4 firm — leaned on its own auditor, Grant Thornton, to pass along AI savings, and got its audit fee cut 14%, from $416,000 to $357,000 (per UK Companies House filings surfaced in early 2026). KPMG saved fifty-nine grand and, in the process, handed every client on earth the script for demanding the same thing. When the arsonist starts running fire-safety seminars, you should probably listen.

An HFS Research survey of 1,002 senior executives across 16 industries and 14 countries, published in November 2025, put the mood in numbers: 65% said traditional consulting models fail to deliver real value, only 13% rated traditional consulting “highly effective,” and while 49% of contracts today are still tied to headcount, only 16% of leaders expect to be using that model within two years. HFS president Saurabh Gupta didn't hedge: “Consulting as we've known it is over. AI has blown up the model where armies of consultants spend months producing recommendations no one implements. If your consulting partner can't deliver measurable outcomes at the speed of AI, they're obsolete.”

Okay. So who survives, and why? Here's the thesis, and it's almost boringly simple once you see it: survival isn't predicted by how good your AI is. It's predicted by whether a government forces a specific, named human being to be personally, legally accountable for the output. Capability is cheap and getting cheaper. Accountability is a moat. Let's walk the ranking.

Big 4: audit is a fortress, advisory is standing in a field naked

The Big 4 aren't one business, they're two businesses in a trench coat, and AI is treating the two halves very differently.

The audit half has a moat that has nothing to do with technology and everything to do with the law. A public company must be audited, and a specific engagement partner must personally sign off, staking their license and their firm's liability on the opinion. You cannot fire the requirement. AI is genuinely transforming how the work gets done — EY's Helix, PwC's Halo and GL.ai now let auditors test 100% of transactions instead of pulling a 5–10% sample, and a PCAOB board member acknowledged in a 2025 speech that full-population testing beats manual sampling on coverage. But notice what that does. It makes audit better and cheaper without making it optional. Audit survives — as a regulated utility. Commoditized, margin-squeezed, AI-run, but structurally un-killable because the state says so. A boring fortress is still a fortress.

The advisory half has no such luck. Advisory is consulting wearing an accounting firm's lanyard, and it is standing in an open field with no moat, no mandate, and no signature requirement. Which is exactly why the cuts are landing there. KPMG cut about 4% of its US advisory staff, shut its federal audit practice (~450 roles), and — in the genuinely startling move — axed roughly 10% of its US audit partners, around 100 people, a level of seniority Big 4 firms almost never touch. Over September 2024 to May 2025, KPMG shed more than 3,300 US roles. Deloitte, posting a healthy 8% US revenue growth, still decided to slash benefits for its internal “Center” staff: parental leave halved from 16 weeks to 8, PTO down 5–10 days, a $50,000 IVF-and-adoption fund killed, and pension accruals frozen after 2026. When a firm growing at 8% is cutting your parental leave, that's not about this year's revenue. That's a firm quietly redrawing the line between who's inside the long-term contract and who's a cost to be optimized.

The tell is at the bottom of the pyramid, in graduate hiring. In the UK, KPMG slashed its graduate scheme 29% (1,399 down to 942), Deloitte 18%, EY 11%, PwC 6%. Accountancy graduate job adverts fell 44% year over year. Australian Big 4 partner ranks are down 15% — about 500 partners, of whom fewer than half were replaced — in two years. The base of the pyramid is being kicked out, and I'll come back to why that's the scariest number in this whole essay.

MBB: the brand is the moat, and the brand is quietly eating itself

Strategy consulting — McKinsey, Bain, BCG — has no regulatory moat at all. Nobody is legally required to hire McKinsey. What they have instead is brand-as-insurance (“nobody ever got fired for hiring McKinsey”) and access to the CEO's ear. That's a real moat. It's just not a technological one, which means AI can't kill it but also can't be sold as the reason it survives.

So the Big Three are doing something fascinating: they're cannibalizing themselves as fast as possible and calling it strategy. BCG reported $14.4 billion in 2025 revenue, up 7% (its 22nd straight growth year), with 25% of that — roughly $3.6 billion — coming directly from AI work, and tech-and-AI services now north of 40% of the total. CEO Christoph Schweizer's line: “AI has turned out to be highly value accretive and not dilutive for BCG.” Bain says AI- and tech-enabled work is about 30% of revenue, heading for 50%. McKinsey's internal AI, Lilli, runs over 500,000 prompts a month, has 72% of the firm using it, and reportedly delivers up to 30% time savings on knowledge work.

Here's the catch nobody at these firms enjoys discussing. If Lilli saves 30% of the time, and three associates plus Lilli now do what ten associates used to do, the firm needs fewer associates. Full stop. McKinsey trimmed around 200 tech roles in late 2025, and reports point to larger reductions ahead as it “right-sizes” back-office and non-client functions over the next two years. The associate factory — the thing that made these firms money-printing machines — is shrinking.

And the pricing model is confessing the crime. About 25% of McKinsey's global fees now come from outcome-based pricing rather than hours. McKinsey's UK managing partner Michael Birshan says “we're doing more performance-based arrangements with our clients,” and the firm's own AI leader Kate Smaje admits “many of the fundamentals of the professional services model are coming under challenge.” Translate that from consultant into English: the client no longer believes the hour is the product. When you switch from billing hours to billing outcomes, you're not innovating on pricing. You're admitting that the hour — the entire historical unit of value — has been exposed as a fiction. Outcome pricing is the ransom note the industry is writing to itself.

Indian IT: the purest pyramid, the thinnest moat, the biggest human cost

Now the part that matters most, because it involves the most people and the least protection.

If the billable pyramid has a spiritual homeland, it's Bengaluru, Hyderabad, Pune, and Chennai. TCS, Infosys, Wipro, HCLTech, Tech Mahindra, Cognizant, LTIMindtree — this is the pyramid in its purest, most beautiful, most exposed form. It was never about a signature or a license or a relationship with the CEO. It was about bodies and hours: take a huge supply of Indian engineering graduates, train them, bill them to a Western client at a fat markup, and repeat a few hundred thousand times. Labor arbitrage. That was the miracle, and it lifted millions of Indian families into the middle class.

Ask the accountability question here and the answer is brutal: there is no moat. Nobody is legally required to use TCS. There's no signing partner staking a license. Which makes Indian IT arguably the most exposed of all five industries in this essay — and the fastest to sell the rope, because when your whole business is “we'll do the labor cheaply,” pivoting to “we'll do the AI-delivery cheaply” is a lateral move, not a reinvention.

Watch the decoupling, because it's the whole story in one motion. In July 2025 TCS announced it would cut about 12,200 roles — roughly 2% of its 613,000-strong workforce — its largest reduction ever, primarily middle and senior management, even as revenue held. CEO K. Krithivasan went out of his way to insist “this is not because of AI giving some 20% productivity gains. We are not doing that… It is not because that we need less people” — which is exactly the kind of triple-denial that makes you check your wallet. Infosys paused fresher onboarding; Wipro turned cautious. Across the top five firms, reportedly 80,000+ roles vanished in the eighteen months to mid-2025. Revenue steady, headcount falling, revenue-per-employee rising. For twenty years those lines moved together. Now they've divorced, and only one of them got the house.

Then there's the quieter, crueler number. Entry-level pay at TCS and Infosys — that famous ₹3.5 lakh fresher package — has barely moved in nearly two decades. TCS still starts freshers around ₹3.36 lakh; Infosys around ₹3.6 lakh. Inflation has roughly halved the real value of that number since the late 2000s. Indian outlets now run the genuinely humiliating comparison that a skilled plumber in an Indian metro can out-earn a fresh engineering graduate at a marquee IT firm. The escalator that carried a generation upward hasn't just slowed. It's frozen, and the machine that made it move is being unplugged.

The final indignity comes from the clients themselves. The Global Capability Center — the captive in-house tech center multinationals run directly in India — is eating the outsourcers' lunch from both ends. In FY26, GCCs added a net ~200,000 people in India versus ~110,000 for the whole IT services sector, the third straight year GCCs out-hired the industry that trained their talent (per Xpheno data). There are around 1,700–1,800 of them, employing 500,000+ professionals, paying (per NASSCOM/EY benchmarks) 25–40% more than equivalent IT-services roles at the mid-senior level, and projected to be a $100 billion sector employing 2.5 million people by 2030. When Goldman Sachs runs its own 9,000-person operation in Bengaluru, that work simply never becomes a TCS contract. Clients figured out they can hire the arbitrage directly and skip the middleman's markup. The middleman built the talent pool; the client is now shopping in it — and paying more to poach from it.

Accenture is the bellwether, and the bell is loud. In one three-month stretch of 2025 it cut 11,000+ people (791,000 down to 779,000) in an $865 million restructuring, with GenAI bookings of $5.9 billion for FY2025 — nearly double the prior year. CEO Julie Sweet's line will age like milk or like prophecy: the firm is “exiting on a compressed timeline people where reskilling, based on our experience, is not a viable path for the skills we need.” Bodies out, AI bookings up. That's the template, and every Indian IT major is being measured against it.

The counterexample that proves the rule: BigLaw

Now compare all of that to the one profession getting richer during peak AI adoption. Am Law 100 profits per lawyer are up 53.7% since 2019, per Thomson Reuters and Georgetown Law's 2026 Report on the State of the US Legal Market. The fiscal-2024 numbers were obscene — $158.3 billion in revenue, up 13.3%, profits per equity partner of $3.15 million — and 2025 was better still. Law firms are adopting AI aggressively; legal AI startup Harvey went from $100 million ARR in August 2025 to roughly $190 million by year-end and a $200 million raise at an $11 billion valuation (co-led by GIC and Sequoia) in March 2026, now used across most of the Am Law 100. Its CEO Winston Weinberg says AI “isn't just assisting lawyers. It's becoming the system through which legal work gets done.”

So why is law thriving while consulting sweats? Because a lawyer signs the brief, carries personal liability, can be sanctioned or disbarred, and operates in an adversarial system where the other side is paid to catch your AI's mistakes. When judges started sanctioning lawyers for AI-hallucinated citations, they weren't slowing AI down — they were hard-coding the requirement that a licensed human verify every word. Accountability is the moat. You can't hallucinate your way past a bar association.

Here's the whole essay in one table:

Notice the ranking is almost perfectly ordered by one variable, and it isn't AI capability. Law firms don't have better AI than McKinsey. They have better law.

The problem every single one of them is pretending not to have

Here's where I stop being glib, because there's one thing all five share, and it's genuinely frightening in a slow-moving way.

The partners, principals, signing auditors, and lead architects of 2035 were supposed to be doing grunt work right now. That's the whole point of the pyramid — it was never just a profit machine, it was an apprenticeship. You did four years of soul-crushing document review or deck-building or bug-fixing, and somewhere in that misery you absorbed judgment. You learned what a weird transaction smells like, why a client is actually angry, when the model is confidently wrong. Nobody teaches that in a seminar. You catch it, like a cold, from proximity to hard, boring work.

AI just ate the hard, boring work. So the industries are gleefully cutting the exact cohort that was supposed to become the expensive, irreplaceable seniors — the accountability layer — a decade from now. The Big 4 cut grads by up to 29% into a demographic cliff where the AICPA estimates 75% of US CPAs are within 15 years of retirement and CPA exam candidates have fallen 43% since 2016. BigLaw runs entry-level hiring roughly flat and lets lateral hires (49% of associate hires in 2025) outnumber fresh graduates (38%) while everyone assumes the senior pipeline refills itself by magic. Indian IT freezes fresher intake. Consulting contracts its associate factory and pays the ones it let go $110 an hour to train the replacement.

Every one of them is optimizing a beautiful short-term margin by quietly eating its own seed corn. AI can draft the memo, run the population test, and build the deck. It cannot yet be the named human who stakes a license, a reputation, and a liability on the answer being right — and that named human is exactly the person these firms have stopped growing. The moat that protects the survivors is made of accountable senior humans, and the industry just defunded the factory that produces them.

The rope sellers, it turns out, aren't just teaching the rope to tie the knot. They're also declining to raise the next generation of people who'd know when not to pull it.

References

Consulting & the billable pyramid

• Bloomberg — Ex-McKinsey Consultants Are Training AI Models to Replace Them (October 2025). Project Argentum; ~150 ex-MBB consultants at ≥$110/hour.
• HFS Research (with IBM) — Consulting that delivers, not just recommends (November 2025). Survey of 1,002 senior executives; 65% say traditional consulting fails to deliver real value; headcount contracts expected to collapse.
• PwC / Bloomberg coverage of client “AI discounts” and KPMG–Grant Thornton audit-fee cut (UK Companies House filings, early 2026).

Big 4, MBB, and Indian IT

• KPMG / Deloitte / EY / PwC graduate-scheme and partner-rank cuts (UK and Australia coverage, 2024–2025); US advisory and audit partner reductions — same buyer pressure documented in the HFS survey above.
• Moneycontrol — K. Krithivasan on TCS's ~12,200-role cut (“not because of AI”).
• NASSCOM / EY GCC benchmarks; Xpheno FY26 GCC vs IT-services hiring — NASSCOM.
• Accenture — Q4 / FY2025 earnings: $5.9B GenAI bookings; ~$865M optimization program; Julie Sweet on compressed-timeline exits.
• BCG / Bain / McKinsey AI-revenue and Lilli / outcome-pricing disclosures (2025 firm annual reports and partner interviews) — same structural pressure as Bloomberg's Argentum reporting.

BigLaw & the accountability moat

• Thomson Reuters Institute & Georgetown Law — 2026 Report on the State of the US Legal Market. Am Law profits-per-lawyer trajectory cited in the essay.
• AICPA demographic estimates (CPA retirement cliff; exam-candidate decline since 2016) — AICPA.

On this site

• The Rope Sellers Buy a Rope Machine — Series · Part 2: Indian IT wakes up, buys partnership gym memberships, and calls it strategy.
• Vibes All the Way Down — same pyramid, different angle: what happens when orgs outsource judgment to the model.
• Anatomy of an Agentic AI System — what building the rope machine actually looks like.