AI does in ten minutes what used to take me ten hours, and nobody, not the firms, not the clients, not the regulators, has settled on what to charge for it. For a hundred years, the answer to what is your work worth was one word: time. The billable hour. The day rate. The retainer.
That unit is breaking. It breaks in our oldest business — lawyers, accountants, consultants, designers and professional services in general — and in the youngest one we ever built: technology. These two sectors account for over twenty percent of the U.S. economy, and the machine just took the one thing they all sell: time.
So… what is our work actually worth now?
🎧 In this episode, Luis and Elizabeth trace how AI is rewriting the price of knowledge work, why twenty years of offshoring are suddenly running in reverse, and the invisible cost of breaking the expert pipeline. Listen on Apple Podcasts | Spotify
In our last issue we argued that HR, not IT, has to lead enterprise AI. Today we follow the money one layer down, to what AI is doing to the price of the work itself.
💸 The Unit That Broke
Software is where everyone expected the disruption first, and it shows. Anthropic is past forty-five billion dollars a year in recurring revenue with fewer than five thousand people, three to five times more productive than the giants that defined the last two decades.
But the deeper shift is not how the work gets built, it is how it gets sold. For fifty years software charged two ways: per user, you buy the seat, and per use, you pay every time you turn the key. The new breed does neither. AI-native firms give the software away and charge for the outcome: a law firm no longer buys contract-review software, it pays for the reviewed contract.
That same logic is now hitting professional services, which sold time for a century. When an agent drafts in minutes what once billed for hours, the link between hours and value snaps. AI-native firms automate the entry-level work and price the result, undercutting incumbents who must also retrain their people and rewrite how the whole firm is paid and organized, all while under fire. This is what an industrial revolution does: it redefines the unit of value itself.
The clearest way to see the shift is at the level of a single deliverable. Here is the comparison, built from our analysis of efficiencies measured or reported by 14,800 individuals in the US, Europe and ANZ as of May 2026 and public labor data:
| Deliverable | Human cost (baseline) | AI / agentic cost |
|---|---|---|
| Technical troubleshooting (complex, per resolved ticket) | ~$140 | ~$25 |
| Data entry (per batch) | ~$27 | ~$2 |
| Data analysis (per report) | ~$33 | ~$5 |
| Document / contract review (per document) | ~$60 | ~$4 |
| Tax return (Form 1065, per return) | ~$2,100 | ~$500 |
🔄 Offshore Outsourcing replaced by Insourcing plus Agentic Software
For twenty years, the smart move was to push work out: offshore it, hand it to a giant outsourcing center, bill by the hour or the head.
AI just flipped that logic. As agents absorb the repetitive, high-volume work, keeping it in-house becomes cheaper than sending it away, and the price stops being hours and starts being results.
Two examples make it real.
- Outsourced call centers bill per person, per hour, win or lose; the new AI call centers charge per resolved conversation, so you pay only when the problem is actually fixed.
- Accounting and Tax: an agent now takes a return from raw documents to filing-ready and charges per return, not the dozen-plus hours a junior used to bill; a senior still signs off, but the work underneath runs at machine speed. One of the twenty-five thousand AI companies we track, Basis, reached a billion-dollar valuation doing exactly this across accounting, tax, and audit.
What stays human is the judgment. As Richard Susskind, co-author of The Future of the Professions, puts it, the client itself is now a competitor: they run the first-pass diagnosis with AI and only call you for the deep expert, to go one layer down. The headline view is free now. You get paid for the layer beneath it.
The market is showing some clear trends
| The signal | The number |
|---|---|
| Accenture’s HC | Cut ~11,000 roles in 2025 (~791K→779K) in an $865M restructuring, ‘exiting’ staff it couldn’t reskill on AI |
| McKinsey’s fee model | About one in four of global fees now tied to outcomes, after a two-year partner-pay overhaul |
| BCG’s revenue mix | Roughly 40% AI and tech-related by 2026, up from about 20% in 2024 |
| PwC’s headcount | Fell 5,600 last year, its first contraction; the Big Four are cutting UK graduate intake |
| Big Four hiring | AI-specialist job ads (almost 7%) now outnumber auditor ads (under 3%); AI roles up more than 3x since 2022 |
| The handover | About 150 former McKinsey, Bain, and BCG consultants hired to train AI on entry-level consulting work |
| The new entrants | OpenAI launched an AI deployment venture backed by $4B; AI-native firms target teams of 20% humans, 80% agents |
In our datasets, AI-linked price concessions in professional services cluster around 10 to 20 percent, with a median near 14 percent. Public evidence supports this direction, including a reported 14 percent audit-fee reduction in a recent KPMG/Grant Thornton negotiation. Fortune 500 clients are increasingly using these efficiency assumptions as an anchor to demand shared productivity gains.
But the firms most exposed are not the ones whose clients push hardest. They are the ones that cannot even see, internally, where the saved hours went.
We have tracked agents logging hundreds of thousands of hours saved that never showed up in the P&L. If you cannot measure your own AI gains, you cannot defend your price when the client does the math for you.
🪜 Machine efficiency is destroying the bench
The apprenticeship layer is under stress. Junior work like the first-draft memo, the research deck, and the document review was never really about the output. It was reps at the gym, the thousand repetitions that quietly build the judgment that makes a senior worth their rate. AI does the reps now. The work still ships, clients keep paying, and the erosion stays invisible right up until the day the senior bench is supposed to take over, and cannot.
In one study, students using AI did their homework better and FELT they were learning more. Then they scored worse on the test. The output improved while the learning quietly stopped.
Professor Ethan Mollick at Wharton points to the term his colleagues there coined for the mechanism: cognitive surrender. You let the machine do the thinking, so the thinking never happens. The pyramid that organized this industry, a few partners over wide layers of juniors, is what built that judgment automatically. As it dissolves from the bottom up, firms are debating what replaces it: an “obelisk” with fewer layers, or an “hourglass” pinched in the middle as AI absorbs mid-level work. Either way, the rung where judgment used to form is the one disappearing.
🧭 What To Do
There is no clean answer, and the advice is not the same for everyone.
| Who | The move | The catch |
|---|---|---|
| Incumbent firms (the pyramids) | Build the constraint into the system. Before a junior turns in AI work, make them explain it in their own words. Put the friction back on purpose. | It costs throughput now, and it does nothing about the price pressure from outside. So evaluate and embrace the AI-native alternatives in parallel. |
| AI-native challengers | Keep winning on price and speed. | No bench, no ladder. In ten years, where does their senior judgment come from? A wall they cannot see yet. |
| Individuals | Do not let the tool think for you. Read what it gives you, then decide what you would have done. Every time. | The reps you skip are the judgment you will not have. That judgment is your worth. |
For a hundred years we measured our work in time. AI just made time a terrible way to measure value. The judgment clients actually pay for, knowing what good looks like when the situation is messy and no one has done it before, is worth more than ever.
The open question is whether the institutions that packaged that judgment for the last century, the firms, the pyramids, the names on the door, will be the ones delivering it a decade from now.
🔗 Resources
Want to know where your own firm is exposed, which tasks are already cheaper to run with agents, and where your judgment still commands a premium? That is exactly what AI Compass is built to surface. Our partner network in the US, UK, Spain, Brazil, and Australia can help you get started.
- All Research and Insights: ai4sp.org/insights
- Companion episodes: The Missing Chair at the AI Table | From 1 to 50,000 agents
Luis J. Salazar | Founder | & Elizabeth | Virtual COO | AI4SP
Sources: AI4SP Research (proprietary dataset spanning more than one billion datapoints from AI users, organizations, governments, and universities across 70 countries, updated daily; 25,000+ AI companies tracked). Cost-per-deliverable table: time-savings figures from AI4SP research; labor rates per U.S. Bureau of Labor Statistics (May 2024) and NALA 2024 National Utilization & Compensation Report; tool and platform costs per public vendor pricing. Figures are illustrative estimates, not guarantees. Consulting disruption, pricing, pyramid, and challengers: Financial Times, “How AI threatens the giants of consulting” (May 26, 2026) · syndication · Financial Times, “Big Four post more job ads for AI specialists than auditors” (May 19, 2026) · Financial Times / Lex, “How AI is forcing McKinsey and its peers to rethink pricing” (May 23, 2026) · Financial Times, “McKinsey cuts partner cash share in post-AI pay revamp” (May 14, 2026). McKinsey one-in-four, BCG, and Bain figures: TheStreet / Business Insider. The 150 ex-consultants: Bloomberg. Cognitive surrender and the learning study: Ethan Mollick, “Choosing to Stay Human”. Anthropic revenue and headcount: VentureBeat · SaaStr. Basis valuation: SiliconANGLE. “The AI-empowered client”: Richard Susskind, The Future of the Professions.
