Last week, the Financial Times reported that KPMG demanded a 14% reduction in audit fees from Grant Thornton. The reason? AI-driven efficiencies.

Grant Thornton pushed back. Their argument: “High-quality audits rely heavily on expert human judgment.” Fees represent “the cost of people plus technology support.”

They lost the negotiation anyway. KPMG’s audit fees dropped from $416,000 to $357,000 in a single year.

Here’s what struck me about that story. KPMG didn’t deploy AI to do their own audit. They didn’t replace Grant Thornton with a machine. They simply pointed at what AI can do now and said: if the work is getting easier, the price should go down.

AI didn’t have to replace anyone. It just had to exist.

The Negotiation Leverage Problem

This is the part most professionals aren’t thinking about yet.

We’ve been debating whether AI will replace us. But the more immediate threat isn’t replacement—it’s repricing. Your client doesn’t need to believe AI can do your job. They just need to believe it makes your job easier. And then they’ll expect to pay less for it.

KPMG’s move wasn’t an anomaly. It was a negotiation tactic that will become standard across every professional service. Legal. Consulting. Advisory. Audit.

The conversation in every procurement meeting will sound the same: “We know AI tools exist. We know they reduce the hours required. So why are we paying last year’s rates?”

And what do you say to that? “We don’t use AI”? That’s not a defense—that’s an admission of inefficiency. “We use AI but it doesn’t reduce hours”? Good luck selling that when the evidence says otherwise.

The Evidence Is Piling Up

While KPMG was negotiating audit fees down, a six-lawyer litigation firm in San Francisco was proving the math works from the other side.

Ad Astra Law Group had an eighth-year associate leave. Instead of hiring a replacement, they leaned on AI tools. Costs dropped 27%. Profits went up. And their managing partner, Katy Young, said something that should make every professional services firm pay attention:

“It used to take two days to draft a complaint. Now it takes me two and a half hours.”

Two days to two and a half hours. For the same deliverable. Same quality—arguably better, because the AI-assisted complaints ran to 45 pages and signaled enough seriousness that settlement responses improved.

This isn’t a Big Four firm with a million-dollar AI budget. This is six lawyers in San Francisco who decided not to hire and ended up more productive and more profitable.

Now multiply that across every professional service. Every firm that discovers it can absorb a departure without replacing the headcount. Every partner who realizes the work still gets done—faster, cheaper—with AI in the mix.

That’s not disruption. That’s arithmetic.

The Hourglass Is Forming

There’s a pattern emerging in how AI reshapes professional workforces, and it looks like an hourglass.

The top stays. Senior professionals with deep judgment, client relationships, and the ability to navigate ambiguity—they remain essential. Someone has to design the engagement, interpret the findings, make the call on ambiguous issues, and sit across the table from the CFO.

The bottom adapts. Junior roles shift from doing the work to overseeing AI that does the work. Entry-level positions become about reviewing outputs, managing agent workflows, and learning to exercise judgment—sooner and faster than before.

The middle gets squeezed. The experienced-but-executing professional—the one who’s valuable primarily because they can do the work reliably and independently—that’s where AI bites hardest. That’s the associate who didn’t get replaced at Ad Astra. That’s the audit senior whose testing can now be orchestrated by a system. That’s the consultant who’s great at applying frameworks but doesn’t yet have the client relationships or strategic judgment of a partner.

Harvard Business Review recently published a piece arguing that companies need “agent managers”—people who oversee fleets of AI agents the way a manager oversees a team. Salesforce’s agent platform already resolves 74% of customer support cases autonomously. Their sales development capacity went from 150 meetings in 30 days to over 350 in a single week.

The “agent manager” isn’t a theoretical future role. Salesforce has them now. And the skill set—AI operational literacy, systems thinking, prompt design, hybrid workflow orchestration—looks nothing like what most professional certifications test for.

The hourglass workforce means fewer people doing the work, more people overseeing the machines that do the work, and a shrinking middle where experience alone used to be enough.

What This Means for Fees

Follow the logic:

If AI reduces the hours required for a deliverable, clients will demand lower fees. That’s the KPMG story.

If firms can deliver the same quality with fewer people, they’ll either cut prices to win more work or keep prices high and pocket the margin. Either way, the competitive dynamic changes. The firms that adopt AI fastest can undercut on price while maintaining profitability. The firms that don’t adopt will be forced to match prices anyway—or lose clients.

This creates a race. Not a race to the bottom on quality, but a race to the bottom on the cost of execution.

And here’s where it connects to what I’ve been writing about. In Article 1, I built an AI orchestrator that could do a significant portion of audit work. In Article 2, I showed you that the SaaSpocalypse validated the same conclusion at scale. In Article 3, I argued that the knowledge advantage is collapsing—your auditee will soon have the same capabilities you do.

This week’s point is simpler and more brutal: the fees are falling. Not because AI replaces the auditor. Because AI makes the argument for the auditor’s current rate indefensible.

The Standards Are Coming

On February 17th, NIST—the U.S. National Institute of Standards and Technology—announced the AI Agent Standards Initiative. They’re developing interoperability and security standards for AI agents that operate autonomously for extended periods—agents that write code, manage communications, conduct transactions.

When the government starts standardizing how AI agents operate, that’s not a signal of experimentation. That’s a signal of permanence. Standards don’t get written for things that might go away. They get written for things that are becoming infrastructure.

For professional services, standardization removes the last major adoption barrier: trust. Once there are accepted standards for how AI agents operate in regulated environments, every client and every regulator will expect firms to use them. “We prefer the human-only approach” stops being a quality argument and starts being a liability.

The Journal of Accountancy reported this month that audit firms are already using agentic AI for compliance workflows, real-time reconciliation, and continuous monitoring. One firm described spending over a year building their AI implementation—standardizing workpapers, creating knowledge bases, deploying agents across their practice.

That’s not tinkering. That’s infrastructure investment. And firms making infrastructure investments in AI aren’t planning to charge the same rates for the same work.

So What Do You Do About It?

I’ve spent four weeks writing about what’s changing. This week I want to talk about what you should actually do.

Because the worst response to falling fees is to sit still and hope the market comes to its senses. It won’t. The second worst response is to panic. The best response is to start building the skills that justify your rate when the rate resets.

Here’s what that looks like in practice.

Start training. Now.

Most professionals I know have access to AI tools through their employer’s enterprise plans and have barely touched them. Your company is paying for Claude, Copilot, ChatGPT Enterprise, or some combination—and you’re using it to summarize meeting notes. That’s like buying a Formula 1 car and using it to drive to the shops.

Log in. Explore. Break things. Feed it a real work problem and see what happens. Use the tools you’ve already been given before complaining you haven’t been given tools.

And if your employer doesn’t provide access? Get your own subscriptions. The paid tiers unlock capabilities that the free versions don’t come close to—better reasoning, longer context, tool use, file analysis. If all you’ve tried is free ChatGPT, you’re forming opinions about AI based on the demo version. Playing with AI costs money. But so does irrelevance.

Stop expecting magic from a single prompt.

This is where most people get stuck. They type one question into ChatGPT, get a mediocre answer, and conclude: See, AI isn’t good enough.

That’s like opening Excel, typing a number into cell A1, and concluding spreadsheets are useless.

Working with AI is delicate. It requires deliberate prompting. It requires using reasoning models for complex problems and fast models for simple ones. It requires understanding which tool fits which situation. If all you have is a hammer, everything looks like a nail—and right now, most professionals have a full toolbox and are only using the hammer.

Learn to use reasoning. Learn to provide context. Learn to break complex problems into steps. Learn when to use Claude versus ChatGPT versus Copilot versus a specialized agent. The difference between “AI is underwhelming” and “AI just changed my workflow” is almost always how you use it, not what it can do.

Chase your own gains.

Think about the blockers in your own process. The things that slow you down, frustrate you, eat your time. The document review that takes two days. The reconciliation you dread. The report template you rebuild every quarter.

Now ask: can AI help me solve this?

Not “can AI do my entire job.” Can it fix this one annoying thing? And then the next one. And then the next.

Start chasing your own productivity gains. Your own quality improvements. Your own efficiency wins. Document what works. Build on what doesn’t. This is how you develop the fluency that will matter when the fee conversation reaches your desk.

Because the professionals who’ve spent a year experimenting will have an enormous advantage over the ones who start when they’re forced to. Compounding experience is real.

Who Captures the Value?

There’s a tension here that Steve Yegge captured perfectly:

Scenario A: You decide you’re going to impress your employer, and work for 8 hours a day at 10x productivity. You knock it out of the park and make everyone else look terrible by comparison. Your employer captures 100% of the value from you adopting AI. You get nothing. Everyone hates you. And you’re exhausted.

Scenario B: You decide you will only work for an hour a day, and aim to keep up with your peers using AI. Nobody notices. You capture 100% of the value. But your company goes out of business, because a competitor took them out while everyone was slacking off.

The answer has to lie somewhere in the middle. And I think that middle ground is exactly what the fee compression story is about.

The value from AI adoption can’t flow entirely to the employer (burnout). It can’t flow entirely to the employee (company dies). And it can’t flow entirely to the client (firm becomes unprofitable). It has to be shared.

But here’s the thing about shared value: it flows to whoever has leverage. And leverage comes from capability. The professional who understands AI deeply—who can design workflows, interpret outputs, and exercise judgment that the machine can’t—that person has leverage. They can negotiate from strength. They can demonstrate value that justifies their rate even as the commodity work gets cheaper.

The professional who hasn’t built that capability? They’re on the wrong side of the negotiation. They’re Grant Thornton arguing that high-quality work requires humans, while the client points at the tools and says: prove it.

The Series So Far

Four weeks ago, I had my Oh Fuck moment—the personal realization.

Three weeks ago, I showed you the orchestrator—and connected it to the market’s $285 billion reckoning.

Two weeks ago, I argued that your auditee doesn’t need you anymore—the knowledge advantage is collapsing.

This week: the fees are falling. The economics are being rewritten. And the only sustainable response is to start building the skills that make you worth more than the hours you bill.

Ready to start? I’m building a structured playbook for knowledge professionals navigating this transition — from recognizing what’s changing to building the skills that justify your rate. The Knowledge Professional’s AI Playbook →

If you’re watching fees compress in your profession—or if you’re the one doing the compressing—I want to hear from you. Get in touch.

Sources and further reading: