Your Competitors Are Choosing. Are You?

In May 2026, Jack Clark, a co-founder of Anthropic (the company behind the Claude AI models), stood in front of an Oxford audience and said: “Change is inevitable. Autonomy is not.” He was talking about people. He could just as easily have been talking about firms.

He told the audience that when he came back from paternity leave in February, his own company looked different than the one he had left. Engineers who used to write code were now managing fleets of Claude instances. The bottleneck had moved from production to verification. The company felt like it had ten times the headcount it actually had. The work hadn’t gotten smaller. The humans had moved up the stack, to the places the AI couldn’t yet reach.

That’s not a forecast. That’s a snapshot from inside one of the companies that is exploring the AI-changed future. The rest of the economy sits somewhere on a line between that posture and what Clark called retreating from the present: pretending the technology isn’t here, that it won’t reach you, that this round will plateau before it does.

I work with leaders of services firms, the kind in the 10-to-100-person range, and I’ll tell you what I see with no editorial flourish. The firms doing well right now are exploring. The firms losing ground are retreating.

The technology isn’t the variable. The choice is.

This piece is about how to tell which one you’re doing, and what it quietly costs you if you’ve already chosen the second one without saying so out loud.

What exploring actually looks like inside a firm

It helps to anchor the word in something concrete, because “exploring AI” has been used to mean everything from a strategy offsite to a free ChatGPT login.

Anthropic is the loud version. Per Clark’s own account: engineers managing Claude instances rather than writing code by hand, the bottleneck moving from making the thing to validating that the thing is right, an organization of a few thousand people behaving like one many times its size. I don’t raise this as a model to copy. You are not Anthropic, and neither am I. I raise it because it is the cleanest available proof that AI changes how work is structured, not just how fast it gets produced.

The quieter version is happening in professional services, and it’s the one that should get your attention, because it’s closer to your business. There was a piece this spring framed as “AI is forcing the big consulting firms to rethink pricing.” The framing was a little naive, and people inside the industry said so: top-tier consulting moved off the pure billable hour a long time ago. The honest version of the story is more interesting. Serious transformation work has been priced on outcomes, not hours, for the better part of a decade. What AI is doing is accelerating the migration of more and more work into that bucket, and pushing the size of those outcome-based engagements up, not down.

Put those two pictures next to each other and they’re the same shape. Commoditize the analytical work. Price the part that survives. For Anthropic, the part that survives is judgment about what to ship. For a consulting firm, it’s execution and the knowledge of a specific client’s situation. Both get priced on outcomes, because both are reliable enough to stake a price on.

That’s exploring. It isn’t buying a tool. It’s noticing that the technology moved the reliable part of your work, and repricing around the new line.

What retreating actually looks like

Retreat is rarely a decision anyone announces. It’s a set of reasonable-sounding sentences that give a leadership team permission to keep doing what it was already doing. I hear the same four, almost verbatim.

We looked at AI. It’s not ready yet. Usually this means someone tried a chatbot once in 2024, it hallucinated, and the firm filed the whole category under “monitor.” That wasn’t an evaluation. It was a single bad first date that the firm has used to justify staying home for two years. The capability got materially better every quarter in the meantime, and the firm hasn’t looked again since the day it was disappointed.

Our work is too specialized for AI. This one is almost always partly true and dangerously over-applied. The judgment in your work is genuinely hard to automate. The production work that surrounds the judgment, the drafting, the formatting, the first-pass research, the reconciliation, is exactly what the technology is eating. Confusing the two, treating the whole job as if it were all judgment, is how a firm talks itself into standing still.

We’ll let competitors prove it first. This sounds prudent and is the most expensive posture of the four. By the time a competitor has “proven it,” they haven’t just validated the idea. They’ve built the workflow, trained their people, and signed clients at the new price point. Fast-follower works in a market where the leader stands still. It does not work in a market where the leader compounds.

Our clients aren’t asking for it. True, right up until the day it isn’t. Clients don’t ask for AI. They ask for faster turnaround, a lower fee, or a richer deliverable, and they take it from whichever firm figured out how to provide it. The retreating firm sees a price-sensitive client. The exploring firm sees a structural shift in what the work should cost.

I want to be careful here. Every one of those sentences is said in good faith by a smart leader. None of them is stupid. The trap is that each one is locally reasonable and globally a way of not deciding. Naming the trap is not the same as insulting the people caught in it.

What that posture is actually pricing

I want to come back to the third retreat, the one about letting someone else go first, because I heard it in its purest form not long ago.

I sat with the chief executive of a company doing a few hundred million dollars a year. Sharp, candid, not remotely a technophobe. His position on AI was simple: he did not want to be the one who spent the money to learn the new thing. Let someone else fund the experiments, take the losses, and discover what actually works. He’d adopt the proven version later, once the risk had been wrung out of it. To him, being first meant being the one who pays to be wrong.

Here’s the thing. He was pricing a world that no longer exists.

In the world he was picturing, a new capability meant one of two things: a long, expensive internal build, or a vendor walking in with a finished platform, a sales team, and a per-seat price. He’s right that both of those are expensive and risky. But the bought platform carries a second cost people forget. When you buy the thing that came through the door, you reshape your operation to fit the software. You pigeonhole your actual problem into whatever the product happens to do, and you live with the awkward fit for years. Most of us have done it.

That is no longer the only option. The cost of building something custom, fit to your specific problem, has fallen through the floor. You can build to the problem now instead of buying a solution and bending the business around it. The expensive, bet-it-all, first-mover software project he was so reasonably avoiding is not the thing on offer anymore.

He was declining a price the market had already stopped charging.

That’s the quiet flaw in “let them prove it first.” It assumes the cost of moving is what it used to be. For most services work, it isn’t, and the gap between what he thinks it costs to explore and what it actually costs is exactly the room a faster competitor is moving into.

The pricing tell

If you want one observable thing to check in your own firm, check how you price.

Pricing follows reliability. You can only price something on the outcome when you’re confident enough to stake the price on the outcome. That single sentence connects a software company charging fifty cents per resolved support conversation to a consulting firm charging a percentage of the value a transformation creates. Same logic, wildly different scale.

The pricing tell. When a firm refuses to consider outcome-based pricing on work that could support it, it’s telling you something: that it doesn’t quite trust its own delivery, that the work is too uncontrolled to commit to, or that it has quietly become so commodified the firm doesn’t believe the outcome is worth what it would have to charge. All three are signals worth sitting with.

I made this same argument for law firms in an earlier piece on legal billing. When the production cost of the work compresses but the billing model doesn’t move, the firm simply captures less of the value it creates. The arithmetic doesn’t care what industry you’re in. Legal was an early and visible case. It is not a special one.

Changing your pricing model is not mainly about getting paid more, though sometimes it does that. It’s a way of telling the market, and yourself, that you stand behind the result. That’s the explore move. Refusing it, when the work would support it, is the retreat.

How to tell which one you’re doing

Three questions. They’re uncomfortable on purpose.

1. Re-evaluate against your real work. When did your firm last formally test AI against your actual deliverables? Hand three standard ones to a current model with a serious prompt and a senior person’s review, and ask what fraction is now genuinely AI-doable. If it’s been more than six months, you’re closer to retreat than you think; the capability moves faster than your memory of the last time you checked.
2. Stress-test the plateau. What changes in your firm if the technology gets thirty percent better next year, and sixty percent better the year after? If the honest answer is “nothing,” you are betting on a plateau. Every plateau bet on this technology in the last three years has lost.
3. Find the slowest, costliest step. Where is the slowest, highest-cost-per-hour step in your delivery, and what would your firm look like if it were five times faster and you could redirect the time it freed? Most firms haven’t asked, because the answer is unsettling. The exploring firms ask it relentlessly, then reorganize around the answer.

None of these is a technology question. They’re operating questions, which is the whole point. The tools are the easy part. Knowing what to change because of them is the hard part, and it’s the part that doesn’t come in a box.

The part you choose

The choice between exploring and retreating isn’t made in one board meeting. It’s made in a hundred small decisions about who you hire, what you charge, what work you take, what you invest in, and what you wave off as hype. Most of those decisions get made without anyone naming what they’re actually choosing.

The retreating firm doesn’t lose all at once. It loses slowly, in ways that are easy to file under ordinary noise, and then it loses fast. The early signs look like nothing in particular. Clients leaning harder on fees. A quiet drift of your best junior people toward firms that feel more current. A strategic plan that reads almost word-for-word like the one from two years ago. Each one is easy to explain away on its own, right up until they stop being explainable.

The exploring firm doesn’t have to be Anthropic. It has to be honest about what changed, and willing to restructure pricing, staffing, and offerings around the change rather than bolting a tool onto the side of the old shape and calling it done. That kind of redesign is harder and less glamorous than buying software. It’s also the only version that holds.

Change is inevitable. Whether you stay autonomous through it is the part you choose.

If you’ve read this and you’re not sure which one your firm is doing, that’s worth a conversation. Get in touch.