I keep running the same uncomfortable math

A few weeks ago I was looking at METR's latest time horizon report. The headline number is that the length of tasks frontier AI agents can complete with 50% reliability has been doubling roughly every seven months for the past six years. Current frontier models handle work that takes human experts around an hour. If the trend holds, autonomous agents handling longer-horizon tasks become plausible within the next few years.

I am not someone who reads these reports and panics. I run a community of product leaders. I consult. I use AI heavily in my own work every day. I am, by most definitions, on the right side of this curve.

And yet, I keep running the same uncomfortable math.

The direction, not the prediction

The argument is not that knowledge work disappears next year. The argument is about direction. If the marginal cost of competent analysis, writing, code, design, and synthesis keeps falling, then what those activities can command in the market falls with it. Not to zero, not soon, but the trend line points down.

Anthropic's latest Economic Index report shows Claude usage diversifying across an increasing range of tasks, with augmentation (AI as a thinking partner) rising in both consumer and API traffic. The average hourly wage of tasks performed on Claude.ai sits around $47.90, well above the U.S. average of $37.30. Translation: the bulk of people using these tools are exactly the workers whose work the tools are getting better at doing.

This is the part nobody wants to say plainly. The more capable AI gets, the more leverage shifts from people who do the work to people who own the systems that do the work. From labor to capital.

Why this changes the planning question

If you believe even a softer version of this, the rational response is uncomfortable. The asset that produces income most reliably in this scenario is not your skill or your time. It is capital. Equity in a business. Ownership of something productive. Cash that can be deployed.

Which means the honest question for someone in their 30s or 40s is not "how do I stay relevant" but something closer to "how do I accumulate enough capital, fast enough, while my skills still command a premium, to be okay if the premium compresses."

I find this uncomfortable to write. It sounds mercenary. It also sounds like the kind of thinking that, taken to its conclusion, hollows out the meaning of work itself. But pretending I am not running this math would be dishonest.

Where I might be wrong

I should name what I might be wrong about.

The METR trend could flatten. Reliability at long tasks is much harder than 50% success rates suggest, and the gap between "AI can do this" and "AI can do this in a way you would actually trust" has consistently been larger than predicted. Daron Acemoglu, in The Simple Macroeconomics of AI, has argued that aggregate productivity gains from AI will be modest for years. The Economic Index data itself shows people are using AI more as a collaborator than a replacement, which is not what pure substitution looks like.

And the "stockpile capital now" instinct has a problem of its own. If everyone with foresight does it, the edge compresses. The actual durable advantage in a capital-dominated world is probably not cash. It is owning productive things: equity in businesses, infrastructure, networks, IP that compounds.

What I am actually doing

I do not have a clean answer. I am doing a few things in parallel: building businesses that compound (Product Circle, the consulting work), trying to be at the frontier of using AI rather than competing with it, and being more deliberate than I used to be about ownership versus income.

These are my current thoughts, but sometimes I wonder if I am going crazy and what other people are seeing.

So I want to ask, genuinely.

If you accept the direction, even partially, what are you actually doing differently? And where do you think I am wrong, or too pessimistic, or missing something I should be paying attention to?