AI Is Eating the Billable Hour

The phrase is a16z's. They ran it as a chart this week, and it is the right phrase, so here it is with the argument underneath it.

Consider what an hourly contract actually says. It says: the longer this takes, the more I pay you.

That was tolerable when effort and output were roughly proportional. A developer who worked twice as fast was maybe twice as expensive, the market cleared, and the arrangement was inefficient only at the edges.

Now a competent operator with a good harness does in twenty minutes what used to take a day. Under an hourly contract, that operator has just cut their own revenue by roughly 95% for producing an identical outcome. The one who refuses to adopt gets paid more.

As one investor put it this month, in the age of AI, paying people hourly is complete and utter misalignment.

The exposure is concentrated exactly where you would expect: agencies, law, accounting, consulting, outsourced development. Every industry whose unit of account is an hour is now selling something that is getting cheaper to produce while pretending it is not.

Where the money goes instead

Three things replace it, and they are not equally good.

Fixed scope. A defined outcome for a defined price. The vendor keeps the gains from getting faster, which is exactly the incentive you want, and carries the risk of getting it wrong. Works whenever the scope can actually be written down, which is more often than people claim.

Outcome pricing. Payment per closed ticket, per reviewed contract, per qualified lead. The cleanest alignment available and the hardest to agree, because both sides have to trust the measurement. Worth the negotiation on high-volume work and rarely worth it below that.

Retainer for access. A monthly fee for someone's judgment and availability rather than their hours. This is what most good advisory relationships already were, dressed in a timesheet because procurement wanted a number to divide.

What all three share: none of them pay more when the work takes longer.

My own position, stated plainly

This is how nativefirst is priced, so read the argument knowing that.

A First Build is $8,000 for two weeks and one shipped agent. If it takes four days instead of ten, the price does not move. An Install is monthly, for embedded access, not for hours logged. There is no timesheet in either.

That is not a virtue. It is self-interest that happens to point the same way as yours. I am paid to make the loop work, and every hour I save is mine to keep. Under an hourly contract I would be paid to take longer, and you would have to trust me not to.

Better to remove the temptation than to trust anybody about it, including me.

What to do on Monday

Take your largest hourly contract and ask one question.

If this vendor doubled their speed tomorrow using AI, what happens to my invoice?

If the answer is that it halves, the contract is working and your vendor is the one exposed. They will need to move to fixed or outcome pricing to survive, and the good ones already are.

If the answer is that nothing happens, they simply bill the same hours, then you are already paying for a productivity gain you will never see. That gain is real, it is happening, and under your current contract it accrues entirely to them.

The hour was always a proxy for value. It was a decent one for a century. It stopped being one about eighteen months ago, and most contracts have not caught up.

Pay for the outcome.