The firm that stopped billing hours
Tenex, an applied-AI engineering firm that has grown to 75 people serving Fortune 2000 and mid-market customers, pays its engineers for the work they deliver instead of the hours they spend at a keyboard. On Latent Space's podcast this month (published Oct 21, recorded at AI Engineer New York), co-founder Alex Lieberman explained the logic plainly: if AI makes an engineer faster, an hourly rate punishes the speed. CJ Hess, the firm's special projects lead who walked through its delivery system, added the detail that matters most to buyers — project plans and decisions get checked into Git alongside the code, so the record of why survives the work.
Per the same episode, the feared failure modes — padded estimates, code shipped fast to rack up points — have not been the biggest problem in practice. The model works. That is your signal: this is not a fringe experiment, and the way your dev shop or agency prices work is going to change with it.
This is not a build-vs-buy question. It is about the labor you already hire — and whether the productivity dividend lands in your invoice or stops at the vendor's margin.
Why hourly rates were quietly protecting you
Hourly billing has one honest feature buyers stop appreciating the day it disappears: when work goes slowly, the vendor's inefficiency comes out of their margin, not your pocket. A capped budget is a natural ceiling, and when AI collapses the delivery hours, competitive pressure pushes hourly rates down.
Outcome pricing changes where the dividend goes. A vendor who quotes you a fixed price for a defined outcome keeps every efficiency they find. That can be perfectly fair — they took the risk, they found the speed. But it is only fair if you bring the three terms below to the table. Without them, "outcome-based" does not mean the work is done. It means the argument starts.
The three terms to demand
1. Acceptance criteria written before the work starts
Objective, testable, attached to the contract — named pages, named behaviors, named integrations, what "done" does under real load. Not "a modern website," but the specific things you will check. This single document is what makes "outcome" mean something, and it does double duty in an AI-delivery world: if the vendor's agents build your project, the acceptance criteria are the only thing separating "shipped" from "demonstrated once, in front of a salesperson."
2. A revision-bound cap
How many rounds of "not quite" are included, and what rounds beyond the cap cost. Unlimited "we iterate until you are happy" sounds generous and is priced at the vendor's worst case — which you pay for. A named cap with a named overage price is honest arithmetic on both sides, and it kills the two classic failure modes: the vendor declaring victory at your first "looks good," or the meter running on round nine.
3. Full context ownership — repo, plan, decision log
You get everything a successor needs to continue the work tomorrow: the repository, the plan, the specs and prompts, the decision log, admin credentials, integration configs. Tenex checks plans and decisions into Git beside the code precisely because agents can then read not just what was built but why the team built it that way. Demand that same record be delivered to you in a form you can hand to the next vendor — or to your own agent when the maintenance contract ends and the tooling becomes yours to run. A deliverable without its context is a black box you will pay to reopen.
Make the savings reach your invoice
If their delivery costs dropped, the question of who keeps the difference is a negotiation, not a law of nature. When comparing bids for the same defined outcome, ask each bidder to quote it two ways: a time-and-materials estimate and a fixed outcome price. The gap between the two is their efficiency assumption — and your negotiating room. Add a re-benchmark clause at renewal too: in an AI-delivery world, outcome prices should trend down, not up, and your contract should notice.
Honest boundaries
- Expect to pay a premium for outcome pricing. It moves delivery risk onto the vendor, and vendors price risk. Demanding hourly-level prices with outcome risk on their side does not get you a deal — it gets you change orders and scope fights dressed up as "outcomes."
- Not everything fits an outcome price. Discovery, strategy, and genuinely open-ended design cannot be priced as outcomes yet — pay time for that phase, cap it, and flip to outcome pricing the moment the target is defined. CJ Hess's point that the real work is increasingly shaping customer context into a form the model can use is worth taking seriously: that labor is real, and paying for it honestly keeps the relationship real.
- Small jobs burn the same way. A $5,000 build with vague acceptance criteria fails exactly like a $50,000 one. All three terms scale down without any added cost.
Tonight's version
Pull your most recent vendor invoice — agency, dev shop, designer, whoever builds for you. Draft the acceptance criteria for the next job before you send the brief, then add the revision cap and context-ownership clauses to your contract template. Twenty minutes, and the next "outcome-based" quote you receive arrives as a price instead of a riddle.
Want the templates? The Boring Automation Pack includes the acceptance-criteria worksheet and the vendor-terms checklist — the same three clauses, ready to paste into your next contract.
Top comments (0)