Drawdown Is Not Adoption

Sierra announced this week that it is a launch partner in the OpenAI Marketplace. The line that matters for anyone who owns an AI budget is that eligible customers can now build agents with Sierra using their OpenAI commitments. I think that is a good deal for both sides, and I think it is about to make one bad number look better than it should.

The short version

Per Eric Eyken-Sluyters on the Sierra blog, eligible customers can apply a portion of their OpenAI commitment toward building with Sierra, whose first selling point on the same page is paying for outcomes, not tokens. swyx's AINews recap of DevDay adds that the marketplace also lets enterprises apply commits to open models through Baseten. So a commitment is now easier to consume, and the drawdown percent will rise. In my own per-seat sunset, 12 accounts were on plan, paying committed minimums, and using less than the minimum, and we lost two of them. Behavior beats commitment. Kill commit utilization as the adoption number and report three lines: committed and drawn, outcomes bought with the price of each, and active utilization flagged separately.

What was announced

Sierra's post is short. It is a launch partner in OpenAI's new B2B marketplace. Eligible customers can build agents with Sierra using their OpenAI commitments, out of their existing budget. It builds on Sierra's participation in OpenAI Frontier, and it lets Sierra reach enterprise customers looking to apply a portion of their OpenAI commitment toward building with Sierra.

The same post makes Sierra's usual case, and the first item is the pricing: pay for outcomes, not tokens. A loan refinanced, an insurance claim paid, a sale closed, a customer retained.

The DevDay recap in swyx's AINews gives the wider frame, from the free portion of the issue. The B2B Marketplace lets enterprises apply OpenAI commits to open models via Baseten, and the recap relays one reading of the move: OpenAI competing to own the enterprise AI budget.

Put those together. A spend commitment made to a model company can now buy outcomes from an agent company. That is a sensible thing to offer. A buyer with a commitment would rather spend it on refinanced loans than let it sit. I wrote about why Sierra's pricing works in The Sierra Playbook: What They Actually Do Differently, including Bret Taylor's point that token usage and value are not strongly correlated.

Neither source says commitments are going unspent. I am not claiming that either.

The number this makes prettier

Every AI budget with a commitment in it has an easy number. Committed, consumed, percent. It is the first line of the slide because finance already has it.

A marketplace gives the commitment more places to go. The percent goes up. And the percent is the same whether the commitment was drawn by an agent that refinanced loans or by a pilot nobody opens.

The 12 accounts

I have been on the vendor side of that exact number. It is Break 1 in Field Report: What Broke When We Killed Our Per-Seat Tier.

Eighteen months into moving a $40M ARR business from per-seat to hybrid outcome pricing, our migration tracker had 12 accounts classified as on plan. They had migrated on paper. They were paying the new committed minimums. Their actual usage was below the committed minimum, so there was no overage, and their account teams had not escalated because the dashboards said on plan.

When the legacy tier sunset, those 12 did not notice. Three weeks later they went looking for the old product surface and it was gone. The escalation was angry and we lost two of them.

The lesson I wrote at the time was three words: behavior beats commitment. Migrated is not the same as using. The tracker needed active utilization as its own dimension, separate from committed minimum, and it did not have one.

A committed minimum is a fine thing. It is one of the five terms I would not sign an outcome contract without, and The Pricing Migration Sequence: An 18-Month Quarterly Playbook calls it what it is, the floor under the forecast. A floor under the vendor's forecast. It was never evidence that the customer got anything.

Kill it, and report three lines

So retire commit utilization as the number that says the AI budget is working. Keep it as a finance number and put two lines under it.

Committed and drawn. Finance needs these and they are already in the contract.

Outcomes bought. A count, and the price of each. Loans refinanced, tickets resolved, claims paid. If the vendor prices on outcomes, as Sierra does, this line comes off the invoice. If the vendor prices on tokens or seats, somebody has to build it, and that somebody is product.

Active utilization, flagged separately. Anything paid for and not used gets its own row and an owner. This is the line my tracker was missing.

If the second line is blank, the first line is a payment schedule.

The vendor-side mirror of this, where a meter gets stacked on a seat and nobody moves to the outcome, is in Kill the Per-Call Agent Meter.

What to do this week

Ask whoever owns your model-vendor commitment for the three-line version before the next renewal conversation. If you sell agents, put the outcome count on the invoice so your customer can fill in line two without asking.

It belongs to the argument on AI Business Models: software priced per seat is priced against a labor cost that AI removes, the pricing has to move to the outcome, and the margin structure moves with it. A budget measured in drawdown has not made that move yet. It has changed who gets paid.

Related answer: Is commit drawdown a good measure of AI adoption?

Sources: Sierra joins the OpenAI Marketplace, Eric Eyken-Sluyters, Sierra, September 29, 2026. [AINews] OpenAI DevDay 2026, swyx, Latent Space, September 30, 2026, free portion only. Field Report: What Broke When We Killed Our Per-Seat Tier, falkster.com, May 11, 2026.