The company
Software moved its cost inside the product. Almost no one is measuring it.
Margin is the economic control layer for AI-agent spend. We price every agent operation against the outcome it produced — a merged PR, a passed test, a resolved ticket — and, once a person approves it, hold the cheaper path only while it keeps quality. This page is the argument for why that job has to exist.
The shift
For twenty years, serving one more customer cost almost nothing. That assumption broke.
SaaS earned 75–85% gross margins because the marginal cost of another user rounded to zero. AI-native companies run a model on every request, at a real per-query cost that lands in cost of goods sold rather than overhead. The more of the product an agent does, the more of the P&L that cost claims.
- Classic SaaS
- ~80% gross margin · ~20% cost
- AI-native
- ~52% gross margin · ~48% compute, now in COGS
The more AI-native a company becomes, the more its profit is simply how efficiently it runs its agents.
Why now
AI has a unit for how much you spend. It has none for whether the spend paid off.
Tokens already measure consumption — how much you used. A second unit, the task, is emerging to measure how much you improved a model; Benchmark’s Everett Randle laid that out in July 2026, pointing at a task marketplace past a $2B run-rate. The unit still missing measures value: did the spend produce the outcome. That is cost per outcome. Measuring it is turning into table stakes, and enterprise players are claiming that ground now. What stays open is acting on the measurement — closing the loop, neutrally, for the lean teams those players are not built to serve.
You can measure cost per outcome in a weekend. Our framework says so plainly. The part that takes a year is trusting the number enough to change production on it, and to change it back.
Measurement is where this category starts. The defensible ground is acting on the measurement, and undoing the action the moment it stops paying.
What we believe
Three commitments the product is built to keep.
Neutral by construction.
We sell no tokens and resell no inference. A cheaper route costs us nothing, so “cheaper” is never an argument we have a reason to lose. The other tools that touch this spend mostly make money on the tokens flowing through them.
A number you can’t reproduce isn’t a number.
Every figure in the console and on this site traces to a committed run you could rerun. Sample data is labelled as sample data, everywhere, without an exception we make for a better-looking screenshot.
A person ratifies the loop.
The system proposes a change and proves it first; a human approves it; it reverts itself the moment quality drops below the bar. Autonomy is a dial you set, not a default we assume on your behalf.
The harder path
We could have shipped a router that guesses. We built the gate that checks.
The quick version of this product sends every call to a cheaper model and reports the saving. It is a weekend of work, and it is wrong often enough to be dangerous, because nobody checked the cheaper model still did the job. We made the opposite bet, and it cost us the speed. Three checks are where that bet lives:
- Five tests fail the build if a claimed saving can’t be reproduced from a committed run. A cost win that dropped the pass-rate fails the same way — cheaper-and-worse is not a win we let ship.
- Provenance reads fail-closed. A metric is real or it does not render. There is no path where a hand-typed number reaches you wearing a
provenlabel. NOT COMPUTEDis a shipped answer. When the evidence isn’t there, the honest output is the absence of a number, not a plausible one drawn to fill the space.
Measuring is a weekend. Standing behind the measurement enough to act on it, and to reverse the act, is the year. We spent the year.
Watch the loop run on real spend.
The console runs the whole thing live on open-source agents we metered ourselves: measure, recommend, act, revert — with the parity proof on every change and every number tagged real or sample.