Software moved its cost inside the product. Almost no one is measuring it.

Margin prices every agent operation against the outcome it produced and, once a person approves it, holds the cheaper path only while it keeps quality. When no cheaper model holds, it says so: on our founder’s own agents it recommended a dearer one.

AI moves cost off the fixed overhead line and onto the variable one that scales with every request.Figure 1

The more AI-native a company becomes, the more its profit is how efficiently it runs its agents.

Classic SaaS80% gross margin20% costAI-native52% gross margin48% compute + COGS
analysts call it the OpEx-to-COGS inversionclassic SaaS drawn at 80%, inside the 75–85% bandexternal research, not our meter

Source: ICONIQ, 2026 State of AI (Bessemer puts the margin band 50–60%); Cloud Capital, Cost of Compute 2026; PwC, 2026 Global CEO Survey.

why now

AI has a unit for how much you spend. It has none for whether the spend paid off.

Measurement is where this category starts. The defensible ground is acting on the measurement, and undoing the action the moment it stops paying.

Tokens measure how much you used. The next unit counts what the spend produced: cost per outcome, the number we meter, neutrally.

One unit sits past all three, and nobody owns it. Cost per outcome is a counter: a 99%-accurate extraction and an 85% one score as one outcome each, so it can never tell you a costlier route earns its bill. That takes what an outcome is worth, a judgment only you can make, and one your own records already hold.

Cost per outcome is the honest default of a value-per-outcome model. Until you price an outcome, every figure Margin computes is cost per outcome, and it is labelled as one. Price yours and the meter records it beside the cost, every outcome counted once and unweighted. A value you did not supply is never invented.

what we believe

We could have shipped a router that guesses. We built the gate that checks.

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.

Neutral by construction.

Our revenue does not rise with your bill. We sell no tokens and resell no inference, so “cheaper” is never an argument we have a reason to lose. Most other tools that touch this spend make money on the tokens flowing through them.

A number you can’t reproduce isn’t a number.

Five tests fail the build if a claimed saving can’t be reproduced from a committed run, or if it cost pass-rate. Provenance reads fail closed, and when the evidence isn’t there the answer is NOT COMPUTED, never a plausible figure.

A person ratifies the loop.

The system proposes a change and proves it first; a human approves it; when quality drops below the bar it opens the revert as a second pull request. Autonomy is a dial you set, not a default we assume.

who, and how far along

One founder, taking on three design partners.

Every number on this site traces to a run you can open, and says whose run it was.

Margin is built in Chicago by Subh Mukherjee. What exists is the product, the tests that keep its numbers honest, and metered runs of open-source agents we operate ourselves: the source of every live figure on the console.

The next step is three teams with agents in production who meter them through Margin for a pilot. If that is you, write to subh@trymargin.io, or find me on X at @subh_mukherjee.

Watch the loop run on real spend.

The console runs it on open-source agents we metered ourselves, every number tagged real or sample.