Most teams cannot say how many agents they run, let alone what each one costs.
Once calls are landing, the shape falls out of the data: teams, the environments they run in, the agents inside them, the operations each performs. Your AI workforce drawn from what actually ran, not a diagram someone keeps current by hand: accurate to the last metered call, never a promise of “always live”.
See the org chart, at a buyer’s scale ↓Nobody drew this diagram. It derives at build time from the call rows of the three committed artifacts behind this site’s figures. The levels are the ones our estate actually has — a workflow, its task class, the routes inside it; yours would carry teams and environments above these. Three workflows is a small estate. It is also a real one.
- margin-oss-seed (aider-edit, crewai-solve, crewai-pipeline-solve) fit-scoring$1.04
- gpt-4o 332 calls$1.04
- highlightmagic-tape sidekick-tagging$0.0699
- claude-sonnet-4-6 · incumbent 116 calls$0.0535
- claude-haiku-4-5 · candidate 116 calls$0.0164
- docpipeline-reasoning reasoning-answer$0.1322
- gemini-2.5-flash · full thinking budget 115 calls$0.0970
- gemini-2.5-flash · thinking budget 128 115 calls$0.0352
One object, two ways. Every bar is that node’s share of the estate’s total spend, so the org chart above is already the Money Map — the next part is this same tree, ranked by dollars and opened where the bill hides.
is_simulated=false in all threecaptured 2026-08-19 → 2026-09-19Source: the committed artifacts provenance/{oss_cost_per_outcome_usd, autoroute_defended_savings_frac, reasoning_effort_savings_frac}.json. Every node, count and bar derives from their call rows at build time; none is typed.
The org chart, at a buyer’s scale.
Our own estate is three workflows. This is the same inventory drawn on a stand-in company, simulated and labelled so: 5 agents across 4 environments, $59,988.70 over 30 days, with the bill and the cost per outcome on every level. Pick a team, then an environment; open an agent and its operations are priced one by one.
Every node carries its own cost and its own outcome, so the org chart and the Money Map are one object seen two ways. The inventory keeps what stopped: the one agent that produced nothing in 30 days and stays on the list, marked silent, because a chart that hides what stopped is the wrong answer dressed as a tidy one.
What you run
5 agents across 4 environments · 1 produced nothing in 30 days
Completeness unknown — no provider invoice banked, so we cannot say what share of your real spend this covers.
- SIMULATED sample estate — every figure is generated by Margin's demo model, not metered from a real deployment.
- Team and environment are labels a real deployment sends. Here they belong to the sample, chosen to show the org chart.
- Only agents Margin meters appear here. On a real estate, a shared provider invoice says what share of the bill that covers; without one we say it is unknown.
Open the full console →The wage bill, the Money Map and the fix Margin would make first are one click on. Nothing above is enacted: the estate is simulated, and it says so first.
- The Meter: measures every call, its cost, tokens, substrate, and whether it worked.
- The Estate: the org chart of your AI workforce.
- The Money Map: where the spend goes, decomposed by step.
- The Parity Gate: proves the cheaper route held quality before it ships.
- The Governor: acts with a human ratifying, and a revert armed; if quality slips it reverts and the loop re-runs.
Watch the loop run on real spend.
The console shows the loop on open-source agents we metered ourselves: measured spend, recorded face-offs and the gate’s verdicts, every number tagged real or sample. No live route has slipped yet, so the revert has not fired.