Cut the agents.

You are paying for AI seats nobody uses and agents that don’t work. Temno measures both — every seat, per person; every agent, what it costs and whether its work is accepted — and tells you which to keep, fix or cut.

$192,680

a year on seats nobody uses and agents that never ran or whose work gets redone, in the sample company on the next page.

Agents

3 keep · 2 fix · 3 cut8 agents · $34,400 a year on the cut ones · synthetic company

Sample agent ledger: agent, team, runs in the window, people using it, share of outputs accepted, cost per run, cost per year and verdict.
AgentTeamRuns, 90 dPeopleAcceptedCost / runCost / yrVerdict
Lead-enrichment agentRevOpsnever0——$18,000CUT
Weekly KPI narratorFP&A13238%$5.00$12,300CUT
Code-review agentEngineering8601433%$1.20$4,100CUT
Invoice-matching agentFinance1,240961%$2.10$10,400FIX
Support triage agentSupport21,0004669%$0.09$7,600FIX
Meeting recap agentSales3,42011882%$0.14$1,900KEEP
HR policy answer botPeople4,10030277%$0.05$820KEEP
Contract summariserLegal210691%$0.60$500KEEP

Seats

500 people hold 980 seats across three tools · 368 hold more than one · $158,280 a year idle · eight of the rows

Sample seat ledger, per person: seats held, seats used in the window, what the idle ones cost per year, and the verdict.
PersonSeats heldSeats usedIdle cost / yrVerdict
007230$1,560CUT
000720$1,080CUT
000610$360CUT
038631$1,200FIX
000521$720FIX
005333—KEEP
000222—KEEP
003611—KEEP

One word per seat, one word per agent

Agents · by what people do with the output

Cut:

Never ran in the window, or under 40% of its outputs accepted. Cut; the dollars are on the report.

Fix:

Runs, but 40–69% accepted, or costs more per accepted output than the person did.

Keep:

Runs, and 70% or more of what it produces is accepted as is.

Seats · by days since the person last used it

Keep:

Used in the last 30 days. Stays.

Fix:

Idle 30–89 days. Nudge or reassign before renewal.

Cut:

Idle 90 days or more, or never used. Cut at renewal.

Thresholds are printed on every report and can change per engagement. “Accepted” means used without rework, read from the platform’s own logs; where a platform can’t show it, the report says so and grades lower. A 30-day export can’t prove 90 idle days, and the report says so.

How it works

  1. Inventory — Every seat and every agent you pay for, who holds or owns it, and what it costs.
  2. Logs — Who uses which seat, how often each agent runs and for whom, from the exports and logs the platforms already keep.
  3. Output — What people do with the results: accepted as is, edited, or redone.
  4. Cost — Fully loaded per run and per year, including the time spent checking the work.
  5. Verdict — Keep, fix or cut, per seat and per agent, with a confidence grade and the thresholds printed.
  6. Agent P&L — For the big ones: realised return against the human baseline, not a projection.

Temno doesn’t build or sell agents, so it can tell you which to cut.

Fixed fee. The method is printed on every report.

Free waste estimate

Tell us roughly how many seats and agents you pay for. Within two working days you get what is probably idle or not earning its keep, what it costs a year, and what confirming it would take.

We reply once, by hand. No list.