Test an existing agent before it is expanded, and produce evidence that a release decision can rely on.
Agent release assurance is a four-to-six-week assessment that tests an existing AI agent against the system of record, its permissions and repeated trials before it is expanded.
Who it is for
The AI platform lead, quality lead or security lead. You have an agent in pilot or production, built by your team or a vendor, and someone is asking whether it can be trusted with more.
What you get
- Outcome tests against the authoritative system state, not the agent's transcript.
- Permission and denied-action tests, including cross-tenant checks.
- A regression suite with repeated-trial reporting, so one success is not mistaken for reliable behaviour.
- Trajectory review of a sample of runs.
- A reproducible evidence pack and a remediation backlog, ordered by severity.
What we test for
A retrieved document that tries to redirect the agent. An action that targets the wrong tenant. An approved action changed before execution. A retry that duplicates a transaction. A success report with no real change. More access than the task needs. An evaluator that sees sensitive content. The full list and the control behind each is on How you can check us.
How you can check us →Duration and commercials
Typically four to six weeks. Fixed price for the assessment. Remediation, if you want us to do it, is scoped separately so the assessment stays independent.
Boundary
An assessment is not a legal certification and we do not call an agent "safe". Where independence matters, assessment is delivered separately from implementation, by people who did not build the system.
Questions buyers ask
How long does agent release assurance take?
Typically four to six weeks. The assessment is fixed price. Remediation, if requested, is scoped separately so the assessment stays independent.
Is agent release assurance a certification?
No. It is an assessment that produces a reproducible evidence pack and a remediation backlog. It is not a legal certification.
What is tested?
Outcomes against the authoritative system state, permission and denied-action tests including cross-tenant checks, repeated trials for consistency, recovery after injected failures, and a trajectory review of sampled runs.
How to start
Tell us what the agent does, which systems it touches and what decision is waiting on the result.
Request an assessment →Related client stories
Client stories are published as permissions are confirmed. Until then, the method we hold every story to is on How you can check us.
How you can check us →