AI agent monitoring and operations: watched, re-tested, reported
An agent that worked in March can drift by June: a policy changes, a model version changes, the cases change. We run your agents on the Wise Agents platform, watch them, re-test them before any change goes live, and send you a plain report each month. Deployers keep real duties under the AI Act; this service gives you the evidence to meet them.
What the engagement covers
- Agents we built, and agents moved onto our platform after a readiness review.
- Hosting, monitoring, cost limits, re-tests, change control, incident handling and the monthly report.
- The run records kept for the period your contract sets.
- Good fitsAgents we built, and agents moved onto our platform after a readiness review.
- What stays yoursThe people who oversee the agent day to day, your instructions, and the decision to keep using it. The AI Act puts those on the deployer (Art. 26).
What you get, in writing
- Monitoring of every conversation and run, with the cost against a daily limit you set.
- Re-tests on your cases before any new version, prompt or model change goes live.
- Change control: a new version needs your approval, as the first one did.
- Incident handling: we pause the agent, tell you, find the cause and fix it, with the record kept.
- A monthly report: volume, hand-offs to people, test results, cost, changes and incidents.
- Logs kept for the period your contract sets, at least six months for high-risk use (Art. 26(6)).
How long it takes
- Handover and limits set at go-live
- Re-tests before each change every change
- Monthly report every month
Runs monthly, from go-live, with the notice period your contract sets.
What this service doesn’t do
- The day-to-day oversight by your people, which the AI Act gives the deployer (Art. 26).
- New agents or new processes. Each one starts as its own pilot.
- Model usage, billed at the provider’s price without mark-up and capped by your limits.
- Round-the-clock staffed support, unless your service agreement sets it.
How the rules apply here
Each control points to the article it answers. We read them on 9 November 2026; your counsel confirms how they apply to you.
Monitoring
AI Act Art. 26(5)Art. 72Deployers monitor operation and report serious risks; providers run post-market monitoring. We give both sides the data.
Logs
Art. 12Art. 26(6)Each run is recorded and kept for the agreed period.
Stop and override
Art. 14(4)A person can pause the agent at any time. Paused means paused: no new runs start.
Cost limits
PlatformDaily and per-run spending caps. Lowering a cap refuses new work; work already admitted is not cancelled.
Questions about this service
What does a monthly report contain?
Volume, hand-offs to people, test results on your cases, cost against your limit, every change made and any incident. It is written for the business owner, not for engineers.
How fast do you react to an incident?
The response times are in the service agreement, set per agent. Any person with access can pause an agent at once; that never waits for us.
Can we leave?
Yes. The contract sets the notice period and what we export for you, by hand and on request: your knowledge sources, your test cases and results, and the run records the law requires you to keep.
How do we estimate the running cost?
The free agent cost calculator on this site gives a first estimate from your volumes and your provider’s list prices. The proposal replaces it with figures for your agent.
Bring one process. Leave with a risk class, a pass bar and a plan.
A 45-minute scoping call with the people who would build it. You get a first view of the AI Act risk class, the controls it needs, and whether an agent is the right tool at all.