Start with a person, not a pipeline.

Two board questions concern AI agentsAgentA program that takes actions for an organisation. NovusLens records what it is allowed to do and what its activity reports show. in the customer’s agent estateAgent estateThe collection of AI agents an organisation operates, together with the systems and providers they use.. Each answer comes from records. You can start collecting them before connecting the systems that run the agents. NovusLens maps and surfaces the evidence. It does not run, stop or change the agents.

The financial question: is any agent acting beyond its mandate?

Collect:

1. The agent inventory: every automation, what it serves and who answers for it. See the agent inventory.
2. The mandate for each agent, from the body that granted it: issuer, operation, resource, unit, upper limit and validity.
3. The system that records each approval with its value and unit. This is the operation sourceSourceA document, interview statement, measurement or export that supports a value. Each source has a date and a stated scope.. It is never the agent.
4. Who may authorise an intervention, and which runtimeRuntimeThe software that runs a model or agent. A model runtime answers model requests; an agent runtime carries out the customer's agent actions. executes one.
5. Who observes the outcome independently, and the criterion they will measure over a stated window.

Then ingestIngestRead supplied records or documents and add their information to NovusLens, after the required checks. the inventory, enrol one source for each record type and activate them. Record the outcome criterion for each intervention case on the evidence policyEvidence policyThe operator’s rules for which agent reports count and for how long. The default period for grouping related events is fifteen minutes. with the agent-source criterion command. It adds to the existing policy and records a new version. Send batches, check each first, and confirm the receiptReceiptA recorded acknowledgement of a request or action, with its identity, time and result. An evidence intake receipt confirms storage. It does not establish acceptance or prove that an agent acted.. See agent evidence records.

The evidence policy has two settings. fleet_window_ms is between one minute and one day. evidence_horizon_ms is between one hour and 366 days, and defaults to 30 days.

It depends on these:

• A grant and its operations must share a unit, or they cannot be compared.
• A fleet incidentFleet incidentA finding about several agents that completed the same operation above their recorded limits in one period. The agents must share something recorded in the model. needs at least two agents over their limits within one window, sharing something on the mapMapThe view of teams, services and their dependencies for the selected part of the organisation. Details appear beside the map.. One agent three times is one case.
• An outcome is only as complete as its window and coverageCoverageWhat the available records cover, and what is missing. A count such as “3 of 5 dependencies measured” names the measured portion and the gap. say. With no observer there is no outcome and no all-clear.

The functional question: is any agent doing work it was never given?

Collect the responsible person for each agent and the written scopeScopeThe work, data, activity or access that a rule, statement or calculation covers. For access, it names the records an account can see. For team measures, “Own operations” covers the selected unit. “Including descendants” adds the units beneath it. they stand behind: purpose, allowed operations, resources and destinations, exclusions, validity, and whether the list is complete. Collect the business-activity catalogue and the source that sees what agents do.

Declare and approve each remitRemitA written list of an agent’s allowed actions, resources and destinations for information. The list states whether it is complete. in the product. This needs no runtime. Then enrol the activity source and send activity records. State which activity the source covers and what it misses. Read the remit panel for the result: inside, outside, review required or cannot be stated. See status messages and states and AI agents.

Push for a complete list, because an incomplete one proves nothing about what it omits. Never mark a list complete only to get a result.