Customer: small businesses and serious individual users deploying agents to email, browse, code, purchase, schedule or invest.
Problem: agents increasingly act across multiple systems, but users have poor visibility into why an action happened, which permission enabled it, what changed and how to reverse it. The more useful agents become, the more painful that missing audit trail becomes.
Product: a lightweight 'flight recorder' that captures each proposed action, the source data that triggered it, the permission used, the external side effect and the available rollback path. Think Snagit history plus Git undo plus a security-camera timeline for autonomous software.
The first version should be deliberately narrow. Pick one workflow such as Gmail. Record every read, classification, label change, draft, send, archive or delete as a human-readable event with timestamp, reason, affected object and before/after state. Do not try to solve observability for every agent on day one.
The interface is the product. A developer log full of JSON is not enough. The user should be able to scan a timeline and answer: what did the agent do, why did it do it, what information did it rely on, and can I undo it? High-risk actions should stand out immediately.
OpenTelemetry is a useful technical foundation because it already provides a vocabulary for traces, spans and events. The business value comes from translating that machinery into an audit experience a normal operator can understand.
Monetization could start as a per-user or per-agent subscription for small teams, with a higher tier for longer retention, policy controls and compliance exports. The buyer is not paying for another agent; they are paying to trust the agents they already have.
First move: instrument one local Gmail automation end-to-end and build the cleanest possible timeline around it. If the timeline makes a confusing automated action immediately understandable, the core idea is working.