Track blocked tasks, corrections, fallback actions, handoff outcomes, and reuse rules so personal AI agents learn how work recovered instead of repeating the same failure.
Preference memory remembers how a user likes work done. Exception memory remembers where the work broke, why it broke, and how the system should route similar tasks in the future.
Exception memory software stores structured evidence: trigger, attempted action, fallback, outcome, and reuse rule. That makes it useful for text assistants, browser automation, publishing agents, and human escalation loops.
Remember unclear asks, skipped approvals, and user corrections inside text message AI assistant workflows.
Pair with computer-use cache to avoid repeated tool and context failures.
Store broken assets, stale deploy checks, and release gates in agent website workflows.
Classify what made the normal path fail: missing approval, stale context, tool error, ambiguous user intent, policy boundary, or failed handoff.
Write down what the agent tried to do in operational language, with enough context for a future agent to avoid guessing.
Track whether the system drafted, retried, downgraded scope, escalated to a human, stopped, or received a user correction.
Convert the exception into future routing: ask earlier, require approval, avoid a tool, route to a person, or package context differently.
Do not bury exceptions in transcript search. Attach them to the workflows that need different behavior next time.
Keep correction and approval exceptions near the conversation where the user will act.
Remember failed selectors, auth gates, brittle sites, and risky action boundaries.
Use exceptions to improve asset checks, source checks, link checks, and release gates.
Remember which context made a human resolution fast enough to reuse.
Is exception memory just an error log?
No. Error logs explain what failed. Exception memory also stores what recovered and what should happen next time.
Should exception memory be private?
Yes. It can include sensitive workflow context, so user control and deletion matter.
Where does Supers fit?
Supers is relevant for personal AI agents that span text, browser automation, publishing, and human handoff.
Exception memory software turns failed or interrupted agent work into reusable routing evidence. It is the operational layer that helps personal AI agents recover better the next time.
Start with the exceptions that create repeated human effort, then wire those records into text, browser, publishing, and handoff workflows.