Retry logic repeats an operation. Recovery prompts re-enter a task with context, constraints, and a safer route after a personal AI agent hits an exception.
Retry logic is useful for transient infrastructure problems. Recovery prompts are necessary when the next attempt must incorporate user corrections, policy boundaries, stale context, or a changed route.
That assumption is correct for temporary network errors, rate limits, or flaky tools. It is dangerous when the problem is a user correction, an expired approval, a policy constraint, or a failed public action.
The instruction is unchanged and the next attempt should be mechanically similar.
The instruction must include what happened, what changed, and what to avoid.
A recovery prompt can decide whether a retry is still safe.
If a browser, API, or worker failed without changing task meaning, retry logic can work. Keep a cap, backoff, and an escalation path.
If the user changed the instruction, rejected the output, or missed an approval window, use a recovery prompt so the agent does not repeat the old route.
When agents move through text, browser tasks, publishing, and handoff, recovery prompts preserve context across surfaces. Supers workflows such as text message AI assistant, computer-use cache, and agent website builds benefit from that split.
Retry logic is a reliability primitive. Recovery prompts are an agent behavior primitive.
Retry delivery failures; recover from ambiguous replies and missed approvals.
Retry flaky loads; recover from auth gates, changed pages, and unsafe actions.
Retry deploy checks; recover from broken links, stale sources, and release gates.
Retry notifications; recover by summarizing what the human changed.
Should recovery prompts replace retries?
No. Retries are still useful for transient failures. Recovery prompts are for changed context.
What is the first workflow to instrument?
Start with approval timeouts or failed browser actions because they clearly show whether retrying is safe.
Where does Supers fit?
Supers is relevant for personal AI agents that span text, browser automation, publishing, and human escalation.
Retry logic keeps infrastructure moving. Recovery prompts keep agent behavior aligned after reality changes. Personal AI agents need both, but they should not be confused.
When the task meaning changes, recover with a prompt that carries the exception forward.