A Slack-Triggered Autonomous Incident-Response Agent
Problem
Incident and PR reports landed in Slack faster than anyone could triage them, and the first responder always repeated the same manual steps: figure out which feature area the incident touched, find the relevant code, and sketch a remediation approach. That repeated analysis was pure latency between 'something is wrong' and 'here is a plan.'
Trial
I built a service that watches the Slack incident and PR channels and, for each new report, auto-builds an 'analysis map' from a full feature-audit report of the codebase, so the incident is immediately located against known feature areas. Each incident is then dispatched to a read-only agent planning queue that drafts a remediation plan automatically, and the whole thing runs as a live local server. Later I merged it into a daily engineering dashboard that auto-surfaces other issues of the same severity, forming an event-driven chain from Slack to analysis to LLM planning.
Result
By the time a human opened an incident thread, a located analysis and a draft remediation plan were usually already waiting, cutting the cold-start phase of response to near zero. The same-severity surfacing on the daily dashboard turned isolated incidents into visible clusters, which caught related issues that would otherwise have been triaged independently. Keeping the planning agents read-only meant the automation could run unattended without any risk of it 'fixing' production code on its own.
Lessons
The pre-built analysis map was the key piece: an agent that has to rediscover the codebase per incident is too slow and too inconsistent, while one that starts from an existing feature audit plans in seconds. Automating the plan but not the fix is the right trust boundary for incident response — humans stay in control of every change while the machine absorbs the repetitive analysis.
Next Action
Next I want the agent to attach confidence levels and link past similar incidents to each plan, so responders can tell at a glance whether to trust the draft or start fresh.