Andrew Norrish, director of field engineering at Sourcegraph, describes the incident-response consequences he sees as teams adopt coding agents.
The pressure on responders
- Norrish argues that faster code generation can shift work toward human review and leave site reliability engineers (SREs) investigating unfamiliar changes.
- His account draws on Sourcegraph customer experience; the talk does not establish a general causal rate for AI-related incidents.
- Large, distributed codebases make it harder for either a reviewer or an agent to see cross-service effects.
Three workflows
- Investigate: search code and history across repositories, and connect logs to the revision actually deployed.
- Prevent: monitor for known risky or noncompliant patterns before they spread.
- Remediate: locate all affected repositories and coordinate changes across them.
- Sourcegraph uses a HackerOne-triggered triage workflow internally; Norrish also describes customers’ incident-response and code-review agents.
These are vendor-presented examples, not independent evaluations. Automated reverts and bulk changes still need permissions, safeguards and verification.
“A fix found in one repo often applies to hundreds if not thousands of others.” — Andrew Norrish