Cognition’s Devin Auto-Triage is a new feature in Devin Automations that the company says can monitor incoming alerts, investigate them automatically, and return with a summary, next steps, or even a PR.
The feature is aimed at the systems engineering teams already use to flag problems: Slack channels, Linear issues, GitHub checks, Sentry alerts, Datadog dashboards, customer escalations, and custom webhooks. According to Cognition, those tools often stop at detection, leaving a human to reconstruct the surrounding context before work can begin.
With Auto-Triage, Devin is positioned to respond to new bug reports, incidents, and requests as they arrive. The company states that Devin can inspect a codebase, check observability tools, review related tickets or threads, ask for missing context, and spin up sub-Devins to investigate in parallel. If it identifies a likely cause, it can post a summary. If a human needs to intervene, Devin can tag the relevant owner. If a fix appears straightforward, Devin can open a PR.
Customers are already using it
Cognition includes a customer quote from Modal’s Hari Subbaraj, who mentions that Modal has been using Devin Automations to triage incidents for its inference team. He affirms that the system monitors the team’s channel without prompting and can "investigate quickly and come back with fixes or next steps."
The company also points to long-running memory as part of the feature. In Auto-Triage, Devin keeps context from prior investigations, recurring issues, and routing preferences, which Cnognition suggests can help it tag the right owner or connect a fresh alert to an earlier thread. That approach appears intended to reduce duplicate incidents and speed up triage when the same issue surfaces again.
Where it runs
Cognition notes that Auto-Triage can respond to Slack messages, Linear events, GitHub activity, schedules, and incoming webhooks. It also works with connected tools such as observability systems, issue trackers, and the codebase itself.
The company describes several use cases: investigating production alerts, routing new bug reports, looking into failed CI runs, summarizing recurring health issues, and opening fixes.
Security and untrusted inputs
Auto-Triage is designed for messy inputs such as Slack messages, alert payloads, tickets, logs, and webhooks. Cognition warns that those inputs can contain arbitrary text and should not be treated as trusted instructions.
To handle that, Devin runs in secure, network-sandboxed environments, and Cognition says the feature includes additional protections against prompt injection and data exfiltration.
Source: Cognition
