Cognition's Devin Turns Customer Bug Reports Into Pull Requests Automatically
Cognition's Devin now plugs into Pylon so customer bug reports flow straight into AI coding sessions, with $15,000 in credits for new teams.
- Cognition launched a Devin integration for Pylon that turns support tickets into Devin coding sessions.
- New Devin customers who set up the integration receive $15,000 in usage credits.
- Workflow: Pylon AI tags a bug, Devin reproduces it, opens a PR, and posts updates back to the ticket.
- Default security profile blocks Git writes and non-Pylon network traffic to limit prompt-injection risk from customer tickets.
- Free codebase search in Pylon is included with no Devin seat required; full sessions use your Devin plan.
- Pylon counts Linear, Modal, and Cognition among 1,000+ customers already running support on its platform.
Devin turns Pylon tickets into coding sessions
Cognition has released a native Pylon integration for Devin. Customer-reported bugs can move from a Pylon support issue into a Devin coding session, with investigation updates and pull request events flowing back to the original ticket. Engineers retain responsibility for reviewing and merging proposed changes.
Cognition calls the target metric Median Time to Fix, or MTTF. First-response time records how quickly support acknowledges a report, while MTTF tracks progress toward a resolution. Infrastructure teams often use the same acronym for Mean Time to Failure, so dashboards should spell out Cognition’s support-specific definition.
A tagged ticket can start the fix
Teams can launch Devin manually from Pylon, ask questions about a connected codebase, or configure automations around issue creation, tags, and status changes. The automated workflow follows six stages:
- Intake: A customer reports a bug through Pylon.
- Triage: Pylon’s AI evaluates the issue and can add an escalation tag when the report contains enough reproduction detail.
- Launch: A configured event, such as a new issue, tag, or status change, starts a Devin session.
- Investigation: Devin reads the ticket, examines connected observability systems for runtime evidence, and attempts to reproduce the failure.
- Proposal: Devin can prepare and open a pull request when its security profile permits Git write operations.
- Synchronization: Session and pull request events appear in Pylon as internal notes, while subsequent customer replies flow into the linked Devin session.
Two prebuilt automation templates, Triage Customer Bug Tickets and Investigate Support Escalations, provide starting prompts for common workflows. Teams can adjust the triggers and instructions to match their escalation process instead of building the orchestration from webhooks and the Devin REST API.
Untrusted text gets narrow permissions
Support tickets are untrusted input because customers can place arbitrary text inside them. Prompt injection occurs when text supplied as data contains instructions that an agent interprets as commands, potentially influencing its investigation or proposed code changes. Cognition addresses that risk with a restrictive security profile enabled by default.
| Surface | Default control |
|---|---|
| Pylon data | Devin can read issues, accounts, and contacts. Administrators may separately grant write access for issues and accounts. |
| Network | Sessions can make network requests only to usepylon.com. |
| Git | Repository access remains read-only. |
| GitHub CLI | The session receives no GitHub CLI token. |
| Enforcement | The profile is mandatory, preventing a sub-automation from silently overriding it. |
An administrator must explicitly relax the documented controls before Devin can open pull requests. Any resulting change still enters the team’s normal review process, and engineers decide whether to merge it.
Support joins Devin’s toolchain
Cognition already distributes Devin across Slack, GitHub, Linear, Jira, and the Devin CLI. Pylon adds the support channel where customer reports often originate before reaching an engineering tracker. Pylon, which describes itself as an agentic support platform, says more than 1,000 companies use its product, including Linear, Cognition, Modal Labs, and Incident.io.
Pylon says its own engineers use the integration daily. When an engineer merges a Devin pull request, the linked support issue receives an internal note, giving support staff the latest status without a separate Slack message or manual ticket update.
Measure the full repair loop
The synchronized events let teams report how many escalated issues produced a Devin pull request and whether each pull request was opened, merged, or closed. Those records provide the inputs for a median-time dashboard and expose where tickets stall between escalation, investigation, review, and merge.
Metric definitions still require care because the announced workflow tracks pull request states rather than deployment completion. A dashboard labeled “time to fix” should specify whether the clock starts at ticket creation or escalation and whether it stops when a pull request opens or merges.
Adoption hinges on four choices
- Trigger policy: Teams must choose which issue events can start a coding session and what reproduction details an escalation requires.
- Access policy: Administrators must decide whether Devin needs Pylon write access, Git write access, or additional credentials.
- Review policy: Engineering owners should define who reviews agent-generated pull requests and how failed investigations return to support.
- Measurement policy: Support and engineering teams need shared start and stop events for Median Time to Fix.
Existing Devin and Pylon customers can replace custom webhook and API glue with the native integration. New Devin customers who configure it are eligible for $15,000 in credits, subject to Cognition’s program terms. The release extends Devin’s working context from developer tools into customer support while preserving explicit permission and review boundaries around code changes.
Comments
