A native agent that synthesizes workspace context, scopes customer feedback into issues, creates follow-up work, runs coding sessions, and can be triggered automatically when issues enter triage; used by Linear's own CX, Product, and Engineering teams.
Customer email to shipped feature: Intercom conversation scoped into an issue, routed by Triage Intelligence, first-pass change by a delegated coding agent, reviewed pull request, and customer follow-up.
Scope customer feedback into an issue
From the Intercom inbox, the agent picks up the full conversation, metadata, and attachments, and creates a scoped issue[Source 2]
A delegated coding agent drafts the change or opens the pull request while the issue stays assigned to the human engineer[Source 2]
Close the loop with the customer
When the issue is done, related requests reopen in Intercom or the originating Slack thread is notified, and a person follows up with the customer[Source 2]
Where people stay involved
People review, approve, or guide the agent’s work here.
Reported: A human engineer makes the final pull-request approval, risky actions pause for confirmation, and the delegated issue stays assigned to its human owner.
Supervision level
Work-product review
A person reviews the draft or implementation.
3/5Level 3 of 5
assigned coding work → agent-created change[Source 1]
Codex was used for internal pull-request review; other model choices are not detailed in the preserved sources[Source 2]Codex is named for internal pull-request review; the captures say only frontier language models elsewhere.
Harness
Separate planning agent (triage/issue creation) and coding agent (code generation); Code Intelligence for codebase knowledge; scheduled/event-driven 'Loops'[Source 1, Source 2]
Sandbox
Not reportedThe captures describe a custom owned stack but no execution or isolation boundary.
Tool access
Triage Intelligence (auto-route, dedup, label); GitHub; testing Code Intelligence + custom MCP servers[Source 2]Code Intelligence and custom MCP servers are named as internal tests with a planned launch.
Knowledge
Semantic/vector search evolved into agentic context acquisition across the workspace; Datadog/Sentry customer context[Source 2]The Datadog and Sentry customer context is verified; the semantic-search history clause could not be located in the captures and is flagged in the source review.
Context management
System skills are progressively disclosed per run; relevant skills load before the task and the agent loads further skills on demand[Source 1]
Credentials
The Agent SDK gives agents explicit identities, scoped OAuth tokens, assignable/mentionable handles, and visible human delegation; issues stay assigned to a human; 'an agent cannot be held accountable'[Source 3]The Agent SDK model gives agents scoped OAuth identities; accountability stays with the human assignee.
Interfaces
GitHub is a system the agent acts on through pull requests; the conversational surfaces are Slack, Intercom, and Linear.
Agents get identities, scoped team access, and visible delegation alongside humans[Source 3]
Loops
Scheduled or event-driven agent runs that execute recurring work[Source 1]
Validation and failure handling
Review the pull request
A coding agent reviews every pull request and a human engineer makes the final approval[Source 2]
Pause for confirmation on risky actions
When an action is risky or hard to undo, the agent pauses mid-run and asks the user for confirmation[Source 1]
Reported observations
Not reported: The reviewed captures report no usage or outcome measurements for Linear Agent; statements about effectiveness are qualitative.
Lessons
Linear's customer-support workflow carries Intercom conversations into issues and delegates coding work from those issues.[Source 2]
Reported opinion: Linear recommends starting with agent suggestions, observing the results, and adding guidance before enabling automatic operation.[Source 2]
Reported opinion: One Linear engineer reports better results when he breaks coding work into small, targeted steps.[Source 2]
Reported opinion: A Linear engineer finds agent mistakes useful for identifying possible failure modes during his review.[Source 2]
Linear completion events notify the originating support workflow, where a person follows up with the customer.[Source 2]
Reported: The capture documents two further workflows (a Slack thread becoming a pull request, and a PM filing and shipping a fix); the email-to-feature run is the representative one because it spans all five stages.
human involvement
Reported: A human engineer makes the final pull-request approval, risky actions pause for confirmation, and the delegated issue stays assigned to its human owner.
implementation
Reported
validation
Reported: The coding-agent review pass with human final approval, and the mid-run confirmation gate for risky actions, are the documented checks.
observations
Not reported: The reviewed captures report no usage or outcome measurements for Linear Agent; statements about effectiveness are qualitative.
