Design how engineering teams review and approve work delegated to coding agents in Linear.

Introduction

Linear has moved from "issue tracker" to what its CEO calls context infrastructure for agents. Agents are installed in roughly 95% of paid workspaces, and the share of issues created by agents has climbed from about 3% to around half in a year. Delegation is now easy: assign an issue to Cursor, Codex, Devin, Claude, GitHub Copilot or the Linear Agent and a pull request comes back. Review hasn't kept up. A human still has to decide whether the agent's work is correct, safe and in scope, and that decision is now the bottleneck. I'll design the review-and-approval experience for agent-delegated work inside Linear: clarify the problem, explain why it matters now, choose a user segment, rank their pain points, then propose and measure a solution.

Step 1

Clarify Questions (3 minutes)

• Are we designing for agents from the agent platform (Cursor, Codex, Devin), for Linear's own Coding Sessions, or both?

• Why it matters: Third-party agents open PRs in GitHub, while Coding Sessions run inside Linear, so we see different amounts of context.
• Hypothetical answer: Both. The review experience should not depend on which agent did the work.
• Impact: I'll design an agent-agnostic "review contract" that every agent reports into.

• Is "approval" about code review only, or also about whether the work solved the issue?

• Why it matters: GitHub already handles line-level code review. Linear's advantage is the issue, the project and the customer context.
• Hypothetical answer: Both, but Linear should own "did this solve the right problem," with Diffs covering code where teams want it.
• Impact: The solution centers on intent verification, not a GitHub clone.

• Who is the customer: small startups, or larger orgs with compliance needs?

• Why it matters: Enterprise customers need audit trails and required approvers. Startups want speed.
• Hypothetical answer: Start with Business-plan engineering teams of 20–200 engineers, with Enterprise controls layered on.
• Impact: Defaults stay fast and opinionated, and policies are optional.

Propose the Goal

The goal is to let engineering teams approve agent work with confidence in minutes, not hours, without lowering quality. The north-star outcome is more agent-delegated issues reaching "Done" with a merged PR that does not get reverted or reopened.

Define the Scope

In scope: the stage from "agent says it's finished" to "human approves and the issue closes." That covers what the agent reports back, how reviewers are routed, how they inspect and approve or send back work, and team-level policies. Out of scope: how agents write code and general CI tooling.