Linear has placed coding-agent work inside the issue tracker where product and engineering teams already define, discuss, and review work. A person can delegate an issue or ask for a change in Linear, Slack, or Teams. The agent works in a managed sandbox, returns a diff and pull request, and can include verification artifacts for review.

The issue carries product context into implementation

Linear's June 2026 launch account says a coding session can begin by assigning an issue to Linear Agent or asking for a change in a chat, comment, or Slack thread. The session can use issue history, customer requests, discussions, and related work already stored in the workspace.

The issue becomes more than a ticket number. It carries the reason for the change, prior decisions, affected customers, and the conversation that shaped the request. This reduces manual copying between product planning and an agent prompt.

Linear distinguishes investigation from implementation. Its coding-session documentation says requests to investigate, debug, research, or plan do not automatically start coding. A direct request to fix, implement, or open a pull request does. That boundary lets a team explore a problem before authorizing a change.

A managed sandbox supplies the development environment

Coding sessions run through Claude Code or Codex inside a managed development sandbox. Workspace administrators configure supported runtimes, tools, setup commands, text files, environment variables, repository instructions, and the selected agent and model.

A sandbox is an isolated computer for one work session. It keeps the agent's processes and files away from an employee laptop and makes the setup repeatable. Linear warns that configured environment values are visible to the agent and should not be treated as hidden secrets.

The agent can install locked dependencies, start local applications, use browser automation, and capture screenshots or recordings. When it prepares a pull request, Linear shows the diff and evidence in its review interface. Reviewers can ask the agent to address comments, rebase the branch, or repair lint failures before merging.

This creates a useful evidence ladder. The diff shows what changed. Tests and checks show whether defined technical conditions hold. A screenshot or recording shows how a user-visible flow behaved. A person still decides whether that evidence answers the product requirement.

Triage can trigger the first repair attempt

Linear lets teams start a coding session automatically when an issue reaches Triage and matches a condition such as a label. Internally, Linear connects Triage Intelligence to this workflow: when the system identifies a new issue as a bug, an automation investigates it and can draft a pull request.

At launch, Linear reported using this workflow to resolve roughly 30 percent of incoming internal bug reports, mostly on the first pass. This is meaningful internal-use evidence because it describes a product outcome rather than raw generation. The company does not publish the number or severity of bugs, the evaluation window, false classifications, review effort, escaped defects, or an independent audit.

The automation can gather evidence from tools such as Sentry or Datadog through MCP. MCP, or Model Context Protocol, is a standard connection through which an agent can call an approved tool. The issue remains the record that connects the report, investigation, proposed change, review, and final decision.

Linear's design refresh shows a human-agent team

Linear's March 2026 design account describes a two-person team using Linear Agent, Cursor, Codex, and Claude Code during a broad interface refresh. The agents helped locate components and prior owners, build internal tools, and prototype alternative directions.

Claude Code built an internal color tool in a few hours. Designers used it to tune interface tokens, then exported the chosen values as JSON into Figma. Feature flags and a development toolbar let the team compare the old and new interface while shipping the refresh in smaller changes.

This example shows agents supporting product exploration and implementation without choosing the design direction. People evaluated the alternatives and invested further work in the chosen result. It also mixes several agent products, so the outcome cannot be attributed to Linear Agent alone.

Product availability and internal evidence are separate

Coding sessions are a customer product on Linear's paid workspace plans and require a GitHub connection plus AI credits. The documentation explains the public workflow and current configuration. Linear's 30 percent bug-resolution figure and design-refresh story describe internal use.

The feature was new in the September 2026 source window. Linear has not published a longer trend for adoption, cost per accepted result, review time, reverts, incidents, or customer outcomes. Model choices and product behavior can also change after this dated account.

What a product team can copy

Keep product context attached to the delegated issue. Separate research verbs from implementation verbs. Configure one repeatable environment per repository, and treat every value inside it as visible to the agent. Require a diff plus evidence from tests or the running product. Automate first passes only for well-classified work, then measure acceptance, review time, defects, and cost rather than counting started sessions.