--- schema: "swft.publication/v1" id: "company-linear-coding-sessions" title: "Linear coding sessions: from product issue to reviewed pull request" description: "How Linear connects issues, Slack, managed coding sandboxes, browser evidence, pull-request review, and automated first passes on incoming bugs." summary: "Linear coding sessions let a team delegate an issue where its product context already lives. Claude Code or Codex works in a managed sandbox, returns a diff and pull request, and can include verification artifacts. Linear also uses automated sessions on incoming bugs, although its outcome figure is a recent company self-report without a full quality or cost series." canonical: "https://swft.io/companies/linear-coding-sessions" author: "SWFT Editorial" author_type: "Organization" published: "2026-09-01" modified: "2026-09-02" kind: "case-study" section: "Companies" tags: ["Linear coding sessions", "Linear Agent", "managed coding sandbox", "issue to pull request", "agentic bug triage", "AI software factory"] evidence_labels: ["INFERENCE", "OBS", "SELF-REPORT"] source_ids: ["linear-coding-sessions", "linear-coding-sessions-docs", "linear-design-refresh"] authorship_disclosure: "AI-drafted from the cited public sources and independently checked by a second AI editorial-review agent (Codex) for source fit, claim boundaries, overlap, and reader utility. SWFT Editorial is responsible for corrections." --- # Linear coding sessions: from product issue to reviewed pull request Linear carries issue and customer context into a managed coding sandbox, then returns code and can include checks and visual evidence in the same workflow for review. > **Authorship:** AI-drafted from the cited public sources and independently checked by a second AI editorial-review agent (Codex) for source fit, claim boundaries, overlap, and reader utility. SWFT Editorial is responsible for corrections. ## Quick answer Linear coding sessions let a team delegate an issue where its product context already lives. Claude Code or Codex works in a managed sandbox, returns a diff and pull request, and can include verification artifacts. Linear also uses automated sessions on incoming bugs, although its outcome figure is a recent company self-report without a full quality or cost series. 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](https://linear.app/changelog/2026-06-11-coding-sessions) 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](https://linear.app/docs/coding-sessions) 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](https://linear.app/now/behind-the-latest-design-refresh) 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. ## How we know - **First-party report (SELF-REPORT)** Linear describes coding sessions that start from issues or conversations, use workspace context, run in the cloud, return diffs, and can include preview evidence for review. Sources: [Coding sessions in Linear](https://linear.app/changelog/2026-06-11-coding-sessions). - **Observed artifact (OBS)** Linear's public documentation specifies managed environments, administrator controls, investigation-versus-implementation behavior, browser testing, verification artifacts, and pull-request review. Sources: [Coding sessions](https://linear.app/docs/coding-sessions). - **First-party report (SELF-REPORT)** Linear says an internal triage automation resolved roughly 30 percent of incoming bug reports and describes a two-person design team using several coding agents during its interface refresh. Sources: [Coding sessions in Linear](https://linear.app/changelog/2026-06-11-coding-sessions); [A calmer interface for a product in motion](https://linear.app/now/behind-the-latest-design-refresh). - **Analysis (INFERENCE)** SWFT treats context-preserving issue intake, explicit authorization verbs, managed environments, and reviewable behavior evidence as the most reusable parts of Linear's workflow. Sources: [Coding sessions in Linear](https://linear.app/changelog/2026-06-11-coding-sessions); [Coding sessions](https://linear.app/docs/coding-sessions). ## Sources - **First-party report (SELF-REPORT)** [Coding sessions in Linear](https://linear.app/changelog/2026-06-11-coding-sessions) — Linear Changelog; published 2026-06-11; accessed 2026-09-01. Linear's announcement of issues and team chat as dispatch surfaces for managed coding sessions, sandboxes, proof artifacts, review, and merge handoff. - **First-party report (SELF-REPORT)** [Coding sessions](https://linear.app/docs/coding-sessions) — Linear Docs; accessed 2026-09-01. Current product documentation for Coding Sessions, execution providers, evidence artifacts, review, revision, and repository integration. - **First-party report (SELF-REPORT)** [A calmer interface for a product in motion](https://linear.app/now/behind-the-latest-design-refresh) — Linear; published 2026-03-12; accessed 2026-09-01. Linear's account of using multiple coding agents during its design refresh, including code archaeology and an in-product color-token tool. ## Read next - [Tempo's software factory: turning product signals into review-ready PRs](/companies/tempo-software-factory) - [Ramp Inspect: how a background coding agent became factory infrastructure](/companies/ramp-inspect) - [How an AI software factory works](/software-factory-architecture)