Guide

AI coding agents that turn tickets into pull requests

Copilot, Cursor, Codex, Devin, Factory and Claude Code compared on the one thing this site cares about: what happens when you hand them a ticket.

An AI coding agent is a model that works on a codebase on its own: it reads the task, edits files, runs commands and tests, and hands back a change for a person to review. The ones worth comparing for a ticket-driven team are the cloud agents you can assign an issue to, the same way you assign a teammate, and get a branch or a pull request back.

This guide compares them on that loop only: where you can start them from (Linear, Jira, GitHub issues), what they read from the ticket, what comes back, and what has to be true of your setup first. It does not rank model quality, and it lists no prices: plans and limits change often, so each entry links to the vendor's own documentation, which is where to check them.

At a glance

AgentFrom LinearFrom JiraWhat comes back
GitHub Copilot cloud agentYes: assign the issue to Copilot (GitHub Copilot for Linear app)Yes: assign the work item to Copilot (GitHub Copilot for Jira app; Jira Cloud with Rovo activated)A draft pull request in your GitHub repository
Cursor cloud agentsYes: delegate the issue to Cursor or mention @CursorYes: assign the work item to Cursor or mention @Cursor (Teams and Enterprise plans)A pull request, linked back to the ticket
OpenAI CodexYes: assign the issue to Codex or mention @CodexNo assign-from-Jira; Codex can read Jira through the Atlassian MCP serverA summary and a link to the finished work; you create the pull request from it
DevinYes: assign to Devin, add a skill label, or @mention DevinYes: through the Devin for Jira app (a Rovo agent) or a Jira service userA pull request
Factory DroidYes: assign the issue to Factory or mention @FactoryYes: assign the issue to Factory or mention @FactoryInvestigation, implementation and tests, reported back in the ticket
Claude CodeNo native assign; reads and updates issues through Linear's MCP serverNo native assign; reads and updates work items through Atlassian's MCP serverCommits and pull requests pushed from a GitHub Actions run

Checked against each vendor's documentation on November 14, 2026.

How to choose

  • Start from your tracker. On Jira, Copilot, Cursor, Devin and Factory can all be assigned a work item. On Linear, all of those plus Codex can. Claude Code only reads tickets, in a session someone starts.
  • Then your repository host. Copilot only opens pull requests on GitHub, and Codex works through a GitHub connection. Cursor's cloud agents also work with GitLab, Azure DevOps and Bitbucket.
  • Then where the agent runs. Most run in the vendor's cloud. Claude Code's GitHub Action runs on your own CI runners, which matters if your security team decides where code executes.
  • Then decide who reviews, and when. Codex stops at finished work and a person opens the PR; Copilot opens a draft PR; the others open or link a PR from the ticket. Agree who reviews agent PRs before the first one lands.
  • Check the plan each one needs in its own docs, linked under each entry above. Several integrations are limited to particular plans or need Rovo turned on in Jira.

The agents, one by one

GitHub Copilot cloud agent · GitHub

Best for: Teams whose code lives on GitHub and who want draft PRs from any of the three trackers.

Copilot's cloud agent (GitHub's earlier name for it was the coding agent) starts from a GitHub issue assigned to Copilot, and from Linear and Jira through GitHub's own apps for each. GitHub opened the Linear app in public preview in October 2025 and the Jira app in March 2026.

Assigned a ticket, it reads the issue and its comments, works in its own ephemeral development environment powered by GitHub Actions, and opens a draft pull request. From Jira it uses the work item's title, description, labels, comments and Atlassian custom fields such as acceptance criteria, so a filled-in acceptance criteria field reaches the agent as written.

Check before you roll it out

  • It needs a paid Copilot plan, and it only opens pull requests on repositories hosted on GitHub.
  • Installing the Jira app takes a Jira site admin and an owner (or GitHub App manager) of the GitHub organization; after that, only Jira users with write access to the repository can trigger it.
  • The Jira integration requires Jira Cloud with AI features turned on and Rovo activated for the organization.

Docs: Copilot cloud agent with Jira · Copilot cloud agent with Linear · Starting Copilot sessions on GitHub

Cursor cloud agents · Cursor

Best for: Teams already on Cursor, and teams whose code is on GitLab, Azure DevOps or Bitbucket.

Cursor's cloud agents run from Linear (since August 2025) and Jira (since May 2026). Delegate the issue to Cursor or mention @Cursor in a comment, and an agent scopes the task from the ticket's title, description and comments plus your team's repository settings. You can follow and steer it from the tracker, the Cursor web app or the editor.

In Linear, triage rules can assign certain issues to Cursor automatically, which pairs with the routing approach in Linear workflow automation for agent ticket assignment.

Check before you roll it out

  • The Jira integration needs Jira Commercial Cloud with Rovo enabled, and is available only on Cursor's Teams and Enterprise plans. Cursor's docs say it is not supported on Atlassian HIPAA or FedRAMP instances.
  • Cloud agents need usage-based pricing turned on, a connected repository provider (GitHub, GitLab, Azure DevOps or Bitbucket) and a default repository, model and base branch.
  • Choose between running agents under a service account or under each user's own account; team MCP servers aren't available in service-account mode by default.

Docs: Cursor in Linear · Cursor in Jira

OpenAI Codex · OpenAI

Best for: Linear teams on ChatGPT plans who want to look at the work before a pull request exists.

In Linear, assign an issue to Codex or mention @Codex in a comment, and it starts a cloud chat in an environment you set up for the repository, posting progress to the issue's activity. Triage rules can assign issues to it automatically. To pin a repository, name it in the comment.

When it finishes it posts a summary and a link to the completed chat, and a person creates the pull request from there. That extra step is a review gate some teams want and others will find slow.

There is no Jira app that assigns work to Codex. Atlassian's MCP server lets Codex read a Jira issue and search with JQL inside a session someone starts, and Jira can open a work item directly in a local coding tool.

Check before you roll it out

  • Codex in Linear is on paid ChatGPT plans; on Enterprise, a workspace admin has to turn on Codex cloud chats and enable Codex for Linear in connector settings.
  • It needs GitHub connected in Codex and an environment for each repository. If a request is ambiguous it falls back to the environment you used most recently.

Docs: Use Codex in Linear · Codex in Linear's integration directory · Atlassian and OpenAI integrations

Devin · Cognition

Best for: Teams that want every ticket of a kind handled the same way, through reusable skills.

Devin takes Linear tickets three ways: assigned directly (it runs your default skill), tagged with a skill label such as @skills:implement (it runs that skill), or mentioned in a comment (your comment becomes the instruction). Automation triggers can start it without anyone assigning the ticket.

Skill labels are the useful part for a backlog: the same kind of ticket gets the same procedure every time, which is what makes clearing a batch of tech-debt tickets repeatable. For Jira, Devin offers two setups: the Devin for Jira app, which works as a Rovo agent, or a Jira service user connected over OAuth.

Check before you roll it out

  • Pick one of the two Jira setups; they install differently, so read both before choosing.
  • Decide which Linear teams Devin can see when you connect it; you can change that later in Linear's app settings.

Docs: Devin Linear integration · Devin Jira integration

Factory Droid · Factory

Best for: Organisations that want one admin-controlled agent setup across trackers and chat.

Factory calls this Delegations: work goes to its agent, Droid, from Linear, Jira, Slack and Microsoft Teams. Assign an issue to Factory or mention @Factory and Droid investigates, implements and tests the change on a remote computer, posting progress and results in the issue.

An administrator sets one shared agent profile (instructions, plugins, a service-account identity and compute) that every delegation uses. That is where to put your repository conventions; see context files and repo conventions coding agents need.

Check before you roll it out

  • Shared delegations run as a service account with the profile your admin configured, so whoever owns that profile owns how every delegated ticket is handled.

Docs: Factory Delegations · Delegate from Linear · Delegate from Jira

Claude Code · Anthropic

Best for: GitHub teams that want the agent to run inside their own CI.

Claude Code's ticket-to-PR path is GitHub: the Claude Code GitHub Action runs in your repository's own workflows, and mentioning @claude in an issue or pull request comment has it analyse the code, implement the change and push commits. Because it runs on your runners, your CI configuration decides what it can reach.

From Linear or Jira it works the other way round: a developer starts a session and Claude Code reads the ticket through the tracker's MCP server. Assigning a Linear issue to Claude Code is an open feature request on its GitHub repository, not a feature.

Check before you roll it out

  • It runs on your own GitHub Actions runners with your own Claude credentials, and your workflow file decides which events and comments trigger it.
  • Nobody can hand it a ticket from the tracker on its own; plan for a person, or your own automation, to start each run.

Docs: Claude Code GitHub Actions · Linear MCP server

Left off this list

  • Jules (Google) started tasks from GitHub issues labelled 'jules'. Google is winding it down: it turned read-only after November 11, 2026 and closes on December 31, 2026, so don't build a workflow on it.
  • Rovo Dev (Atlassian) starts coding sessions from Jira work items and can check a pull request against the ticket's acceptance criteria. Atlassian says it is ending as a standalone product and its features are moving into eligible Jira subscriptions; check Atlassian's Rovo Dev page for what your Jira plan includes now.

Before you assign the first ticket

Every agent above reads the same thing a new hire would: the ticket. Which one you pick matters less than whether the ticket says what done looks like.

Questions people ask

What is an AI coding agent?

A model that works on a codebase on its own: it reads a task, edits files, runs commands and tests, and returns a change for a person to review. The cloud agents in this guide do that from a ticket assigned to them and hand back a branch or a pull request.

Which AI coding agents can I assign a Jira ticket to?

As of November 2026: GitHub Copilot's cloud agent, Cursor's cloud agents, Devin and Factory's Droid all take a Jira work item assigned to them. Copilot's and Cursor's Jira apps need Jira Cloud with Rovo enabled, and Cursor's is limited to its Teams and Enterprise plans.

Which AI coding agents work from Linear?

Codex, Cursor, GitHub Copilot, Devin and Factory can all be assigned a Linear issue. Claude Code can read and update Linear issues through Linear's MCP server, but it can't be assigned an issue.

How much do they cost?

This guide doesn't list prices because they change often and depend on the plan. Each entry links to the vendor's documentation, which says which plans include the integration; check current prices on each vendor's own pricing page.

Sources

  1. GitHub Copilot coding agent for Jira is now in public preview (GitHub changelog, March 5, 2026)
  2. GitHub Copilot for Linear available in public preview (GitHub changelog, October 28, 2025)
  3. Bringing the Cursor Agent to Linear (Cursor, August 21, 2025)
  4. Cursor in Jira (Cursor changelog, May 19, 2026)
  5. OpenAI Codex agent (Linear changelog, December 4, 2025)
  6. Linear's agent directory
  7. Linear integration: assign issues to Claude Code (feature request)
  8. Jules changelog: Jules turndown
  9. Rovo Dev (Atlassian)

← Front page