Three weeks ago, GitHub's Copilot cloud agent for Linear went generally available. The same week, the Cursor changelog confirmed you can type @Cursor fix the login issue in any Slack channel and get a pull request back without opening an IDE. Both shipped quietly, but together they mark something concrete: the interface for delegating coding work has moved from the editor to the chat window.
How coding agents in Slack actually work
The pattern is the same across Cursor, Copilot, and OpenClaw's code-agent plugin: you send a natural-language message, the agent spins up an isolated environment, works against your repo, and posts back a PR when it finishes. No tab-switching, no terminal.
With Cursor, you mention @Cursor in any channel; agents read the thread, understand context, create PRs in GitHub, and post updates - including links - when the work is done, running remotely in a secure environment.
Cursor reads the entire Slack thread before starting, so the background agent understands the full context when you reference previous discussions or issues.
GitHub Copilot takes the same approach through Linear rather than Slack threads. You assign issues in Linear to Copilot cloud agent - it analyzes the issue contents and opens a draft pull request.
It works independently in an ephemeral GitHub Actions environment, exploring code, making changes, running tests and linters, streaming updates to your Linear activity timeline in real time, and requesting review when the work is ready.
The Linear integration is now in general availability, with teams able to select models, use repository-defined custom agents, and control base and working branches directly from Linear. Workspace and team-level guidance lets engineering leads standardize how delegated issues are handled, and developers can redirect an agent mid-run by mentioning Copilot in a Linear comment.
Beagle in action#eng, 10:31amThe three tools now competing for this workflow
The Slack-and-Linear coding-agent pattern is no longer a single vendor's bet. Three distinct approaches are live:
| Tool | Trigger surface | Execution environment | PR output |
|---|---|---|---|
| Cursor Background Agents | @Cursor in any Slack channel |
Isolated VM per task | Yes, with live status |
| GitHub Copilot cloud agent | Assign issue in Linear or @GitHub in Slack |
Ephemeral GitHub Actions | Draft PR, awaits review |
| OpenClaw + code-agent plugin | Any channel OpenClaw supports (Slack, Telegram, Discord) | Worktree isolation per session | Merge, PR, or discard |
The GitHub app for Slack now works with the Copilot coding agent, letting you generate pull requests directly from Slack conversations - and the new functionality is optional, requiring new permissions only if you want the coding features.
OpenClaw's code-agent plugin runs Claude Code, Codex, and experimental OpenCode as managed background sessions from OpenClaw chat, adding plan approval, session lifecycle, wake routing, worktree isolation, and merge/PR follow-through on top of those agent backends.
The latest OpenClaw release - v2026.7.1 - brought expanded model support including GPT-5.6 compatibility and stronger Codex and connected coding-agent workflows.
The release drew 3,063 contributions from 532 contributors.
What breaks at the handoff
The workflow sounds cleaner than it runs. A few friction points come up consistently across all three tools.
Context length matters more than it looks. Copilot cloud agent captures the entire Slack thread as context for the request - and that context is stored in the pull request. That sounds helpful, but a long thread full of tangents or stale proposals becomes noise the agent has to filter. The PR inherits the mess.
Redirect mid-run is limited. Cursor lets you reply with follow-up instructions inside the thread. Copilot's Linear integration lets developers redirect an agent while it is still working by mentioning Copilot in a Linear comment
- but that requires you to catch it early, before it has committed down a wrong path.
Security posture varies sharply. Cursor launches an isolated VM to work with your repo in a remote dev environment; the agent reads context in the Slack thread and writes code.
Governance is preserved - every PR follows your review and approval rules, regardless of origin. OpenClaw's self-hosted model is different: giving an AI agent shell access to a machine carries real risk, and Bitdefender found exposed OpenClaw instances leaking Anthropic API keys, Telegram bot tokens, and Slack OAuth credentials on the open internet.
The non-obvious cost here is not the per-task inference spend - it is the review load. Each agent-generated PR still needs a human read. On a team running five concurrent background agents, PR review can quietly become the bottleneck the agents were supposed to remove.
Where this leaves the Copilot and Cursor pricing math
Cursor reached $2B in annualized revenue by February 2026 and serves roughly one million daily active users. Plans run from a free Hobby tier through Pro ($20/month), Pro+ ($60/month), Ultra ($200/month), and Teams ($40/seat/month); in June 2025, Cursor moved from request-based to credit-based pricing, so actual cost depends on which models you use.
Background Agents burn credits faster than chat. A task that spawns multiple review/implementation passes - common on bugs that touch several files - can exhaust a Pro plan's monthly budget in a handful of sessions. The Teams plan at $40/seat is the practical floor for engineering teams running agents regularly; at that rate, a 10-person team pays $400/month before any model premium.
Copilot's coding agent is available with all paid Copilot plans , which currently start at $10/seat/month for Individual and $19/seat/month for Business - roughly half Cursor Teams pricing. The trade-off is that Copilot's agent is tighter in scope: it is asynchronous and autonomous, best for offloading tasks like bug fixes, incremental features, test coverage, and refactors. Cursor's agent handles more exploratory work but costs more.
For most teams, the honest decision is: use Copilot's agent for the high-volume, well-scoped tasks that live in Linear, and reach for Cursor when the brief is messier and needs more iteration.
Coding agents in Slack: common questions
What does @Cursor do in Slack?
Mentioning @Cursor in a Slack channel launches an isolated virtual machine that works against your repo, reads context from the Slack thread, writes code, and when done pushes changes and shares a GitHub PR for review. You can specify the repo, choose a model, and reply with follow-up instructions.
Does GitHub Copilot work in Slack?
Yes. The GitHub app for Slack works with GitHub Copilot coding agent. Mention @GitHub in any Slack thread with a prompt, and the coding agent works in the background, then replies when the pull request is ready for review.
It is available on all paid Copilot plans.
How is the Copilot Linear integration different from Slack?
The Linear integration is generally available and lets teams select models, use custom agents, and control base and working branches directly from Linear. Workspace and team-level guidance standardizes how delegated issues are handled. The Slack path is better for reactive bugs surfaced in conversation; Linear is better for planned, scoped work already in the backlog.
Do coding agents in Slack bypass code review?
No. The GitHub Copilot coding agent can be assigned issues, spin up a secure environment via GitHub Actions, make commits, and open draft pull requests - it enforces branch protections and other security policies, and awaits your review before merging.
Is OpenClaw a viable alternative for teams that want more control?
OpenClaw's scale has spawned a managed ecosystem. KiloClaw at $49/month offers hosted deployment so teams can run a production agent without managing infrastructure.
OpenClaw itself carries 369K GitHub stars and is channel-agnostic across Slack, Discord, Telegram, and iMessage, with self-hosted sub-agent orchestration. The security model requires more attention than Cursor or Copilot - but teams that want model-agnostic, messaging-first agent control have a real option here.