Slack Code makes AI coding multiplayer, but Git must remain the source of truth
Slack’s new code channels put Claude, Devin, Copilot and other agents in front of the whole team. That is worth piloting for small changes, not treating as a replacement for engineering controls.
What launched
Salesforce launched Slack Code on August 20, 2026, giving supported coding agents a dedicated, temporary channel for each task. Teams can follow the conversation, plan, code diffs and previews, then retain the archived channel as searchable context.
Initial integrations include Claude, Devin, GitHub Copilot, ChatGPT and Vercel agents. Slack says Code is available on any Slack plan, although customers need separate access to each partner agent and availability can vary by region and agreement.
Where it helps
The useful change is not that more employees can request code. It is that intent, business context and agent execution no longer need to remain trapped in one person’s session. A bad assumption can be corrected while the agent is working, before an expensive review-and-rework loop.
That should reduce latency for small defects, prototypes and maintenance. It also expands the number of people who can initiate changes, making permissions and ownership more important rather than less.
What we would do
Virtual Arc would start with a constrained pilot on low-risk repositories, with no automatic merging and no direct production access. Every result would still pass through code owners, automated tests, security checks and the existing release process.
We would expand only if time to a clean merged pull request, review rework and total cost per completed task improve. If those numbers do not move, a multiplayer channel is merely a clearer place to watch the same bottleneck.
Slack Code matters less as another coding interface than as a change in where control sits. Moving agent work from a private terminal into a channel shared by product, design and engineering can cut handoff latency, preserve intent and expose a wrong turn before it becomes an expensive diff. But Virtual Arc would not let Slack become a new production approval system. Searchable conversation is not a reproducible release control, and a human clicking approve is not a substitute for tests, code ownership and least-privilege access. We would run a narrow pilot on documentation, small UI defects, tests and other reversible work. We would measure time to a clean merged pull request, review rework and agent cost per completed task. If Slack Code merely creates more pull requests for engineers to repair, it has moved the bottleneck rather than removed it. Intent and coordination can live in Slack; Git, CI and the existing production gates must still decide what ships.