Writing

Slack Code moves AI coding out of the terminal into group chat

AI coding has mostly been a solo act in the terminal. You prompt an agent, receive a diff or a terminal output, and no one else sees the exchange unless you decide to share it. That has always felt like the weak point to me: the work happens in a private loop, and review only enters after the fact.

Slack Code changes that. It brings agents like Claude Code, Devin, GitHub Copilot, and Vercel into shared channels where teams can prompt, review, and steer work together. Agents run with the invoking user's permissions, which keeps accountability tied to a person rather than a bot with its own access. The channel preserves diffs, previews, and activity as a searchable record.

The practical shift is that code generation becomes a team-visible activity, not a hidden one. When a change lands in a shared channel, a teammate can question it before it becomes a pull request or gets deployed. Instead of a developer quietly generating a change and hoping it fits the codebase, the team can watch the change take shape. That reduces the risk of unvetted changes and makes review part of the loop, not an afterthought.

Shared context prevents silent drift. If every agent interaction leaves a trace in the channel, the team can see why a change was made, which agent produced it, and who invoked it. No one has to dig through a terminal or ask in a direct message to reconstruct what happened. The shared history becomes the source of truth for how the code evolved.

The searchable record also changes how a team accumulates knowledge. Instead of each developer keeping a private collection of prompts and patches, the channel builds a shared memory. New people can read through it to understand how agents are being used, which lowers the cost of getting up to speed. That kind of context is easy to lose when everything lives in individual terminals.

There is a difference between review at the end and review during the work. In a channel, someone can steer direction while the agent is still generating, rather than reacting to a finished result. That is a subtle but important change. The agent becomes a participant in a conversation, not a remote process that returns a result hours later.

Not every coding task needs to move into a shared channel. Some exploration still belongs in a private workspace. But for work that affects shared code, visibility beats a hidden loop. The VentureBeat breakdown describes this as dragging AI coding out of the terminal and into the group chat, and that framing feels right. It makes the process legible to the people who will eventually have to live with the code.

Originally covered at venturebeat.com ↗