Slack Code brings coding agents such as Claude Code, Devin, GitHub Copilot, and Vercel's agent into shared channels where teams can assign, watch, review, and archive software work. For managers, the launch points to a shift from private AI use toward visible workflows with human review built in.
For employers trying to move AI from personal experiments into daily operations, Slack's latest product points to a practical design choice: put the agent where the team already works, and make its output visible.
The Salesforce-owned collaboration platform has introduced Slack Code, a product that connects AI coding agents directly to dedicated Slack channels. At launch, the supported agents include Anthropic's Claude Code, Cognition's Devin, GitHub Copilot, and Vercel's agent. Slack Code is available on any Slack plan, although customers still need their own access to the partner agent services.
Slack turns AI coding into a team sport | Hybrion Insights
The core idea is straightforward. Instead of a developer working privately with an agent in a terminal, a user can tag an agent from a Slack conversation. The system then creates a project-specific code channel where the agent's plan, code changes, previews, and discussion are visible to the people involved. When the task is finished, the channel is archived, leaving a searchable record of what happened.
Rob Seaman, Slack's interim CEO, described the shift as a change in where scarcity sits in software work. "One of the things I love about this is that code is no longer the bottleneck," he said in a press briefing before the launch. "Ideas, taste, judgment, craft, those are the things that are the bottleneck, and you've effectively extended the population that can contribute ideas, taste, judgment, and craft to anybody that exists in your Slack."
From private agent sessions to shared software work
Slack and Cognition executives positioned Slack Code as more than a convenience layer. Jeff Wang, president of new enterprise at Cognition, demonstrated a workflow in which a broken feature is reported in an engineering channel, Devin responds, investigates the issue, and opens a pull request. Wang said Devin can identify the code owner and bring that person into the discussion.
The important management detail is not only that an agent writes code. It is that the surrounding team can see the work form in real time. In Wang's demonstration, a designer added a Figma file while the task was underway, and the agent folded that input into the work. The agent later posted code changes, screenshots, and a recorded demo showing that the feature worked.
That kind of visible handoff matters for teams adopting AI. A private agent session can increase one person's throughput, but it can also hide assumptions, missed requirements, or weak verification. A shared channel gives product managers, designers, engineers, and other stakeholders a place to inspect the task before it reaches a formal review step.
Slack is also expanding how agents appear across the product. The company is adding agent direct messages, an Agents tab with session status and a stop button, and an "Add to Slack" flow for deploying agents from platforms including Lovable, n8n, OpenAI, LangChain, and Airtable. Slack said OAuth and configuration are automated in that setup.
Human review remains the operating model
The product arrives as companies are still struggling to convert agent enthusiasm into measurable value. Gartner predicted last year that more than 40 percent of agentic AI projects will be canceled by the end of 2027, citing rising costs and unclear business value. McKinsey's latest State of AI research found that 62 percent of organizations are at least experimenting with AI agents, but only about a third have started scaling AI overall, and 39 percent report bottom-line impact.
Slack's argument is that public work can reduce some of the risks that come with broader access to code generation. Katie Steigman, Slack's vice president of product, said the shared model helps teams challenge weak output. "The multiplayer part is a guard against that, actually, because people can see your work, people can comment on your work," she said.
Steigman, who is a product manager rather than an engineer, said she normally tags an engineer when she raises a pull request. "Almost every time, an engineer will say something like, 'Come on, you can make that a little bit tighter,' or they'll actually give it some specific technical guidance, and the agent will take one more rev and produce code that has been touched by an engineer to a certain extent."
Cognition also presented internal evidence of speed gains. Wang said the company's merged pull request count has risen 10 times in recent months, while headcount has increased by about 40 percent. He said employees can start a Devin task, move to another piece of work, and then launch additional agent tasks.
For business leaders, the practical lesson is not that every employee should ship software without oversight. It is that AI-assisted work needs a visible workflow, clear ownership, and review points that match the risk of the task.
Permissions and governance will decide adoption
Enterprise buyers will likely focus on Slack Code's security model. Seaman said agents act on behalf of the user who invokes them and inherit that user's access controls in Slack and connected systems. "There's no god permissions or bot-level permissions," he said.
Steigman added that when an agent creates a code channel, it receives the conversation context that triggered the work. Wang said Devin runs in isolated sandboxes with "minimum viable access," including an optional configuration with no internet access.
Those details are important because agent deployments often create governance friction. If every agent requires a new identity, new permissions, and separate monitoring, IT and security teams face another sprawl problem. Slack is instead pitching agents as extensions of existing users, with standard pull requests continuing through GitHub review and release processes.
That does not remove the need for policy. Managers will still need to decide who can invoke agents, what kinds of repositories they can touch, when human approval is mandatory, and how archived channel records should be used in audits. Slack Code's bet is that AI coding becomes safer and more useful when it is pulled into the open. For human-led teams, that is the right question to test: not whether an agent can produce code, but whether the organization can supervise the work well enough to trust the result.
Reported by Hybrion Insights with reference to VentureBeat AI.
hybrion
Better together. Built for what's next.
Stay updated
Get insights on AI, teams, and the future of work.
