What Slack Code is
Salesforce announced Slack Code’s general availability on August 24, 2026, adding AI coding agents as first-class citizens inside Slack channels. Supported agents at launch include Anthropic’s Claude Code, Cognition’s Devin, GitHub Copilot, OpenAI’s ChatGPT in coding mode, and Vercel’s agentic tools. The channel interface is available on every Slack plan; the underlying agent subscriptions are sold separately through each provider’s partnership programme.
The core tension Slack Code addresses: teams building software with AI coding agents typically juggle two disconnected contexts — the IDE where the agent runs, and Slack where the team coordinates. Slack Code collapses that gap by hosting the entire agent session inside a dedicated channel visible to everyone on the team.
How Code Channels work
Each Slack Code session runs inside a “Code Channel” — a standard Slack channel with four purpose-built tabs that track the agent’s progress in real time:
- Conversation. Standard Slack threading. Any teammate can talk to the agent here: ask it to change approach, flag a constraint, or redirect the task entirely in plain language.
- Plan. The agent’s step-by-step breakdown of how it intends to complete the task. Non-technical teammates review and approve the plan before the agent touches any code. This is the explicit human-in-the-loop gate Salesforce is emphasising for enterprise buyers.
- Code. A line-by-line diff view showing every file the agent has changed, created, or deleted, updated in real time. Engineers can comment on specific lines exactly as they would in a standard pull-request review.
- Preview. For front-end and full-stack tasks, a live render of the output — a webpage, an API response, a test report — directly inside the channel, without leaving Slack.
When the task is complete, the channel auto-archives. The full conversation, plan, diffs, and agent output are retained and searchable — turning completed agent sessions into institutional memory rather than disposable ephemera.
Who can use it
Slack Code is explicitly designed to be used by non-engineers. A product manager can open a Code Channel, describe a feature in plain English, and let the agent produce a plan for the team to review — without writing a line of code. Designers can trigger an agent to implement a specific UI change from a Figma link. Marketing can request data-export scripts. The agent handles the translation from intent to implementation; the Plan tab surfaces the logic for non-technical approval.
This is a meaningful positioning shift from most AI coding tools. Products like Cursor or standalone Claude Code target individual engineers working in an IDE. Slack Code targets the whole team — the people who commission software and review it, not just the people who write it. Whether the Plan tab’s plain-language output performs well for non-technical reviewers on real enterprise tasks is the key adoption question to watch.
Enterprise controls and audit
Salesforce is leaning hard on enterprise security as a differentiator. Slack Code sessions run entirely within Slack’s existing permission model: IT admins control which agents are approved for use, which channels or workspaces can run Code sessions, and what data sources the agent is allowed to access. Every action the agent takes is logged under the same compliance exports and Data Loss Prevention policies that govern all other Slack content.
Enterprise Grid customers with Enterprise Key Management (EKM) retain control over their own encryption keys, meaning agent-produced code and conversation logs are encrypted with keys the customer manages — a hard requirement for regulated industries that cannot route code through a third-party model without data-residency guarantees.
For teams evaluating enterprise software development with AI agents embedded in the workflow, the audit trail is as important as the code itself: Slack Code sessions produce a complete, searchable record of what the agent was asked to do, what plan it proposed, what code it wrote, and who approved each step.
What it means for dev teams
Three things to factor in if you are evaluating Slack Code for your engineering organisation:
The friction point moves from the tool to the license. Slack Code itself is free on all plans. The cost is the agent subscriptions — Claude Code, Devin, and Copilot are each separately priced at meaningful enterprise rates. Evaluating Slack Code is really evaluating which coding agent to standardise on and at what seat count, with Slack as the common interface on top.
Institutional memory is the under-discussed feature. Most coding agent sessions are ephemeral: context lives in the IDE and disappears when the window closes. Slack Code’s auto-archiving model means every completed task is a searchable record. Over time, that becomes a form of organisational knowledge — what tasks the team delegates to agents, which plans worked as proposed, which required course corrections, and which agent outputs needed substantial rework before shipping.
The review bottleneck is now visible. IBM’s Bob platform made a similar observation earlier this month: AI agents generate code faster than senior engineers can review it. Slack Code’s Code and Plan tabs are structured to surface review decisions explicitly, but they do not expand review capacity. Teams scaling AI coding agent use will need to think about review workflows and reviewer time, not just generation speed.
Frequently asked questions
What is Salesforce Slack Code?
Slack Code is a Salesforce product that embeds AI coding agents — including Anthropic Claude Code, Cognition Devin, GitHub Copilot, OpenAI ChatGPT, and Vercel agents — directly inside dedicated Slack channels called Code Channels. Each channel has four tabs: Conversation (team discussion), Plan (the agent’s step-by-step task plan), Code (line-by-line diffs), and Preview (live output). The product reached general availability on August 24, 2026.
Which AI coding agents work with Slack Code?
At launch, Slack Code supports Anthropic Claude Code, Cognition Devin, GitHub Copilot, OpenAI ChatGPT in coding mode, and Vercel’s agentic tools. Agent licenses are sold separately through each provider’s partnership programme; the Slack Code channel interface itself is available on all Slack plans including Free.
Do non-engineers need to understand code to use Slack Code?
No. Slack Code is designed so that product managers, designers, marketers, and other non-technical teammates can trigger an agent, review its plan in plain language, and approve or redirect it without writing or reading code. Engineers use the Code diff tab for technical review; non-engineers work in the Conversation and Plan tabs.
How does Slack Code handle enterprise security?
Slack Code runs inside Slack’s existing enterprise controls: IT admins manage which agents are permitted and which channels can run Code sessions. All agent activity is logged and searchable, channels auto-archive with a complete audit trail on completion, and Enterprise Grid customers with Enterprise Key Management retain control over their own encryption keys for agent-produced content.
Sources
Salesforce / Slack — Introducing Slack Code (August 24, 2026)
VentureBeat — Salesforce launches Slack Code, bringing AI coding agents into team channels (August 2026)
The Register — Salesforce adds AI coding agents to Slack with Code Channels (August 2026)