Slack, owned by Salesforce, has introduced Slack Code, a feature that embeds AI coding agents directly into dedicated Slack channels so teams can observe, steer, review and ship code together. Launch partners mentioned include Anthropic (Claude Code), Cognition (Devin), GitHub (GitHub Copilot) and Vercel. Slack Code is available on any Slack plan at launch, but customers must have their own access to the partner agents.
How it works
Slack says the product shifts AI-assisted coding from a one-person, one-agent interaction in a terminal to a multiplayer workflow inside Slack. When someone tags an agent in a conversation, the agent creates a project-specific code channel. The agent’s work is visible in the channel — code diffs, live previews and a running plan appear in dedicated tabs — and when the task is finished the channel is archived, leaving a searchable audit trail.
In demonstrations described by Slack and partner Cognition, a typical flow begins with a bug reported in an engineering channel. The agent acknowledges the issue, investigates and may open a pull request. In Cognition’s demo, Devin identified the code owner, used Chrome and DevTools to test the feature, and then posted code changes, screenshots and a recorded demo proving the fix worked. The work can then continue in a code channel where engineers, product managers and designers can all contribute; in the demo a designer dropped a Figma file into the channel mid-task and the agent incorporated it.
Cloud-based agents in this model generate an auditable verification loop: because the agents run in the cloud rather than on a developer’s laptop, they can produce records that the work completed is correct.
Product and integration updates
Slack also announced a broader rework of how agents appear in the product: agent DMs that act like colleague conversations, an Agents tab with live status and a stop button for each session, and an "Add to Slack" flow that lets teams deploy agents from platforms such as Lovable, n8n, OpenAI, LangChain and Airtable with OAuth and automated configuration.
The strategic case: code is no longer the bottleneck, humans are
Slack executives argue AI has inverted the economics of software development: writing code is less of a scarce resource and the bottleneck has moved upstream to human judgment, taste and creativity. Rob Seaman, Slack’s interim CEO, said Slack Code expands who in an organization can contribute those human elements.
Cognition provided internal metrics: Jeff Wang, president of new enterprise at Cognition, said their internal merged PR count increased tenfold in recent months while headcount rose roughly 40 percent. Wang described a workflow where engineers fire off tasks to Devin, switch to other work, and run many agents concurrently.
Wang predicted that routine toil — fixing bugs, CI/CD issues or vulnerabilities — will be increasingly automated, leaving tasks that require creativity and business logic to humans. He also suggested that while humans gate merges today, that could change to more automation within a year.
Quality concerns and the ‘AI slop’ risk
A key objection to democratizing agentic coding is quality: if anyone can summon an agent, will organizations be flooded with plausible-looking but poor outputs known colloquially as “AI slop”? Slack’s leaders argue visibility mitigates that risk because multiplayer channels let colleagues see, comment on and correct agent work. Katie Steigman, Slack’s VP of product, said product people tagging engineers on PRs and getting technical guidance results in agents producing code that has been refined by humans.
Seaman framed the idea more broadly, saying multiplayer AI tends toward higher quality and less duplication; he cited Shopify’s approach of restricting agentic coding to public channels to disseminate work across the company. Still, independent research offers a cautionary view: Gartner predicted more than 40 percent of agentic AI projects will be canceled by the end of 2027 due to rising costs and unclear business value. McKinsey’s State of AI survey found 62 percent of organizations are experimenting with AI agents, but only about a third have begun scaling AI and 39 percent report any bottom-line impact.
Security and permissions: agents act with user permissions
For enterprise buyers, Slack emphasizes its permissions model. Seaman said agents inherit the permissions of the user who invokes them and operate on behalf of that user using the user’s ACLs within Slack and connected systems. There are no “god” or bot-level permissions that grant access beyond what the user already has.
Steigman added that when an agent spins up a code channel, the agent only receives the conversational context that triggered it. Cognition’s Wang said Devin runs in isolated sandboxes with “minimum viable access,” and offered an optional no-internet mode. Because agents create standard pull requests into GitHub, existing release gates and review processes remain in place.
This model contrasts with standalone agent platforms that require new service identities and separate permission management; Slack positions agents as extensions of existing users to reduce shadow-IT and preserve existing governance.
Timing and strategic importance for Salesforce
Slack Code arrives while Salesforce faces investor scrutiny: the company’s shares fell roughly 18 percent over the year through January, and Wall Street has questioned how AI might affect demand for traditional enterprise software. In December, OpenAI hired Slack’s former CEO Denise Dresser as chief revenue officer, and Rob Seaman — Slack’s former product chief — became interim CEO.
Salesforce has been positioning Slack as the AI entry point for work. In January it shipped a rebuilt Slackbot powered by Anthropic’s Claude, which the companies said became the fastest-adopted feature in Slack’s 27-year history. Slack Code extends that strategy from answering questions to producing artifacts — working code, prototypes and documents — inside Slack itself.
Notably, the announcement marks a shift from a defensive posture earlier in 2025 when Reuters reported Salesforce had moved to block rival AI firms from accessing Slack data. Today’s approach is more open: Slack is courting AI companies as partners and plans to open code channel APIs to third-party developers. Slack says software engineering is the first use case; the company expects marketing campaigns and legal document reviews to follow in dedicated agent channels.
What this could mean for software development
Slack does not claim terminals will disappear. Seaman expects a division of labor: most work will move to multiplayer environments, while deep, single-person, immersive engineering work will continue in terminals. Early adopters may inspect every diff and step, then progressively trust the system and reduce oversight.
McKinsey’s research suggests organizations that redesign workflows around AI — rather than merely tacking AI onto existing processes — capture value. Slack Code is presented as precisely that: a workflow redesign that makes the team, rather than the individual, the unit of AI adoption. If it succeeds, Slack could change not only where colleagues talk but where software is made; if it fails, transparency may simply make poor outputs more visible to everyone.
As Jeff Wang put it, the bottlenecks have shifted. Slack Code attempts to find out whether making development a crowd activity will shrink those bottlenecks or only amplify them.
