Salesforce Launches Slack Code to Host AI Coding Agents in ChannelsSalesforce Launches Slack Code to Host AI Coding Agents in ChannelsSalesforce Launches Slack Code to Host AI Coding Agents in ChannelsSalesforce Launches Slack Code to Host AI Coding Agents in Channels
August 28, 2026
Salesforce has launched Slack Code, a service that turns any Slack conversation into a dedicated, auto-created channel where a tagged coding agent, Anthropic's Claude, Cognition's Devin, GitHub Copilot, ChatGPT, or a Vercel agent, takes on a development task while the people who
Salesforce has launched Slack Code, a service that turns any Slack conversation into a dedicated, auto-created channel where a tagged coding agent, Anthropic's Claude, Cognition's Devin, GitHub Copilot, ChatGPT, or a Vercel agent, takes on a development task while the people who requested it watch, comment, and intervene in real time. The more consequential shift is not the AI plumbing but the org chart: Salesforce is betting that collapsing the standard Slack-to-Jira-to-IDE relay into one shared surface changes who gets to see software get built, not just how fast it gets built. For enterprise IT leaders already juggling multiple coding-agent subscriptions, that is a workflow decision with real handoff-cost implications, not a feature toggle.
What's new
Salesforce, the CRM software provider that owns Slack, detailed Slack Code in a press release dated August 24, 2026. Katie Steigman, VP of product management at Slack, presented the feature in a video walkthrough, describing a mechanism where tagging a supported coding agent from any conversation automatically spins up a project-specific code channel and pulls in the relevant team members along with context carried over from the originating thread. That auto-creation step is the mechanical core of the pitch: instead of a stakeholder manually opening a new channel, inviting a developer, and waiting for a status update, tagging the agent handles channel creation, team invites, and context-carrying in one action. For an enterprise rollout, that removes one of the more mundane sources of delay in getting a fix started, someone forgetting to loop in the right reviewer.
Inside that channel, participants get separated views: one for the agent's running conversation, one for its plan, and one for the code diffs it produces. Anyone in the channel can review the agent's changes, leave feedback, or steer, pause, or stop the work outright. When the task wraps, the agent packages the work for human sign-off and, depending on the workflow and the agent involved, opens a pull request, a formal request to merge the finished code into the project's main codebase. The channel then archives itself automatically, leaving behind a searchable record rather than a conversation that has to be manually closed out or migrated into a ticketing tool.
Why it matters
Salesforce is explicitly positioning Slack Code against what it calls the traditional one-person, one-agent approach: an issue gets flagged in Slack, a ticket gets raised in a tool like Jira, and the work is then handed to a single developer working alone with a coding agent in an IDE, short for integrated development environment, or terminal. Each of those handoffs is a place where context gets lost or a non-technical stakeholder loses visibility into what is actually happening.
Manoj Chandra Jha, principal analyst at Nord-IQ Research, frames that as the upside: "The model around Slack Code could mean faster fixes, better cross-functional input, and fewer costly handoffs between technical and non-technical staff." Advait Patel, senior site reliability engineer, the role responsible for keeping production software running, at Broadcom, does not dispute the theoretical benefit of that visibility, calling it genuinely useful when it works: "Stakeholders watching how a good prompt gets constructed and corrected is genuinely useful, and right now none of that is observable."
The two assessments describe the same feature from opposite ends: Jha is weighing the promise of the workflow, Patel is weighing whether developers will actually work inside it once the visibility becomes an audience. But Patel is skeptical the visibility survives contact with how developers actually work. "Coding is deep work, and Slack is the interruption machine. Putting them on the same surface is not automatically a win," he said. His sharper concern is that the transparency becomes theater rather than substance: "If developers carry out the messy part privately and paste the clean result into the channel, it defeats the whole purpose. The transparency you get is the performance, not the process." That tension, promise versus performance, is likely to define how enterprise buyers evaluate the product long after the launch headlines fade.
Competitive Landscape
Slack Code's defining structural choice is that it does not run on a single proprietary agent. Instead, Salesforce built it as an orchestration layer that lets teams tag whichever coding agent they already use, positioning Slack as the shared surface rather than a competitor to the agents themselves. That is a different bet than the AI vendors are making with their own IDE-native tools, where the product is the agent itself rather than the room the agent works in.
- Anthropic's Claude, supported natively for tagging inside a Slack Code channel.
- Cognition's Devin, supported for tagging inside the same channel structure.
- GitHub Copilot, supported alongside the other named agents rather than as a separate integration path.
ChatGPT and a Vercel agent round out the list of agents Slack Code supports at launch. That breadth is the pitch to enterprise buyers: rather than standardizing on one coding agent and building process around it, teams can keep whichever agent contracts and workflows they already have and route the collaboration layer through Slack instead. For a procurement team already paying for more than one of these agents, that architecture sidesteps a familiar problem: being forced to standardize on a single vendor's coding assistant just to get channel-based collaboration. Whether that agent-agnostic framing holds up depends on details Salesforce has not yet published, including how authentication and permissions are handled per agent inside a shared channel, and how conflicts are resolved if a team wants to run more than one agent against the same task.
What's next
There is no stated general-availability timeline, and Salesforce has not named launch customers or published adoption figures.
The open questions that matter most are behavioral, not technical. Amit Kumar Jena, AI development manager at IT consulting firm Kanerika, echoes Patel's concern: whether a shared channel actually changes how developers work, or just adds a visible layer of ceremony on top of work still done privately. Patel's notification-scaling worry is unresolved too. Salesforce has not detailed how the platform handles the volume of new per-task channels a busy engineering org could generate, or how that squares with Slack's existing reputation for notification overload.
Slack built its name on making conversation persistent and searchable. Slack Code is a bet that the same trick works on code review, but Patel's warning cuts to the center of it: a channel only tells the truth about how software gets built if developers actually build it there.
For an engineering team lead managing handoffs between product, support, and engineering, the pitch is fewer dropped threads between a bug report and a merged fix, the kind of gap that today costs a written ticket, a Slack thread, and a follow-up meeting just to confirm scope. Whether Slack Code actually shortens that chain, or just makes the existing chain more visible without changing its length, is the question a pilot will answer, not a press release.
-- Aria Lin, Enterprise Technology Analyst
Sources: Slack | Salesforce, "Introducing Slack Code" | GitHub, Copilot Workspace