Platform
Slack Code: code channels, the agents inside them, and what security will ask
Slack Code turns a coding agent into a channel member. Tag Claude, Devin, Copilot or another agent from any conversation and it opens a code channel with diffs, live previews and a searchable record of what it did. This is what ships, which agents are in it, what it costs, and the questions your security team will put to you before the first channel opens.
Most coding agents work in a terminal or an editor that only one person can see. The pull request is the first moment anyone else finds out what the agent did, and by then the reasoning is gone.
Slack Code moves that work into a channel. You tag a coding agent from any Slack conversation, and it opens a dedicated code channel for the task where the whole team can watch the diff build. Salesforce launched it on August 20 and used Dreamforce to add a second wave of agents.
Rob Seaman, who runs Slack, put the pitch in one line: code channels bring the work into the open, where the rest of the team can see it, shape it and build on it.
The pitch is right, and it is also the reason your security team will have questions. A channel that can reach a repository and a deployment target is a new surface. This post covers what a code channel is, which agents you can put in it, what it costs, and the five questions to answer before you switch it on.
What a code channel is
A code channel is a Slack channel that an agent creates for one piece of work. You tag an agent in an ordinary conversation, describe the task, and the agent spins up the channel. Inside it you get a conversation tab, a diff view showing old and new lines side by side, and a live HTML preview for anything with a front end.
When the task is done the channel archives itself, and the history stays searchable.
The design choice worth noticing is the archive. A coding agent’s transcript is usually the least durable artefact in the process. Here it is a Slack channel, which means it is retained, searchable and subject to whatever retention policy your workspace already applies. Six months after a change ships, the conversation that produced it is still there.
Work in the channel is visible to everyone in it, which is the point. For anything going to production, the agent packages the change for a person to sign off. Slack describes this as expert sign-off before shipping, and the wording matters: the agent prepares, a human approves.
The agents you can put in a channel
Slack Code is not an agent. It is a place for agents you already license. At launch the roster was Claude from Anthropic, Devin from Cognition, GitHub Copilot, ChatGPT, and Vercel’s agents. At Dreamforce, Slack added ten more: Cedar, Coderabbit, Datadog, Factory, Lovable, Mistral, NanoClaw, Replit, Rhythms and Snowflake.
Two things follow from that list. The first is that a code channel can hold more than one kind of agent. A coding agent writes the change, a review agent like Coderabbit comments on it, and an observability agent like Datadog’s can tell you what the change did after deploy. Slack is positioning the channel as the place those three meet.
The second is licensing. Slack Code itself is available on any Slack plan, but each partner agent needs its own access. If you have a Claude Enterprise seat, Claude works in the channel. If you do not, tagging Claude gets you nothing.
Budget for the agents, not for Slack Code. Vercel’s CTO Malte Ubl gave the reason the arrangement appeals to vendors: Slack is already where their teams work, so it is where working with agents makes sense.
Claude is the default model underneath Slack’s own AI since the Claudeforce partnership, which is separate from Claude appearing as a coding agent in a code channel. The first is Slackbot’s brain. The second is a licensed agent you invite.
Where it fits for Salesforce work
For a team building on Salesforce, a code channel is a shared window onto the same agents you might already run in a terminal. If your developers use the Claude Code plugin for Salesforce locally, the same Claude can work a task in a channel where the admin and the product owner can see the Apex and the metadata change before it reaches a sandbox.
The loop differs from Agentforce Vibes, which lives inside Salesforce’s own tooling, and from Agentforce DX, which treats an agent’s definition as source you deploy. Slack Code does not replace either. It sits in front of them as the conversation, with the artefact still ending up in a DX project and a deployment pipeline.
The practical win is review. A change to a validation rule or a trigger that an agent proposed can be discussed by the people who understand the business rule, in the place they already discuss it, before anyone runs a deploy. The practical risk is the same sentence read the other way. The people in the channel can steer an agent that has a path to your org.
The five questions security will ask
Slack says Slack Code inherits Slack’s existing security model, permissions and admin controls. True, and it answers less than it sounds like it does, because the agents bring their own access. Work through these before the first channel.
Which repositories can an agent reach from a channel, and under whose credentials. A coding agent in a code channel is authenticated to your source control somehow. Decide whether that is a service identity with a narrow scope or a person’s token, and prefer the first.
Which orgs and environments. For Salesforce work, the agent needs an authenticated org to deploy or retrieve against. Make that a sandbox, and make production deploys something the agent packages for a person to run through your normal release process, not something the channel can trigger.
Who may open a code channel and who may be in one. Slack’s permissions decide who can tag an agent. A code channel with a customer-facing consultant, an agent with repo access and no engineer in it is a configuration you can create by accident.
What the agent can see in the channel. An agent reads the conversation to do its job. If someone pastes a credential, a customer record or a production log into the channel, the agent read it. The same rule as any channel with an AI member applies, and the permission review you would run before turning on an Agentforce employee agent applies here too.
How long the archive lives and who can search it. The archived channel is the audit trail, which is good. The archive is also a searchable record of every diff an agent produced, which your data-classification policy may have an opinion about.
None of these is a reason not to use it. All of them are decisions you should make once, in writing, before the first developer discovers the feature on their own.
How I would introduce it
Start with one team and one class of work: internal tooling, test coverage, documentation, anything where a bad change is cheap to reverse. Give the agents a dedicated identity per environment and keep production out of reach. Let the team run a few tasks end to end, including the archive, and read the archived channels as if you were an auditor.
After a month you will know three things the announcement cannot tell you: whether the diffs are readable enough for non-engineers to review, whether the agents’ repository access is scoped the way you thought, and whether people prefer the channel to the terminal. The last one decides adoption.
If the answer is yes, widen to Salesforce metadata work, with the channel feeding a sandbox and a person running the deploy. The employee agents you already run in Slack and the surfaces that show live Salesforce data are the natural neighbours: one channel where the change is proposed, one where its effect on the pipeline is visible.
Understanding the basics
What is Slack Code?
Slack Code is a feature that lets a coding agent work inside a dedicated Slack channel. You tag an agent such as Claude, Devin, GitHub Copilot or ChatGPT from any conversation, and it opens a code channel with a conversation tab, a side-by-side diff view and live HTML previews. The channel archives itself when the task is done and stays searchable.
Which agents work in Slack Code?
At launch: Claude, Devin, GitHub Copilot, ChatGPT and Vercel’s agents. At Dreamforce 2026 Slack added Cedar, Coderabbit, Datadog, Factory, Lovable, Mistral, NanoClaw, Replit, Rhythms and Snowflake. Each agent is licensed separately by its vendor.
How much does Slack Code cost?
Slack Code is available on any Slack plan. The cost is in the agents, because each partner agent requires its own access through that vendor. A team without a licence for a given agent cannot use it in a channel.
Can Slack Code deploy to production?
Slack describes production changes as packaged by the agent for expert sign-off before shipping, so a person approves. What the agent can reach depends on the credentials and environments you give it, which is the decision to make before enabling the feature.
The channel is the audit trail
The strongest argument for Slack Code is not speed. Coding agents were already fast. The argument is that the reasoning, the diff and the review happen in one place that is retained and searchable, in front of the people who own the business rule being changed.
Make the access decisions first, keep production behind a person, and the channel becomes the record you wished you had for every change an agent ever made.
If you want coding agents scoped properly before they reach your Salesforce orgs, with sandbox-only access and a review path that ends in a person, talk to us, or run the Apex security scanner on what the agents have already written.
