Slack Just Made Agent Work Multiplayer. Your Website Needs the Same Upgrade.
Slack launched code channels: dedicated spaces where AI coding agents work alongside a team, with live diffs and previews in the channel, approvals routed to a person before high-stakes changes ship, and a searchable archive when the work is done. It is the strongest statement yet from a major platform that agent work belongs in the open, under governance — and it sharpens a question every marketing team should ask about their website.
What did Slack actually launch?
The mechanics, briefly. Mention an agent and a code channel spins up around the task. The team watches the work happen — code changes, live HTML previews, agent status — instead of one person watching it alone in a browser tab. Changes that matter can route to a human for approval inside the channel. When the project completes, the channel archives but stays searchable, so the work leaves a record. Permissions inherit from the Slack workspace, and a new Agents and Tools tab manages agent conversations in one place. Launch partners include Claude, Devin, GitHub Copilot, and Vercel, with ChatGPT announced as coming.
Slack frames the shift as moving agent work from single-player to multiplayer — in their words, "work is better, faster, and more creative when it's done together, in the open." They also shared a telling internal metric: most of their code channels open and close within a single day, from idea to merged pull request. Speed and governance, in the same sentence — that pairing is the story.
What does this launch validate?
Look at the primitives Slack chose to build: visibility for the whole team, approval by a named person before consequential changes ship, attribution of who and what did the work, and a durable record afterward. Those are governance primitives. When the largest workplace-communication platform ships them as the default way to work with agents, the debate about whether AI work needs oversight is over. The question that remains is where each kind of oversight should live.
The converging consensus
Webflow brought agents into the CMS. Netlify and Vercel brought them into the deployment pipeline. Slack brought them into the conversation. Every layer of the stack now agrees: agents do the work, humans approve it, the record survives.
What's the difference between governing a task and governing a website?
A code channel is built around a unit of work. It is born when the task starts and archives when the task ends — a clean, self-contained record of one change. That is exactly right for a task. But your website is not a task. It is an asset that accumulates hundreds of changes, whose pages compete for AI-search citations, whose brand voice has to survive every future edit, and whose worst day is the one where something breaks and nobody can say which change did it.
Chat-layer governance vs. website-layer governance
| Question | Code channel answers it | Website management layer answers it |
|---|---|---|
| Who approved this change before it shipped? | Yes — in the channel | Yes — role-gated, recorded on the request |
| What did the agent change, live, as it worked? | Yes — diffs and previews | Via deploy previews on each request |
| What are all the pages on the site right now? | — | Yes — synced inventory, mapped to source |
| How ready is each page to be cited by AI search? | — | Yes — scored per page, tracked as it changes |
| What context should every future change follow? | — | Yes — approved brand voice, claims, personas |
| What did the site look like 40 changes ago — and can we go back? | — | Yes — full version history, one-click restore |
| Which changes were AI-authored, across months? | Scattered across archived channels | One attributed feed per site |
None of that is a criticism of Slack Code — a chat platform should not maintain your page inventory or score your structured data, any more than a website layer should host your conversations. The point is that archiving conversations about work is different from governing the asset the work accumulates into. Both matter. They are different layers.
A channel archives a task. A website is an asset. Govern the conversation in chat — govern the site where the site lives.
How do the layers work together?
Cleanly, because they meet at the same primitives. A team discusses a website change wherever it already talks — increasingly, in a channel with an agent present. The change itself flows through a governed request: written in plain language, translated into a scoped agent handoff carrying the organization's approved brand context, previewed on a real deploy, approved by a named role, shipped, attributed, and restorable forever after. The conversation explains the moment. The management layer holds the history.
This is also why the direction of travel favors integration rather than competition. Approval requests belong wherever the approver already is — Slack's launch makes that case well — and a website management layer that can notify a channel the moment a preview is ready gets the best of both: chat-speed decisions, asset-grade records. That integration is on our roadmap for exactly this reason.
What should your team take from this?
Questions to ask about your agent workflow
- Can the whole team see agent work in progress, or does it live in one person's tab?
- Does anything require a human approval before it reaches production?
- Is every change attributed — human or agent, requester and approver?
- Does the record survive the task: months later, can you reconstruct what changed and why?
- For your website specifically: do you have a page inventory, AI-search scores, and a one-click way back to any prior version?
- Do approvals reach people where they already work, instead of demanding another tab?
If the first four questions have good answers and the fifth doesn't, you have governed tasks and an ungoverned asset — the common state of teams adopting agents quickly. The tools to build and run your website with AI already exist. Dispatch makes them manageable for your organization: the plain-language requests, attributed activity, per-page AEO scores, and one-click restore that turn agent speed into organizational capability.
What are Slack code channels?
Code channels are dedicated Slack spaces where AI coding agents work alongside a team. A channel spins up when an agent is mentioned, shows code diffs and live previews as the agent works, routes high-stakes changes to a person for approval, and archives into a searchable record when the work completes. Launch partners include Claude, Devin, GitHub Copilot, and Vercel.
Does Slack Code replace a website management layer?
No — they govern different things. A code channel governs a task: one conversation, one piece of work, archived when done. A website management layer governs an asset: the ongoing inventory of pages, their AI-search readiness, the full version history, brand context for every future change, and one-click restore. Teams that run their website with AI benefit from both, at different layers.
Why are approvals becoming standard for AI agent work?
Because agents produce real production changes at a pace no team can review after the fact. Every serious platform is converging on the same primitive: the agent does the work, a human approves before it ships, and the decision is recorded. Slack routing approvals into channels, Vercel's approval-gated architectures, and Dispatch's role-gated change requests are all versions of the same conclusion.
What does 'multiplayer' agent work mean?
It is the shift from one person running an agent privately — in a terminal or a browser tab nobody else sees — to agent work happening where the team can watch, comment, approve, and learn from it. The output stops being one person's secret productivity and becomes organizational capability, with visibility and accountability built in.
How should a marketing team combine chat-based agent work with website governance?
Let conversations live in chat and let the website's record live in the management layer. Discuss and iterate wherever your team already talks; make the change itself flow through a governed request — scoped to a branch, previewed, approved by a named role, attributed, and restorable. The chat thread explains the moment; the management layer holds the history the site accumulates for years.
Agent work went multiplayer this week. The websites those agents increasingly run deserve the same upgrade — visibility, approvals, attribution, and a way back — attached to the asset itself. Move beyond the limitations of your CMS without giving up the visibility, control and governance your team depends on.
Dispatch Team
Writing about AI governance, collaboration, and operations — helping teams turn AI from scattered experiments into shared organizational capability.