Slack Code: Multiplayer AI Coding and the Agent Control Plane Landscape

🌏 中文版

On August 20, 2026, Salesforce announced Slack Code at Dreamforce 2026. Marc Benioff's launch post read: "Don't code alone. Humans and agents. Same channel. Same work." The core claim is simple — most AI coding work today happens between one person and one agent in a terminal, invisible to everyone else. Slack Code makes that work visible, steerable, and approvable by the whole team.

This post covers three things: how Slack Code works, the competitive landscape it sits in, and the platform calculus behind the fight over who becomes the agent control plane.

How Slack Code Works

The flow: @tag a coding agent in any Slack conversation → the agent creates a project-specific Code Channel and pulls in relevant team members → the channel has separate tabs for conversation, plan, code diffs, and live HTML preview → anyone can pause, redirect, or stop the agent → high-stakes actions like merging to production require human sign-off → the channel auto-archives on completion, leaving a searchable audit trail.

Five launch partners:

Slack's blog post claims over 70% of code channels close within a single day, from idea to merged PR. Slack Code is available on all Slack plans; agent access is a separate purchase.

This didn't come out of nowhere. TNW reported in May 2026 that Salesforce expected to spend $300 million on Anthropic tokens this year, and Benioff wanted coding inside Slack next. Three months later, here it is.

This Isn't a Slack-Only Idea

The "multiplayer AI coding" space Slack Code enters is already crowded in 2026. According to Nori's control plane comparison and the Agent Cockpit Wars analysis, the market has stratified into at least four layers.

Chat-First: Direct Competitors to Slack Code

Amika lets engineers and non-engineers spin up cloud agents from Slack, Linear, GitHub, or CLI, with multiplayer chat in the same live session. It supports Claude Code, Codex, and OpenCode with no model lock-in. The key difference from Slack Code is that Amika provides its own sandboxed cloud execution environment, rather than just showing results in a chat interface.

Replicas (YC Spring 2026) follows a similar pattern: delegate tasks from Slack, Linear, GitHub, or an API, and the agent runs Claude Code or Codex inside an isolated VM, returning a pull request. Billing is per active workspace minute.

Dedicated Multiplayer Workspaces

These products don't embed agents into existing chat tools — they build the workspace from scratch for teams and agents.

Superconductor has the broadest coverage: it supports Claude Code, Codex, Amp, Factory Droid, Grok Build, OpenCode, Pi, and Cursor agents, with shared sessions, live previews, guided code review, and a unique feature — benchmarking agents against real PRs from your own codebase. Currently free; users bring their own API keys.

Shake takes a kanban approach: the AI agent sits alongside human teammates as a first-class assignee on the same board. Assign a card to Shake, and it reads the codebase, writes code, opens a PR, deploys, and reports back — all in a single thread per card. PMs and engineers track progress in one interface.

Poly is more focused: it's "multiplayer Claude Code." A shared room where each person picks their own model and thinking effort, but every prompt, diff, and dollar is visible to everyone. Currently free during open beta.

Other notable entries include: Delta, which unifies conversation and worktree into a single replicated space where review comments stay anchored as code evolves; Nimbalyst, an open-source visual workspace where agents edit shared docs, mockups, and diagrams; YappJam, which adds built-in voice and video chat; and Modulus, which focuses on agent-to-agent collaboration with cross-repo shared memory.

Control Planes and Schedulers

This layer doesn't build the frontend — it builds the operating system for agents: scheduling, triggering, organizational memory, and cross-session management.

Tembo launches agents from Slack, Linear, GitHub, schedules, or webhooks. Foreground development happens in live cloud sessions; background agents report back when ready. Nori's comparison considers Tembo the closest to a complete control plane competitor.

Traycer calls itself the "nerve center for agentic coding," running Claude Code, Codex, OpenCode, and Cursor in the same workspace with built-in planning, debugging, reviewing, and documentation workflows.

Nori Sessions charges a flat $50/runtime/month with native cron and webhook triggers, Slack and Discord control, and portable organizational skillsets — its main pitch is "you don't need to change your workflow when you change your agent."

Platform Players: They Are the Agent and the Control Plane

Cursor 3, after SpaceX's acquisition, runs up to 8 parallel agents in cloud VMs. Codex App is OpenAI's desktop command center with a Skills system (Figma, Linear, Vercel integrations) and Automations (scheduled agents). These companies don't wrap other people's agents — they are the agents, and naturally the control plane too.

The Real Question: Who Does Slack Code Help

Back to Slack Code. The core problem it solves is visibility, not productivity.

Constellation Research analyst R "Ray" Wang said it directly in InfoWorld's coverage: "Coding is deep work, and Slack is the interruption machine. Putting them on the same surface is not automatically a win." Multiple stakeholders intervening mid-task could create competing suggestions, and large enterprises might end up managing hundreds of additional channels with unresolved notification problems.

Slack Code's sweet spot is narrow: well-scoped, cross-functional tasks — changing copy, fixing a button, prototyping an internal tool. A PM tags an agent, a designer drops a Figma file into the channel, the agent integrates and produces output, an engineer reviews. This works because the task itself isn't complex enough to require sustained deep focus.

Complex engineering work — refactoring, performance tuning, system design, cross-service debugging — won't improve by moving to a Slack channel. These tasks need one person (or a small group) holding complete context over extended periods, not a channel where everyone watches in real time. VentureBeat's coverage quotes Slack EVP Rob Seaman acknowledging this: "There's going to be deep, immersive, intensive, single-player thought work that's going to happen in terminals."

The same InfoWorld piece raises a security concern: Slack channel membership shouldn't automatically equal code repository access permissions. Adding someone to a code channel could effectively grant code access — a privilege escalation risk that hasn't been addressed.

"The Agent Is Interchangeable. The Control Plane Is the Product."

That quote from the Agent Cockpit Wars analysis precisely describes Slack Code's strategic position.

Salesforce doesn't build coding agents. It lets Anthropic, Cognition, GitHub, Vercel, and OpenAI compete inside Slack, which serves as the front door. Each agent vendor needs Slack to reach enterprise customers — much like apps need the App Store. What Salesforce captures is the workflow context and usage data, not the value of the models themselves.

This logic holds only if enough work actually happens inside Slack. So far, Slack Code captures only the simple-task segment. But Slack's advantage is its installed base — over 500 AI apps already in its marketplace — and the fact that "where the conversation starts" is naturally where agents are easiest to invoke. Software engineering is step one; Slack has already announced that code channel APIs will open to the broader developer community, with marketing and legal agent channels in the roadmap.

The real risk isn't that Slack Code fails to execute. It's that developers don't buy in. Browsers, IDEs, and terminals have all tried to be the "universal front door," and developers almost always return to the tool best suited for each task. Whether Slack can break this pattern depends on whether the volume of work it handles is truly "enough" — those 70% of code channels closing within a day may not indicate efficiency, but rather that the tasks themselves never needed more than a day.

The Bottom Line

Slack Code is Salesforce's platform aggregation strategy, not an engineering productivity tool. It packages the visibility problem of AI coding work as "multiplayer coding," letting management and cross-functional teams see and intervene in agent work processes.

For engineering teams, worth watching but no rush to adopt. Truly complex engineering work won't move from terminals to Slack. If a team already has cross-functional simple tasks for agents — copy changes, bug fixes, prototypes — Slack Code can reduce some tickets and meetings. If the work is deep technical work, staying in terminals and IDEs with Claude Code or Cursor makes more sense.

What's more worth watching is the convergence of the "agent control plane" space. Agents themselves are increasingly interchangeable; differentiation is moving to the control layer — scheduling, triggers, team visibility, organizational memory, cross-agent coordination. Slack Code leverages Slack's installed base to grab the front door, Superconductor and Tembo compete through deep integration for the workspace, Cursor and Codex App use their own models to grab the entire stack. This fight is far from over.

References

• Slack Code: Where Your Team and Agents Build Together
• Slack Code Product Page
• The Verge — Slack is launching collaborative vibe-coding channels
• SiliconANGLE — Salesforce introduces Slack Code
• VentureBeat — Slack wants to drag AI coding into the group chat
• InfoWorld — Salesforce wants to move AI coding into a shared workspace
• The Next Web — Slack launches Slack Code
• The Register — Slack Code taps into collective vibe
• Gizmodo — Slack Has Launched a Vibe Coding Tool
• RuntimeWire — Slack launches Code with Anthropic, GitHub, Cognition and Vercel agents
• Agent Cockpit Wars: Evaluating the New Wave of AI Coding Orchestrators
• The Best AI Coding Agent Control Planes in 2026 — Nori
• Superconductor
• Amika
• Shake
• Poly
• Delta
• Nimbalyst
• YappJam
• Modulus
• GitHub Copilot Coding Agent Guide (in Chinese)
• Internal AI Coding Agent Adoption (in Chinese)