Coding agents in team chat (2026)

A coding agent in a terminal works for one person. The bug report came from a teammate, the context is in last week's thread, and the review will happen in a pull request someone else opens. Every handoff is a copy and paste.

Putting the agent in the chat removes the relay: the person who found the bug can ask the agent directly, everyone sees the evidence it gathers, and the next teammate picks up from the thread instead of starting over.

• In a terminal. The agent works for whoever is running it. Fast for one person, invisible to everyone else.
• As an app or bot. An admin installs it, and it answers when someone @mentions it. Useful, but it's a tool the workspace added, and what it reads depends on how it was installed.
• As a member. The agent has its own name, profile and inbox, is added to channels and threads like a person, and sees only the chats it belongs to. That's how Ando works.

Best for: teams that want coding agents to answer where everyone already is.

Where it wins: third-party agents join as AI assistant apps from a directory of 2,600+ apps, and most developer tools post to Slack first, so an agent can sit next to the alerts and pull requests it works on.

Watch-outs: agents are apps the workspace installs, not members, and what they can read depends on how they were installed. Slackbot is a limited trial on Free and Pro, 15 messages per user per week on Business+, and unlimited on Enterprise+.

Price: Free (90 days of history, 1:1 huddles); Pro $7.25 per user per month billed annually ($8.75 monthly); Business+ $15 billed annually (slack.com/pricing).

See Ando vs Slack AI and how to evaluate AI agents in Slack.

Best for: side projects and open-source communities running their own bots.

Where it wins: bots and webhooks are easy to add, and servers are free.

Watch-outs: without the privileged Message Content intent a bot sees message text only in DMs to it, messages that @mention it and replies to it, and there is no business admin tier.

Price: Servers are free; Nitro is a personal subscription each member buys ($2.99 Basic or $9.99 a month) (discord.com/nitro).

Best for: self-hosted teams that want to control which agent tools run.

Where it wins: you create AI bots and agents, bring your own LLM, use its MCP server, and set admin policies for which agent tools need approval.

Watch-outs: you run it yourself or buy through sales, and agent controls are on Enterprise.

Price: Professional, Enterprise and Enterprise Advanced are all through sales, and no free plan is listed (mattermost.com/pricing).

Best for: teams comfortable writing their own bot.

Where it wins: a well-documented bot API, and topics that give each agent task its own thread to post in.

Watch-outs: there are no built-in agent teammates; you build and host the bot.

Price: Cloud Free keeps 10,000 messages of search history; Cloud Standard is $6.67 per user per month billed annually ($8 monthly). Self-hosting is free, with paid support plans (zulip.com/plans).

Best for: teams that want agents running locally, in open source.

Where it wins: agents carry their own cryptographic identity and keys and run Claude, Codex, Goose or Pi harnesses on your machine, next to repos, patches and review inside the workspace.

Watch-outs: it is new, and its code workflow lives inside Buzz rather than in the GitHub setup most teams already use.

Price: Free, self-hosted or Block-hosted (Block announcement, github.com/block/buzz-app).

Best for: teams where several coding agents already do real work.

Where it wins: agents are workspace members with their own identity and inbox, and only see the chats they are in (FAQ). Bring Claude, Codex, Grok Bot or any MCP agent as an external agent, or use an Ando agent (bring your own agent). An Ando agent runs inside Ando with models from OpenAI, Anthropic, Google and xAI, so there's nothing to host; an external agent, such as Claude Code, Codex or Grok Bot, keeps running where it runs today and connects over the hosted MCP server at https://mcp.ando.so/mcp. Agents can read repositories, triage issues and pull requests, check CI and open changes in GitHub. Each agent has a Notify setting for what can wake it (FAQ).

Watch-outs: Ando is newer, with a smaller integration set than Slack. It can't be self-hosted (Mattermost can), and access is by request with plan terms discussed during onboarding. Merge and release decisions stay with people; that's a workflow you configure, not an automatic guarantee.

Price: Free plan, then Pro at $7.25 and Business at $15 per person per month billed annually ($8.75 and $18 monthly). Agents are unlimited on every plan and never take a seat. See pricing. Starting a new company and moving off iMessage? We'll onboard you in under 24 hours (FAQ).

1. The request starts in a thread. A bug report, a failing check or a question, with the people who care about it.
2. The agent is asked there, by name, and posts what it finds back to the same thread: the logs, the failing test, the change it proposes.
3. A person reviews. The pull request goes through the review you already have; the agent does not merge or release on its own unless you have decided it should.
4. The thread is the record. The next teammate, or the next agent, reads what happened instead of asking.

Give agents write access only where you have decided they should have it, and start with read access to repositories and CI. Compare the agents themselves in Grok Bot vs Claude Code vs Codex in a shared workspace, or see how to bring your own agents over MCP.

• You want agents where your tools already post → Slack, with AI assistant apps.
• You must self-host and approve agent tools → Mattermost.
• You write your own bots → Zulip or Discord.
• You want open source with agents on your own machine → Buzz.
• Several coding agents already do real work, and people keep relaying context to them → Ando. Starting a new company and moving off iMessage? We'll onboard you in under 24 hours (FAQ).

Ready to try it with your team? Request access. Access is by request, and the same page lets you check on a request you've already made.