← agents-deck

agents-deck vs Claude Code Agent Teams

An honest comparison, row by row, following the comparison table on our landing page. As of August 2026. Corrections welcome.

Agent Teams is Anthropic’s native multi-agent mode for Claude Code, shipped in February 2026 and off by default: a lead session spawns teammates with isolated contexts, coordinated through a shared task list, mailbox-style messaging, and file locking.

Let’s be honest about what that is: the coordination primitive, built into the tool itself, at no extra cost. It is the single development in this category we watch most closely. If you live entirely inside Claude Code and run a handful of agents, it works, and there is nothing to install.

But Agent Teams is also, by its own documentation, in-process, single-account, Claude-only, and ephemeral: limited first-class observability, no session resumption, no cross-session memory, and less mature team handoff (documented limitations). Which is why we do not really see it as the thing agents-deck competes with. Agent Teams is the coordination layer inside one tool; agents-deck is the persistence, observability, and governance layer on top, across tools.

Live observability dashboard

Agent Teams: no. Coordination happens inside the session, with limited first-class observability. There is no deck to glance at.

agents-deck: yes. A live, self-hosted dashboard across agents, tools, and projects, with awaiting-input alerts. Watching the work is the product.

Cross-tool (Claude · Codex · Cursor…)

Agent Teams: no. Claude-only, on a single account. That is by design; a vendor builds for its own tool.

agents-deck: yes. Claude Code, Codex, Cursor, Kimi, across repos. Cross-tool is exactly what a single-vendor feature will never prioritize.

Human approval, not --yolo

Agent Teams: partial. The workflow is one developer supervising their teammates in-session, and it works best at two to five agents. Supervision exists, but it is attention-bound: you are the dashboard.

agents-deck: yes. Structured approvals and awaiting-input alerts, so supervision scales past what one person can hold in their head.

Process guardrails / governance

Agent Teams: no. Task list, messaging, and file locking coordinate the work; nothing enforces process.

agents-deck: yes. Validators enforce skills with evidence, message banners, status-flip read confirmations, and per-agent file jurisdiction. Almost no one does this, vendors included.

Git-native audit · zero database

Agent Teams: no. Coordination state is in-process and ephemeral; when the session ends, the record of who did what does not live in your repo.

agents-deck: yes. State is YAML in the repo; the audit trail is git log. Zero database. The repo is the storage. Built for the EU AI Act era: git-native audit, on-process by default.

Cross-session memory & handoff

Agent Teams: no. No session resumption, no cross-session memory, less mature team handoff, per its own documentation.

agents-deck: yes. Tickets, messages, changelog, and roadmap persist in the repo. Tomorrow’s session, or a different tool entirely, resumes with full context.

Self-hosted

Agent Teams: no. It runs in-process inside Claude Code; there is nothing to host, and nothing you can serve to a team.

agents-deck: yes. Self-hosted, no account, zero database.

Price

Agent Teams is bundled with Claude Code, so effectively free if you already pay for Claude. The agents-deck core is free and open source regardless of which agents you run.

Use them together

This is the comparison page where the honest answer is “both”. Enable Agent Teams for in-session coordination when you are deep in Claude Code, and keep agents-deck as the mission-control layer on top: the persistent tickets, the live deck, the guardrails, and the git log audit trail that survive after every session ends, and that keep working the day you add Codex or Cursor to the mix.

Choose Agent Teams alone if

• You work exclusively in Claude Code, on one account.
• Two to five agents, supervised live, covers your workflow.
• You want zero setup and no extra moving parts.

Choose agents-deck if

• You run more than one tool, or more than one project.
• You need the record of what agents did to outlive the session.
• You want observability, approvals, and process governance, not just coordination.

Also comparing?

• agents-deck vs Conductor
• agents-deck vs Vibe Kanban

Sources for the Agent Teams side: the official documentation, re-checked August 2026, and a detailed guide to its limitations, from May 2026. If we got something wrong, tell us and we will fix it.