Agent Teams
A Claude Code multi-agent mode where agents share a task list, can communicate directly with each other, and can assign work to one another — an extension of sub-agents with peer-to-peer coordination.
Key points
- Sub-agents run in parallel with fresh context but can only report back to the main thread — no peer communication [src-011]
- Agent teams: all agents share a task list and can communicate peer-to-peer, not just top-down [src-011]
- Individual agents in a team can be addressed directly by the user, not just through the main orchestrator [src-011]
- More expensive and longer-running than sub-agents, but produce more cohesive outputs on large, interdependent projects [src-011]
- Best used for complex projects requiring cross-component coherence; sub-agents better for isolated parallel tasks [src-011]
- The 10-hour course teaches agent teams as part of the advanced Claude Code layer after skills and sub-agents, alongside browser automation, permissions, context management, and worktrees [src-016]
- Boris Cherny describes massive parallelism as an active frontier: his own setup often has hundreds of agents and overnight thousands of agents, while the product team is working on easier multi-agent delegation, batch, loop, and team patterns [src-054]
- As models improve, Boris expects the model to infer more of when to parallelize work instead of requiring users to manually decide where to spawn sub-agents [src-054]
- Cursor's team-era framing converges on the same pattern from an IDE/product angle: agents get their own remote computer environments, run for hours or days, and let humans supervise many parallel implementation streams [src-080]
- Cursor's autonomous-browser experiment is an extreme version: an agent team worked over roughly a week, generated millions of lines, and produced a mostly functional prototype browser, though Cursor emphasizes this remains experimental [src-080]
When to use
| Scenario | Use |
|---|---|
| Independent parallel tasks (e.g., 5 separate summaries) | Sub-agents |
| Complex projects with interdependent components (e.g., full-stack app with API + frontend + tests) | Agent teams |
| Single isolated tasks | Main thread |
Related entities
- Claude Code — the runtime
- Cursor — product framing around the broader team era
Related concepts
- Agent Orchestration — agent teams extend the orchestration model
- Claude Code Decomposition Pattern — how to break a project into team assignments
- Claude Managed Agents — cloud SDK counterpart for orchestrating agents outside Claude Code
- Claude Code Loop Skill — recurring loop primitive that often works alongside parallel agents
- AI-Native Organizational Process — organizational layer where many agents coordinate through shared tools and communication channels
- Coding Agent Team Era
Source references
- [src-011] Nate Herk — Claude Code power features cluster (2026-04-20 to 2026-04-27)
- [src-016] Nate Herk — "Build & Sell with Claude Code (10+ Hour Course)" (2026-03-12)
- [src-054] Sequoia Capital — "Anthropic's Boris Cherny: Why Coding Is Solved, and What Comes Next" (2026-05-04)
- [src-080] Cursor — "The next era of AI coding" (2026-05-12)