How many agents can work in parallel with Claude Code?

Claude Code can run up to 7 subagents simultaneously within a single session. For agent teams, the number of teammates is flexible—Anthropic demonstrated 16 parallel agents building a C compiler, though the optimal number depends on the task complexity and coordination overhead.

Claude Code Now Manages Parallel Agent Teams: 16 AI Agents Built a C Compiler

Claude Code agent teams allow multiple AI instances to work in parallel on a shared codebase. Released with Opus 4.6 on February 5, 2026, this feature enables a team lead to coordinate work across independent teammates. Anthropic demonstrated the capability by having 16 agents produce a 100,000-line C compiler that compiles the Linux kernel—at a cost of $20,000 in API usage over two weeks.

What Are Agent Teams?

"Instead of one agent working through tasks sequentially, you can split the work across multiple agents—each owning its piece and coordinating directly with the others," Anthropic explains. This isn't just parallel processing; it's collaborative AI development.

Agent teams differ from the existing subagent system. Subagents run within a single session and only report back to the main agent. Agent teams, by contrast, operate across separate sessions. Teammates can message each other, share task lists, and self-coordinate—without requiring a human to relay information.

16 Parallel Agents

100K Lines of Code

$20K API Cost

The C Compiler Demonstration

To prove the concept, an Anthropic engineer tasked 16 parallel agents with writing a Rust-based C compiler from scratch—one capable of compiling the Linux kernel. The results exceeded expectations:

100,000 lines of working compiler code
99% pass rate on GCC torture test suite
Compiles Linux 6.9 on x86, ARM, and RISC-V
Builds major projects including QEMU, FFmpeg, SQLite, PostgreSQL, and Redis
~2,000 sessions over two weeks

Source

"With agent teams, multiple Claude instances work in parallel on a shared codebase without active human intervention. This approach dramatically expands the scope of what's achievable with LLM agents."

— Anthropic Engineering Blog

How Agent Teams Work

The architecture consists of four components: a team lead (the main Claude session), teammates (independent Claude instances), a shared task list, and a mailbox system for inter-agent communication.

Best use cases identified by Anthropic include:

Research and review — Multiple agents investigate different aspects simultaneously
New modules or features — Each teammate owns a separate piece
Debugging with competing hypotheses — Agents test different theories in parallel
Cross-layer coordination — Changes spanning frontend, backend, and tests

How to Enable Agent Teams

Agent teams are experimental and disabled by default. Enable them by setting the environment variable:

CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1

Or add it to your settings.json. Once enabled, describe the team structure in natural language: "Create an agent team with three reviewers—one for security, one for performance, one for test coverage."

Subagents vs. Agent Teams

Claude Code also supports subagents—specialized AI assistants that handle specific tasks within a single session. The Task tool can run up to 7 subagents simultaneously. Built-in types include Explore (fast codebase search), Plan (research for planning mode), and general-purpose (complex multi-step operations).

The key distinction: subagents report results back to the main agent only. Agent teams can message each other directly, enabling true collaboration. Use subagents for focused tasks; use agent teams when workers need to discuss, challenge, and coordinate.

FAQ

What are Claude Code agent teams?

Agent teams are a feature in Claude Code where multiple Claude instances work together on a shared codebase. One session acts as team lead, coordinating work and assigning tasks, while teammates work independently in their own context windows and can communicate directly with each other.

How many agents can work in parallel?

Claude Code can run up to 7 subagents simultaneously within a single session. For agent teams, the number is flexible—Anthropic demonstrated 16 parallel agents, though optimal team size depends on task complexity and coordination overhead.

What's the difference between subagents and agent teams?

Subagents run within a single session and only report back to the main agent. Agent teams coordinate across separate sessions—teammates can message each other directly, share task lists, and work on a shared codebase without active human intervention.