The Claude Code Feature I Ignored for Months Was the Most Important One
Source: Colby McHenry, I Never Used Claude Code Agents, Now I Can’t Work Without Them (Medium, 2026-04-07)
Key takeaways
- Agents mode is the Claude Code feature most users overlook — lets you spawn parallel sub-agents within a session.
- McHenry ignored it for months; now says he “can’t work without it.”
- Different from Agent Teams — those are separate Claude Code sessions with shared task list. Agents mode runs inside your session.
- Different from Anthropic Managed Agents — that’s a hosted orchestration service, API-only.
The three levels of “agents” in Claude Code — keep them straight
The community uses “agents” for at least three different things. Clarifying:
| Pattern | Runs where | Coordination |
|---|---|---|
| Agents mode (this article) | Inside your Claude Code session | Returns summaries to parent session |
| Agent Teams | Separate Claude Code sessions | Shared task list + peer messaging |
| Anthropic Managed Agents | Anthropic’s infrastructure | First-party orchestration layer |
Agents mode is the lightest-weight pattern — use it for fan-out searches, parallel exploration, independent sub-tasks that don’t need their own sessions.
When to use agents mode
- Fan-out searches across a codebase
- Parallel exploration of multiple hypotheses
- Independent sub-tasks that can run concurrently and don’t need shared context
- Quick “run these three things in parallel and come back to me” workflows
When to escalate to Agent Teams instead
- Truly parallel implementation across multiple files
- When each agent needs its own full context window
- When agents need to coordinate via shared task list / peer messaging
When to use Managed Agents instead
- Production agent pipelines with state-management requirements
- When you want Anthropic to handle infrastructure + permissions layer
- When context-continuity across handoffs matters for efficiency
Why this is worth reading
The delta between “didn’t use agents” and “can’t work without them” is evidence that a significant fraction of Claude Code users are under-using the tool simply because they haven’t tried the feature. If you haven’t used agents mode, spend 30 minutes with it.
Related Playbook pages
- Agent Teams — when to escalate from in-session agents to separate-session teams
- Anthropic Managed Agents launch — when to escalate further to the first-party service
- Multi-Model Orchestration — how agents fit into the broader orchestration picture