FoundationsEXPLAINERHANDS-ONWorking with Claude

Subagents, when a founder actually needs one

Jun 26, 2026 · 3 min

A subagent is Claude Code delegating part of a task to a separate, purpose-built instance of itself — a “researcher” pass, a “reviewer” pass — that works on its own piece and reports back, rather than one continuous conversation trying to do everything at once.

Most founder work doesn’t need this. A weekly review, an investor update, a cold-email draft — these are single, focused tasks that don’t benefit from being split up. Subagents earn their keep on genuinely heavier, multi-part work: research that touches many sources at once, or a task where one pass should sanity-check another’s output before you see it.

The honest founder-relevant guidance: know it exists, don’t reach for it by default. If a task feels like it needs several distinct kinds of work happening in parallel — not just several steps in sequence — that’s the signal to look into it. Otherwise, a plain, well-scoped request to Claude Code does the job with less to think about.

Check current capabilities in Claude Code for exactly how subagents are invoked, since the mechanics are the kind of detail that gets refined over time.

Get plays like this every Sunday