Subagent

A coding agent asked to fix a bug might start one subagent to search the codebase, another to read the failing test’s history, and a third to draft the patch. Each runs with its own context window and returns a short summary, so the parent agent never has to hold every file and log line itself. In Claude Code, for example, a subagent is defined by a name, a description of when to use it, a system prompt and a list of allowed tools.

The pattern is the smallest form of a multi-agent system: one orchestrator agent delegates, subagents execute. The benefits are context isolation, parallel work and tighter permissions per job. The costs are more tokens overall, latency from handoffs, and information lost in the summary a subagent passes back. If the parent needed a detail the subagent left out, the run goes wrong quietly.

In episode 76 at 28:03, Sonar CTO Andrea Malagodi says the benefit of a group of agents, or of one agent that spawns sub-agents, comes from handing them a clearly scoped problem, and that you need to be clear on the boundaries and on how you set up the instructions. Episode 76 credits Svix, Walrus Memory and Inngest as Season 4 presenting sponsors (Svix read at 14:55, Walrus Memory at 16:01, Inngest at 34:17) and carries a G2i episode sponsorship (read at 33:36).