Where the term comes from In this channel's material, sub agents appear as part of the vocabulary around compound engineering — the creator describes building the method out of the prompts, flows, and sub agents he explored while creating Kora, an email assistant built with AI. In that context, sub agents are the delegation layer of the workflow: chunks of work get handed off so the main session keeps moving. How sub agents show up in practice - Delegation: the main session stays responsive while forked background agents take on pieces of work in parallel. - Agent-heavy workloads: in a live model comparison, the testers evaluated how models handled sub-agent-heavy work, noting that one model felt much slower there — mostly because it did more per sub agent. - Compounding work: within compound engineering, sub agents are one of the mechanisms that let every piece of work make the next one easier, since delegated chunks feed results back into the system. Why it matters Understanding sub agents helps you read the channel's agent-workflow material: when hosts talk about a model or method being good (or slow) at sub-agent-heavy tasks, they mean exactly this pattern — many forked background agents handling chunks of work while a main session coordinates. - Sub agents are forked background agents that handle chunks of work so the main session stays responsive. - They were part of the origin of compound engineering: prompts, flows, and sub agents explored while building Kora. - The channel evaluates models partly on how well they handle sub-agent-heavy work. - They enable delegation and parallelism without blocking the main session.