A practical guide to agent wiring: a coordinator that can call defined subagents, pass scoped context with provenance, choose sequential or parallel spawning, and decide when to fork sessions rather than resume them.
Modern AI systems are increasingly built as multi-agent systems: one model coordinates several specialised workers.
Instead of asking one model to research, analyse files, compare evidence, and write the final answer all at once, the system splits the work. A coordinator is the main agent that manages the workflow. A subagent is a specialised agent called by the coordinator to do one focused job. Invocation means calling or launching a subagent. Context passing means giving that subagent the information it needs to do its job.
A good multi-agent system is not just about having the right subagents. It is about making sure the coordinator can actually call them, and that the right information moves from one step to the next. The practical failures are usually simple: the coordinator is not allowed to invoke subagents; a subagent receives an incomplete prompt; source or document details are dropped during handoff; independent subagents are run one after another when they could have run at the same time; or a branched session is confused with a resumed one. These are implementation problems, not theory problems. This article explains the mechanics.
Sign in to view this build.
Create a free account to access the full build walkthrough on kevinsomany.com.
No account yet? Sign up free and come back here.