Single vs Multi-Agent
When does one agent become several? Multi-agent orchestration is powerful, but every extra agent adds coordination cost. Reach for it only when a task truly splits into independent parts.
The Patterns
Most designs are one of three shapes — from simplest to most involved.
- Single agent — one loop does it all, start to finish
- Orchestrator + workers — a lead agent delegates sub-tasks and gathers the results
- Specialists — each agent is focused on one job it does well
Why Split the Work
Splitting a job across agents buys you three things: each agent gets a smaller, focused context, independent parts can make parallel progress, and each agent can specialize.
The Cost of Splitting
Every agent you add has to be paid for — and the bill is real.
- Coordinating the agents — who runs when, and passing context between them
- Merging their outputs into one coherent result
- More to review and debug when something goes wrong
A single good agent often beats a fragile committee.
Default to One
Start with a single agent. Reach for multi-agent only when the sub-tasks are truly independent, the context won’t fit in one agent, or specialists clearly help.
Build It
How to implement: ask of your feature — does it split into independent sub-tasks? If not, keep it one agent. If yes, name the sub-agents and decide who merges their results.
- Weekly AI Tasks tracker — a single agent handles a message end-to-end; multi-agent here is overkill.
- Personal brand site — you could run one agent per section in parallel (from Module 3 L9), then a reviewer agent checks the facts.
What you learned
Three shapes — single, orchestrator + workers, specialists. Splitting buys focus, parallelism and specialization, but costs coordination, merging and debugging. Default to one; split only when the task really does.