When to fan out

Build-along · ~8 min at your own keyboard

Ten subagents at 95% each: if they fail independently, 0.95^10 ≈ 0.60 — about a 60% chance all ten succeed on the first try. That's why fan-out needs a way to spot and re-run failed chunks. Fan out when the task splits into simple, independent chunks; stay single-threaded when steps depend on each other or need judgment. Setup time is real too — small jobs run faster without it.

The long version: Subagents: when they help ↗

Do this

0/3

Checkpoints

▸You should now see four tasks, each marked independent chunks or dependent chain.
▸You should now see a per-chunk reliability guess next to each chunky task.
▸You should now see one best and one worst fan-out candidate with a reason each.

The three ways this goes wrong

Every task looks like a fan-out candidate. — Ask whether chunk two needs chunk one's result. If yes, it is a chain.
You skip the reliability guess. — Ask: could a junior get this right 19 times in 20? If not, do not fan out.
You pick a small job as the best candidate. — If the whole job takes a few minutes alone, the setup costs more than it saves.

Share what happened

Paste your four-task analysis with the best and worst fan-out candidates reasoned.

Strip API keys, passwords, and tokens before you paste. Swap them for [KEY].

🔒 Don't paste confidential, personal-health, or payment info.