Same Task, 10 Claude Code Agents at Once: 50 Minutes and About $32 to See Where the AI Is Guessing
- Level Advanced
- Setup1–3 hrs (estimated)
- Evidence Some evidence
- CostNot stated in source
- No-codeNo
What they did
A developer wanted to know what a coding agent would do if he asked it the same thing ten times. He built multirun, which runs one task through ten isolated Claude Code agents in parallel, each with its own git worktree and its own Postgres database branch, then compares what each one did. All ten agreed completely on the schema but split exactly 5 to 5 on two separate decisions, which showed him the sentences his task description was missing. The run took about 50 minutes and roughly $32 at API rates. He now uses it as a simple eval and before big decisions like a schema or auth design.
How it went
dev.to write-up (Oct 24, 2026) with open-source repo a2wio/multirun: 10 parallel isolated Claude Code runs (git worktree plus Postgres branch each), full agreement on schema, two 5/5 splits that exposed missing instructions; about 50 minutes wall time and about $32 at API rates
Worth knowing
- dev.to write-up (Oct 24, 2026) with open-source repo a2wio/multirun: 10 parallel isolated Claude Code runs (git worktree plus Postgres branch each), full agreement on schema, two 5/5 splits that exposed missing instructions; about 50 minutes wall time and about $32 at API rates.
Prompt
Cleaned version rewritten from the source — not a direct quote.
Before I commit to [design decision, e.g. the database schema], run this same task [10] times in separate, isolated copies of the repo, each with its own branch of the database. Don't let the runs see each other. When they finish, compare them: list where every run agreed, where they split and how (for example 5 to 5), and for each split, the instruction my task file was missing that would have settled it.
Recommended
Try it in Grok Bot
- Copy the prompt (button below).
- Open Grok Bot chat and paste it. Replace any [bracketed] placeholders.
- Connect the listed tools (as needed) as connectors when Grok Bot asks.
- If the task repeats, ask: "Make this a daily routine."
Alternatives: the same prompt works in Claude Code, Claude, ChatGPT Agent, OpenClaw.
Set it up
Stack: Claude Code
- Read the original source for full context.
- Connect the tools your agent needs.
- Paste the prompt and replace [bracketed] placeholders.
- Run once manually and check the output before scheduling it.
Generic suggested steps, not from the source.
Evidence
dev.to write-up (Oct 24, 2026) with open-source repo a2wio/multirun: 10 parallel isolated Claude Code runs (git worktree plus Postgres branch each), full agreement on schema, two 5/5 splits that exposed missing instructions; about 50 minutes wall time and about $32 at API rates
Source: Web · Verified by us: not yet · Report broken or outdated