Claude Code creator uses "opponent" sub-agents to cross-check each other's work
every · x · 2026-10-02
Boris Cherny, creator of Claude Code, runs multiple agents in parallel for big migrations—one for compliance, one digging through git history, one hunting bugs—then spawns five more sub-agents to verify findings and cut false positives. Dan Shipper uses a similar "adversarial" pattern for expenses: one agent argues a case, another plays auditor until they converge. The trick works because sub-agents have isolated context windows, outperforming single-agent self-review.
More from coding & agent
- NVIDIA paper: a better judge lifts terminal agent success from 50% to 68% without retraining — rohanpaul_ai · 2026-10-02
- Neuro-Symbolic Computer Use: agents that turn execution experience into self-healing policies, claimed 99% cheaper — xwang_lk · 2026-10-02
- Stripe now pays gas fees for agent stablecoin payments over MPP — jeff_weinstein · 2026-10-02
- The Flag Game: a toy setting to study agent swarm dynamics and cooperation — Hidenori8Tanaka · 2026-10-02
- Coinbase Link ships API to prove agents act on behalf of verified users — jeff_weinstein · 2026-10-02
- Dev deploys AI agent to 'stand guard' over open-source repo and flag needed changes — IanArawjo · 2026-10-02