Claude Code says agents write 90% of its code — Yoav Goldman offers a skeptical take
PMinervini · x · 2026-11-18
The Claude Code team claims 90% of its code is written and verified by agents without human review, sharing tips on verification via tests and end-to-end runs. Yoav Goldman pushes back: the real enabler isn't superior auto-testing but that Claude Code is a very simple piece of software — minimal efficiency concerns, no competitors, low risk of harmful bugs or data corruption — with many bugs quietly absorbed by the underlying LLM being exceptionally good.
More from coding & agent
- Claude delegates synthetic data generation to local Qwen3.8-27B, citing its own guardrails — maksym_andr · 2026-11-18
- GitHub Copilot app adds Azure DevOps repos and browser automation in public preview — DanWahlin · 2026-11-18
- GPT-6.2 stuck in compaction loop in Codex, burning 4.44M tokens in 9 minutes — Dear_Lock_5280 · 2026-11-18
- Claude Code's context compaction silently turns offhand remarks into agent-constraining rules — nptacek · 2026-11-18
- Shipyard MCP server automates domains, DNS, DB and Stripe setup so AI-built apps launch in an hour — OnlyBullish1 · 2026-11-18
- Infinitty Previews Drag-and-Drop File Transfers and Auto Secure Tunnels for Remote Hosts — jasonkneen · 2026-11-18