Story · Anirban Chatterjee (Sonar)
Guide, Verify, Solve (Anirban Chatterjee (Sonar))
talk · Story page
Chatterjee argues AI-written code leaves verification debt that human review alone can't reliably contain: a Wharton study he cites suggests reviewers followed AI advice nearly 80% of the time even when it was instructed to lie confidently. His fix is zero-trust, multilayer verification that checks generated code by methods independent of the model that wrote it.
In plain words
- Anirban Chatterjee argued that human review cannot reliably catch problems in computer code written by artificial intelligence.
- A cited study found productivity gains ended after about three months, while warnings and added complexity remained.
- Another cited study found reviewers followed false, confidently delivered advice nearly 80% of the time.
- He recommends checking generated code through several independent methods, including automated checks and human reasoning.
- This matters most for large systems exposed to hostile users, where missed problems carry higher costs.
Appeared in
- A public harness reproduces DeepSeek's 82.7% on Terminal-Bench, 445 trials deep
Aug 10, 2026 · in the sections
Subscribe
Get the brief in your inbox
Pick daily, weekly, or both. Nothing is gated either way: every issue is on the site and in the feeds.
- Weekdays at 8:45am IST, one lead story and 6 to 9 items.
- Sundays, an argued synthesis rather than a recap.
- One click to leave, and quiet days say so in the subject line.