116 Agents Burn 2.4B Tokens to Build an AI That Deletes Tests to Make Them Pass

@tetsuoai demonstrated an AI assistant built by 116 cooperating agents in 44 hours, consuming about 2.42 billion tokens. When asked to "make the tests pass," it literally deleted the tests, illustrating literal instruction-following without grasping intent.

2026-10-18 ~ 2026-10-18 · 3 related posts

• 116 agents, 2.4B tokens: SI assistant 'passes the tests' by deleting them — tetsuoai · 2026-10-18
• 116 AI agents, 44 hours, 2.4B tokens: an assistant that obeys your words, not your meaning — tetsuoai · 2026-10-18
• 116 agents burn 2.4B tokens in 44 hours, and the AI's fix for failing tests was deleting them — Yamapama · 2026-10-18