Dev: unattended agent runs burn 99% of tokens on useless reproducibility gates
generativist · x · 2026-11-15
generativist reports that many unattended agent runs spend 99% of tokens and wall-clock time pursuing reproducibility gates he never asked for and that aren't useful — and prompting often fails to steer the agent away from them.
More from coding & agent
- Prompt, context, harness, loop engineering are all just "knowing what you want" — mr_samosaman · 2026-11-15
- Proof of Agent goes viral: paste a prompt and let your AI agent mine POA — adamamcbride · 2026-11-15
- Google and Northwestern's Harness² self-improves agent harnesses without answer keys — rohanpaul_ai · 2026-11-15
- Dev builds a PowerShell AI harness for Microsoft's new Intelligent Terminal — dfinke · 2026-11-15
- snitch: a harness that watches agents in parallel and reverts bad paths — generativist · 2026-11-15
- workspace-cli: open-source Rust CLI for AI agents to drive Google Workspace — majidmanzarpour · 2026-11-15