Do AI agent to-do lists fail because of models or harness design?
pseudobacon · reddit · 2026-09-20
- The author reports that across several agent harnesses, to-do lists get generated fine but the model fails at actually checking items off, which defeats the feature.
- They ask whether this is a model capability limit, a context-size issue, or a harness/plugin design problem — a real pain point for agent workflow builders.
More from coding & agent
- Founder trains AI on 10K malicious domains, builds link scanner in 2 hours — JosephJacks_ · 2026-09-21
- Telex: Open-Source Tool Uses LLM + Tree-sitter to Auto-Fix Dependency Breakages via PRs — Efficient-Passage889 · 2026-09-21
- Two Years of Vibe Coding: This Dev's Daily Stack Is Claude Code + Next.js + Postgres — West_Sound5224 · 2026-09-21
- Devin's native video capture of test runs shows its cloud-first edge — brandon_galang · 2026-09-21
- Swapping the Gaussian Process for an LLM in Bayesian optimization, benchmarked 4 ways — tak3sh8 · 2026-09-21
- Feeding task descriptions to JEV yields workflows across hundreds of apps in seconds — gaganghotra_ · 2026-09-21