Does restructuring context beat adding more of it? An untested agent hypothesis
ClickOk5811 · reddit · 2026-10-22
The poster proposes a hypothesis to be stress-tested: a code-modification agent follows project conventions on simple tasks but ignores them on multi-file tasks — not because information is missing, but because of how the prompt is laid out (conventions buried between file dumps, decision notes written as prose, no signal distinguishing constraints from background).
Two interventions are compared: A) same task with 20% more context (extra examples and rules); B) same token count but reorganized — constraints separated from reference material, conventions placed right before the task, background labeled as background. The expectation: B wins on multi-file tasks while A barely moves anything.
The only supporting evidence is lost-in-the-middle (Liu et al., 2023), which concerns retrieval rather than instruction following. The author flags possible confounds: B might win only by shortening the prompt, gains may not transfer across models, and small samples could be noise — inviting anyone who has measured structure vs. quantity in a controlled way to share results.
More from coding & agent
- Classic 'too many tabs' meme applied to coding agents: send help — mrdrozdov · 2026-10-22
- YC F26 startup Latent reads model internals to catch hallucinations at 530x lower cost — ycombinator · 2026-10-22
- We need an L0.5 for continual learning: agent state stacks up in 4 layers — willcb · 2026-10-22
- DuckDB + Hugging Face Amsterdam workshop: build an AI feedback pipeline over 2.1M Amazon reviews — lhoestq · 2026-10-22
- Veracode report: AI now writes half of committed code but only 56% pass security checks — WeldPond · 2026-10-22
- Will Brown: Continual Learning in the Harness Is Like a Computer Without RAM — willcb · 2026-10-22