Dev uses Gemini Neon to deslop a UI bloated by 271+ agent-inherited disclaimers
HankYeomans · x · 2026-11-20
Testing Gemini Neon, the author discovered 270+ defensive strings in app.hosted.js — guardrails his agent had picked up from earlier iterations. He had the model collapse repeated legalistic disclaimers ("observation is not approval", etc.) into compact badges with on-demand popovers and update the corresponding UI copy tests, "deslopping" the UI in the process.
- Key insight: long-running agents accumulate stale constraints as defensive code/copy over iterations
- Tactic: prompt the model to identify redundant disclaimers and fold them into a single reusable UI component
More from coding & agent
- CapCut plugin connects ChatGPT to auto-cut gameplay highlights, then human picks the payoff — iamfakhrealam · 2026-11-21
- Dev wires a Runway-generated avatar into Nous Research's Hermes Agent, bashing boring agent UIs — alexcovo_eth · 2026-11-21
- Lean compiles, but the natural-language proof is still wrong — jasondeanlee · 2026-11-21
- DeepMind demos ski simulator built by Gemini 4 Neon from real Mont Blanc terrain — GoogleDeepMind · 2026-11-21
- pstack v0.15.46 trims skills, adds Gemini as panel reviewer; pinning question raised — petrusenko_max · 2026-11-21
- Hamel Husain on evals: a good process beats good tools — HamelHusain · 2026-11-21