Writing on Bertram — Judgment & Buildinghttps://bing-hsu.github.io/posts/Recent content in Writing on Bertram — Judgment & BuildingHugoen-USFri, 30 Oct 2026 09:00:00 +0800Why Your Agent on a 4B Local Model Quits on Turn 1https://bing-hsu.github.io/posts/ai-agents/why-your-4b-agent-quits-on-turn-1/Fri, 30 Oct 2026 09:00:00 +0800https://bing-hsu.github.io/posts/ai-agents/why-your-4b-agent-quits-on-turn-1/<!-- span color classes (inserted by :SpanStyles) --> <style> .red { color: #e74c3c; } .orange { color: #e67e22; } .yellow { color: #b58900; } .green { color: #27ae60; } .blue { color: #3498db; } .purple { color: #8e44ad; } .dim { color: #888888; } .highlight { background-color: #fff3b0; color: #7a5c00; } </style> <p>You gave your agent a search tool, a reasoning budget, and a strict Pydantic response format. Why did it stop on Step 1 with a nonsense like &ldquo;I need to perform the calculation after gathering data&rdquo; wrapped in a valid structured output?</p>