AI Agent Calling the Wrong Tool? I Cut It from 6/20 to 1/20
My agent picked the wrong tool in 6 of 20 runs. Cutting tools didn’t help; overlapping descriptions were the cause. Adding ‘when not to use’ cut it to 1.
AI
Claude Code, Codex and Antigravity break in specific ways when you use them on real projects. I’m a backend developer who runs them at work and on my own code, plus local models on an RTX 4090, and I write down each error and the setting that fixed it.
vLLM vs llama.cpp on a Single RTX 4090: Why I Went Back
On one RTX 4090 with one user, vLLM and llama.cpp generate at about the same rate; vLLM pulls ahead with many concurrent users. My trial and the numbers.
Docker Model Runner vs Ollama: Should You Switch? (RTX 4090)
Docker Model Runner vs Ollama on an RTX 4090: 70–90 tok/s on 7–8B models, a 40–60 minute first DMR setup, and why my agent’s JSON parsing broke.
Codex CLI npm install Fails? It’s Sandbox Network Access
Codex CLI edits files but npm install stalls? workspace-write keeps network access off by default. The config fix, why I skip sudo, and my 5-run test.
Codex “Ran Out of Room in the Model’s Context Window” Fix
Codex ran out of room in the model’s context window? Run /compact once and re-pin 5 facts. If it fails again, start /new. Plus when to resume or fork.