Pydantic Ai

MCPs and questionable vibes

How much I can push an LLM to actually be useful in an agentic workflow? I ran the same task with three different models: `Gemini`, `Qwen3-coder-next` and `Qwen2.5-coder` and there was some good, some