Pydantic Ai Agent Creation

This walks you through spinning up PydanticAI agents with proper type safety and structured outputs. You get model selection across OpenAI, Anthropic, Google, and others using a simple provider:model-name format, plus Pydantic models for validated responses instead of raw strings. The dependency injection pattern is clean (deps_type lets you thread API keys or database connections through without globals), and the three verification gates at the end are genuinely useful for catching issues before production. If you're building anything beyond one-off LLM calls, the structured output alone justifies the setup cost. The instructions versus system prompts distinction trips people up initially, but the examples make it clear.