Python for AI: The Only Libraries You Actually Need

0:00 The library problem 0:43 Talking to a model (the provider SDKs) 1:20 Structured output: pydantic + instructor 1:51 Orchestration and agents 2:44 The data and math floor 3:14 Tokens and embeddings 3:40 Vector databases 4:19 Feeding it your documents 4:46 Shipping it as a service 5:14 Async and resilience 5:41 Cloud and serverless 6:09 Running models locally 6:38 Observability and evals 7:06 The workflow tools 7:36 Starter kits by role 8:20 The takeaway