50 Things the Anthropic API Can't Do; Full Breakdown Part 1

Same disclaimer as the intro: Claude helped write this series. I gave it access to our docs at docs.backboard.io, it navigated there itself, read everything, and wrote about its own API's limitations. An AI describing what the AI company that built it can't do. No pushback. No hesitation. Make of that what you will. Anthropic's API is stateless. Every call starts from scratch. No memory of who you are, what you said last time, what your users care about. You pass it a transcript, it responds, it forgets you. That's fine for a demo. It's a disaster for a real product. To simulate a conversation, you have to send the entire message history on every single call: This is what actually happens every time you "chat" with the raw API response = anthropic.messages.create( model="claude-opus-4-6", messages=[ {"role": "user", "content": "My name is Sarah."}, {"role": "assistant", "content": "Nice to meet you, Sarah!"}, {"role": "user", "content": "What's my name?"}, ] ) Works. But only because y