Andrej Karpathy says mastering Claude takes about 10,000 hours. He ran a 14-minute workshop showing how he uses the model day to day. In that session, he walks through his top tips: crafting precise prompts, chaining calls for complex tasks, and integrating Claude’s API into Python scripts. He demonstrates real code snippets and highlights the key settings—temperature, max tokens—and when to adjust them for different jobs. On a separate 25-minute live stage demo, Amazon quants at Anthropic rebuilt an entire framework in three days that others had spent three years coding. They wrote just 44 lines of Claude code and then open-sourced it. The package bundles pre-trained models, helper tools, and prompts you can tweak. With these pieces, they show you can spin up a capable AI agent from scratch—no extra libraries, no proprietary backends—right in your own environment.