⚡️Claude Sonnet 4.5 and Anthropic's roadmap for Agents and Developers — Mike Krieger, Anthropic

Mike Krieger, CPO at Anthropic, details the development and launch of Claude 3.5 Sonnet, highlighting a shift where product teams now actively influence research directions through real-world user feedback. This symbiosis ensures that models evolve beyond raw performance to address specific user needs, such as reducing "laziness" in coding tasks and improving UI design aesthetics. The conversation emphasizes the transition toward agentic workflows, where models manage long-horizon tasks like complex code refactoring or document creation. By evolving the Claude Code SDK into a general-purpose Agent SDK, Anthropic aims to provide composable building blocks that allow developers to integrate sophisticated planning and execution capabilities into their own products. Ultimately, the focus remains on building trust through interactive planning, enabling users to delegate complex, multi-step knowledge work to AI agents while maintaining necessary oversight.