The good, the bad, and the future of AI agents

October 2, 2025·47 min·4 clips

Anthropic's AI built a working chat app with DMs and authentication in 30 hours without human help.

Hayden Fields interviews Anthropic's David Hershey about the new Claude Sonnet 4.5 model, focusing on its breakthrough capabilities in autonomous AI agents for coding. They discuss how the model can build complex applications like a Slack clone in 30 hours, surprising adoption in legal fields, current limitations, and implications for software engineering jobs.

As heard by us

A careful Anthropic conversation about AI agents, coding work, and why new model capabilities only prove themselves once people build with them.

Decoder treats AI agents as a product problem, not a magic trick. Guest host Hayden Fields talks with Anthropic's David Hershey about what a new model might change once people actually build with it, especially around software engineering work.

Read the full review in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Decoder with Nilay Patel