The Changelog: Software Development, Open Source

AI agents can write the code, choose the packages, and run the install. Who's checking what they just invited onto your machine? This week I'm joined by Ahmad Nassri, CTO of Socket and former CTO of npm. We get into the trust built into a package install, the new ways attackers are using agents, and what it takes to keep building without handing over the keys or your wallet.

Dave Kerr created hacker laws, and now he finds himself reviewing agent-written code for them. Back on the show after six years, Dave names what he sees limiting teams today: keeping everyone’s mental model in sync with what’s actually been built. Adam declares it Kerr’s Law on the spot. We get into why Dave’s team treats code review as a deliberate slowdown, how Shellwright lets agents bring evidence to a pull request, and what living with bipolar disorder has taught Dave about noticing when the excitement of these tools stops being healthy.

Craig Kerstiens returns to talk about the ever-expanding world of Postgres: a deliberately boring core, a thriving extension ecosystem, and pg_lake bringing DuckDB and Iceberg into the mix. We dig into the pain of ETL, what Craig is watching in Postgres 19, the impact of AI on open source maintainers, and why the database still has plenty of room to grow.

This week I'm talking with Zeno Rocha, CEO of Resend, about scaling "boring" email to 3 million users, the secret Go rewrite behind it, and building for agents. We cover pricing.md, MCP, spam gangs, and why Resend keeps what's next for email.

Gavriel Cohen, creator of NanoClaw and co-founder of NanoCo, joins Adam to explain how a 40-hour weekend project became a viral open source hit and the foundation of an enterprise AI company. Gavriel describes the security concerns that led him to build a small, container-isolated alternative to OpenClaw, why he released it under the MIT license, and why the project's community and credibility are more valuable than keeping its code private.

They explore the shift from writing code to architecting the environments in which agents work, NanoClaw's "skills over features" approach, the approval and governance layers required inside large companies, and a future where every worker manages agents without surrendering human judgment. They close with the technical simplicity behind NanoClaw, containers, TypeScript, SQLite, and a two-database design, and Adam's case for durable Agent Flow artifacts that keep organizational context from disappearing into chat.