Summary
In this a16z interview, Typesafe founder Diego discusses Jev, the company’s new model for making AI-generated software actually take action rather than just describe it. He argues that even as coding agents like Claude Code and Codex have made writing software faster, the resulting code is functionally the same as what a human would write, and doesn’t expand what software itself is capable of automating.
Diego explains Jev as a new programmable primitive developers can embed inside their code, distinct from a chatbot or coding assistant, designed to make previously unautomatable business logic and decisions automatable at scale. He traces his thinking back to early OpenAI research on RLHF generalization around 2021, and explains why he’s skeptical of recursive self-improvement and SaaS-apocalypse narratives, favoring incremental expansion of what software can do over full automation of software engineers.
The conversation covers Jev’s fast adoption among developers, comparisons to “just in time” software from tools like Claude Code and Codex, and Typesafe’s broader philosophy of building practical, deployable AI systems rather than chasing general intelligence claims. It’s a candid, technical look at how one of the more talked-about new AI tools of the moment thinks about the gap between AI capability and real-world automation.
📺 Source: a16z · Published September 28, 2026
🏷️ Format: Interview
