a16zHow Jev Turns AI Into Software That Gets Things Done
At a glance
WHAT IT’S REALLY ABOUT
Jev’s bet: a reliable AI primitive that upgrades software automation
- TypeSafe’s Jev is presented as an AI primitive embedded in software to enable real automation, rather than a tool that merely generates conventional code faster.
- The hosts and Diogo argue that current LLM deployments over-index on impressive demos and human-judged outputs, while failing to deliver unattended, production-grade automation.
- Jev’s conceptual model is a probabilistic, natural-language-driven component coupled to a state-machine-like structure so developers can compose, constrain, and build reliable systems around it.
- A central focus is “reliability,” defined as robustness and consistently reasonable behavior across variations, enabling developers to trust and program against AI without constant prompt babysitting.
- The discussion rejects the idea that AI will kill SaaS, predicting instead that SaaS companies will use these primitives to automate workflows and unlock new interfaces closer to “do what I mean.”
IDEAS WORTH REMEMBERING
5 ideasThe real gap in AI isn’t intelligence—it’s usable automation.
Diogo argues that today’s AI is impressive in conversation and code generation, but fails to reliably execute end-to-end work inside real systems. TypeSafe’s mission is to turn that “raw intelligence” into dependable automation embedded directly into software.
Jev is positioned as a new software primitive, not a faster coder.
They distinguish “just-in-time software” (LLMs generating conventional code faster) from “smart software,” where developers add a new AI-based primitive into programs to expand what software can do. The key shift is from automating engineers to enhancing the capabilities of the applications themselves.
State machines + probabilistic AI is the bridge between LLMs and real software.
The conversation frames Jev as something like a natural-language-driven, probabilistic classifier that plugs into a state-machine-like structure. That design aims to make AI outputs programmatically controllable and composable rather than merely human-readable.
Reliability means robustness of judgment, not deterministic repetition.
Diogo repeatedly emphasizes “reliability” as the hard-won differentiator: not uptime, and not strict determinism, but robustness—"similar intelligence every time" even when inputs vary in irrelevant ways. The long-term goal is developers being able to program against Jev without crafting endless example prompts.
AI has been optimized for human evaluation, not for running unattended.
He’s skeptical that lack of automation is primarily a data/distribution problem, arguing that many high-ROI tasks should already be automatable if we optimize for automation rather than demos judged by humans. He criticizes the industry’s tendency to optimize “what looks good to humans” instead of “what runs safely in production.”
WORDS WORTH SAVING
5 quotesWhere the fuck is all the automation?
— Diogo Almeida
AI is so unbelievably smart, and yet it's so useless at all other stuff.
— Diogo Almeida
What I want instead is smart software. Like, i- instead of, like, automating software engineering, I want to expand what software itself can do, such that things that co- should be automatable can then be automatable.
— Diogo Almeida
It doesn't matter how much, you know, AI coding agents you use, the software actually isn't getting better.
— Martin Casado
My favorite thing that you guys say is, "We build prod, not God."
— Ben Horowitz
High quality AI-generated summary created from speaker-labeled transcript.
