a16z Podcast: Diogo Almeida on Turning AI Into Real Software Automation

Transcripted.ai (0:01)

So what happens when AI is smart, but still not actually useful enough to change the way software gets work done? That's the tension at the heart of this episode with Diogo Almeida, founder of TypeSafe, and their product Jev.

Right, and Diogo opens with this sharp observation: there's far more intelligence in the world than there is automation. He argues the industry has spent too much time polishing demos and too little time building systems that reliably do real work in the background. That's where his phrase "we build prod, not god" comes in. The point isn't to build some all-knowing general intelligence. It's about creating practical tools that make software more capable and easier to trust. Exactly. And Jev isn't just another coding assistant. Diogo says tools like Copilot help write code faster, but they mostly preserve the same expressive limits software has always had. Jev aims to expand the vocabulary of software itself. How does that work in practice? Think of software as a library powered by natural language, confidence levels, and state machines that choose actions. The goal isn't just generating code—it's creating smart software that can do what humans actually mean to automate. Interesting. Diogo also has an unusual background, right? He was an award-winning mathlete but never loved math for its own sake. Computer science felt useful and real to him, which pulled him toward OpenAI and eventually building TypeSafe.

And he's clearly frustrated with the mismatch between AI's potential and what's actually shipped. He describes the industry as stuck in "gigantic over-promising and under-delivering." Teams need to optimize for automation, not human applause.

That's why his North Star is "intelligence per dollar." Instead of chasing flashy demos, he wants pragmatic, reliable systems capable of handling real automatable work. Reliability is everything here. Each new level of reliability unlocks new use cases. Diogo compares Jev to putting an "electric motor of intelligence" into people's workstations.

He also draws this line between syntax and semantics. Coding agents handle syntax well, but architecture remains deeply human, creative work. The future probably looks like models handling code mechanics while people shape the bigger design.

The support automation example really drives this home. Companies claim high automation rates, but the data often shows only simple cases like password resets. That long tail of exceptions is exactly why robust systems matter.

And rather than killing SaaS, Diogo thinks AI may supercharge it. SaaS companies already know user workflows, so they're best positioned to turn AI into a "do what I mean" layer. The bigger vision is software that's trusted, state-aware, and deeply practical. Diogo hopes these systems become an "anti-frustration" machine, helping people enter flow instead of fighting their tools. That's the future worth building.