Back to Catalog
A16z · EXTRACTED

How Jev Turns AI Into Software That Gets Things Done

7 plays for turning AI into software that actually gets things done. The unlock isn't a smarter chatbot, it's building intelligence into the guts of the system so reliably that a developer can program against it without checking its work.

▶65K views on YouTubePreview, 1 of 7 tactics free

With Diego

"Where the [ __ ] is all the automation? AI is so unbelievably smart and yet it's so useless at all other stuff." — Diego
Diego, on the episode

Diego is the founder of Typesafe, the company behind Jev, sitting down to explain why he thinks AI has been shipped in the night with software instead of actually inside it. The pop framing of AI tools is that they write code faster, a faster human replacement typing the same code a person would have typed anyway. The operating system Diego is actually running is different: Jev is a primitive you embed inside a program, a classifier with a state machine and confidence levels, so software itself gains new capabilities instead of just getting typed out quicker. He traces this back to a specific frustration at OpenAI, watching RLHF generalize beautifully while real automation, the boring stuff like customer service, stayed stuck. This conversation pulls out how he thinks about reliability, where coding agents are actually useful versus dangerous, and when automation is worth building at all.

Tactic 01

Diego, a16z

Design For Robust Intelligence Not Determinism

The play

Test AI-powered components for semantic stability across varied inputs, not exact output matching, so you can stop crutching on example queries.

Diego says

I don't aim for strict determinism inside AI-powered software, I aim for robustness, similar intelligence every time. The real test is whether a human in that exact situation would find the response understandable, that's what a developer can program around. The highest honor of reliability is when people can program against the system without writing example queries anymore.

Try this

Test the system for understandable, humanlike responses instead of exact-match determinism.

Tactic 02

Let Coding Agents Own Syntax, Keep Architecture Human

Diego says

Coding agents are good at syntax and bad at semantics, I'd say incredibly bad at architecture specifically…

Episode quotes and Try this

Unlock
Tactic 03

Run Every Automation Through The ROI Test First

Tactic 04

Use Natural Language State Machines As Code Primitives

Tactic 05

Integrate AI Into SaaS Core Logic Not Just Chatbots

Tactic 06

Target Boring, High-Volume Workflows For Automation

Tactic 07

Optimize For Intelligence Per Dollar, Not Per Second

“It's basically become my shortcut for podcasts I know I'm never going to finish.”
Oliver Thompson · Entrepreneur
NC
“I got Pro mainly because I wanted to use it with my own podcasts. The extra credits make sense if you're actually using it regularly.”
Noah Cohen · Early user
Subscribers only
Unlock the full summary
6 more tactics, the full action plan, and every new summary the day it drops.

$29.99/mo, cancel anytime.

A16z, extracted by Podex