Why Jev Is Changing How We Build With AI with Diogo Almeida
October 6, 2026
AI Summary
5 min readJev's Bet on Machine-Native Intelligence
Diogo Almeida, co-founder and CEO of TypeSafe and former OpenAI researcher who worked on InstructGPT and RLHF, starts with a blunt observation: "AI is just so unbelievably smart, yet so unbelievably useless at the kinds of things you'd really expect it to be useful for." His company's model Jev, which came out of stealth three weeks before this conversation, aims to close that gap—not by making AI faster or cheaper, but by changing what it optimizes for. The core argument is that today's language models are optimized to produce strings for human consumption, while the real automation opportunities require models optimized to make reliable decisions that software can act on.
The Optimization Gap
Almeida's explanation for why AI can solve millennium prize problems in math but cannot automate basic accounting or customer service comes down to a single principle: "You get what you optimize for." All the major language models have been optimized for generating fluent, human-pleasing text. That optimization produces impressive chat experiences and coding assistants, but it actively works against the kind of calibrated, repeatable decision-making that software systems need.
Continue reading the full summary in the app — free to try.
Read Full Summary →Free • No credit card required
Never miss an episode of The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
Get every new episode summarized in your inbox — free, ~5 minutes to read.
No spam. Unsubscribe anytime.
What you'll learn
- 1 (00:58) **The Core Thesis: Where Is All the Automation?** - Diogo Almeida introduces the central problem: AI is superhuman at some tasks yet useless for straightforward business automation.
- 2 (02:40) **Root Cause: You Get What You Optimize For** - Diogo explains the fundamental mismatch between optimizing for human-readable strings versus machine-consumable decisions.
- 3 (05:33) **Testing the "Jagged Edge" Hypothesis** - Diogo demonstrates that reliability is not intrinsic to AI models but is a product of what you optimize for.
- 4 (08:08) **The Bitterest Lesson: Data and Task Matter More Than Compute** - Diogo reframes Sutton's Bitter Lesson, arguing data and the right task are more critical than scaling compute.
- 5 (11:15) **The RLHF Origin Story: Creating a Task From Nothing** - Diogo explains how RLHF created the entirely new task of instruction following from scratch.
- 6 (12:49) **The Obsession That Wouldn't Die** - Diogo traces his journey from OpenAI to TypeSafe, driven by a single unanswered question about AI's uselessness for easy work.
- 7 (14:01) **Why Demos Don't Equal Utility** - Diogo distinguishes between impressive demos and actually reliable, background-runnable software.
+ Full timestamped outline available in the app
Guests on this episode
Show Notes
In this episode, Diogo Almeida, co-founder and CEO of TypeSafe, joins us to discuss Jev, TypeSafe’s recently released model for bringing fast, reliable intelligence directly into software. We explore the idea of “machine-native intelligence” and why Diogo believes models optimized for generating text are poorly suited to many of the decisions required for real-world automation. He explains how Jev differs from traditional classifiers and LLM-based approaches, the role of reinforcement learning from calibrated decisions (RLCD), and why calibration and reliability are central to making AI useful as a software primitive. We also discuss the relationship between models and code, why Diogo believes AI systems should become more engineered rather than relying on a single model to do everything, and how Jev-like models could reshape agents, tool use, and the architecture of AI-powered software. 🗒️ Full show notes: https://twimlai.com/go/779.
More from this podcast
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) →