Why humans are AI’s biggest bottleneck (and what’s coming in 2026) | Alexander Embiricos (OpenAI Codex Product Lead)
Source
Lenny Rachitsky
Author
Lenny Rachitsky
Date
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
Why it matters
If you're building with coding agents, this lays out how OpenAI thinks about Codex's role across the full dev lifecycle and why review — not writing — is becoming the new bottleneck, with a concrete case study (Sora Android in 18 days) to ground the argument.
Key quotes
“It turns out the best way for models to use computers
is simply to write code.”
Alexander Embiricos
“if you want to build any agent,
maybe you should be building a coding agent.”
Alexander Embiricos
“The current underappreciated limiting factor
is literally human typing speed
or human multitasking speed.”
Alexander Embiricos
“the Sora Android app, a fully new app,
we built it in 18 days.”
Alexander Embiricos
“previously it would have taken two to three weeks
for two to three engineers,
and now it's like one engineer one week.”
Alexander Embiricos
Chapters
00:00Introduction to Alexander Embiricos
05:13The speed and ambition at OpenAI
11:34Codex: OpenAI's coding agent
15:43Codex's explosive growth
24:59The future of AI and coding agents
33:11The impact of AI on engineering
44:08How Codex has impacted the way PMs operate
45:40Throwaway code and ubiquitous coding
47:10Shipping the Sora Android app
49:01Building the Atlas browser
53:34Codex's impact on productivity
55:35Measuring progress on Codex
58:09Why they are building a web browser
01:01:58Non-engineering use cases for Codex
01:02:53Codex's capabilities
01:04:49Tips for getting started with Codex
01:05:37Skills to lean into in the AI age
01:10:36How far are we from a human version of AI?
podcastThe 100-person AI lab that became Anthropic and Google's secret weapon | Edwin Chen (Surge AI)
podcastThe coming AI security crisis (and what to do about it) | Sander Schulhoff
podcastWhy most AI products fail: Lessons from 50+ AI deployments at OpenAI, Google, and Amazon
podcastThe non-technical PM’s guide to building with Cursor | Zevi Arnovitz (Meta)
podcastOpenAI Codex lead on the new shape of product work | Andrew AmbrosinoLenny Rachitsky
podcastAI’s third era: the rise of persistent AI coworkers | Tara Seshan (OpenAI’s product lead)Lenny Rachitsky
podcast“Engineers are becoming sorcerers” | The future of software development with OpenAI’s Sherwin WuLenny Rachitsky