Why Humans Are AI's Biggest Bottleneck (and What's Coming in 2026) | Alexander Embiricos (OpenAI)
•December 14, 2025
0
Why It Matters
Human interaction speed limits AI productivity; removing that bottleneck with agents like Codex could automate large portions of software development and hasten the arrival of truly autonomous AI systems by 2026.
Summary
The conversation centers on Alexander Embiricos’s work leading Codex, OpenAI’s coding assistant, and his thesis that human limitations—particularly typing and multitasking speed—are the primary bottleneck to realizing fully autonomous AI agents. Embiricos describes Codex as an “intern” that can write, test, and even manage its own training infrastructure, illustrating its impact with the rapid development of the Sora Android app, which went from concept to top‑ranking App Store release in under a month.
Key insights include Codex’s explosive 20‑fold growth, its shift from pure code generation toward proactive, end‑to‑end software‑engineering tasks, and the observation that the most effective way for AI agents to leverage computers is by writing code. Embiricos also highlights OpenAI’s uniquely “bottom‑up” organization that emphasizes speed, empirical iteration, and hiring world‑class talent, allowing product cycles that would be impossible at traditional tech firms. He notes that even without further model breakthroughs, product innovation alone can unlock massive value.
Notable quotes underscore the theme: “The current underappreciated limiting factor is literally human typing speed,” and “Codex caught configuration mistakes that even senior engineers missed.” He also likens OpenAI’s culture to “ready‑fire‑aim,” stressing rapid deployment and learning from real‑world usage. The Sora example—built in 18 days with Codex’s assistance—demonstrates how AI can compress development timelines dramatically.
The implications are profound for developers, enterprises, and investors. If AI agents can overcome human interaction bottlenecks, software creation could become a largely automated pipeline, reshaping talent demand and accelerating the path toward more capable AGI systems projected for around 2026. Companies that integrate coding agents early may gain a decisive competitive edge, while the broader industry must grapple with new workflows, security considerations, and the shifting role of human engineers.
Original Description
Alexander Embiricos leads product on Codex, OpenAI’s powerful coding agent, which has grown 20x since August and now serves trillions of tokens weekly. Before joining OpenAI, Alexander spent five years building a pair programming product for engineers. He now works at the frontier of AI-led software development, building what he describes as a software engineering teammate—an AI agent designed to participate across the entire development lifecycle.
We discuss:
1. Why Codex has grown 20x since launch and what product decisions unlocked this growth
2. How OpenAI built the Sora Android app in just 18 days using Codex
3. Why the real bottleneck to AGI-level productivity isn’t model capability—it’s human typing speed
4. The vision of AI as a proactive teammate, not just a tool you prompt
5. The bottleneck shifting from building to reviewing AI-generated work
6. Why coding will be a core competency for every AI agent—because writing code is how agents use computers best
Brought to you by:
WorkOS—Modern identity platform for B2B SaaS, free up to 1 million MAUs: https://workos.com/lenny
Fin—The #1 AI agent for customer service: https://fin.ai/lenny
Jira Product Discovery—Confidence to build the right thing: https://atlassian.com/lenny
My biggest takeaways (for paid newsletter subscribers): https://www.lennysnewsletter.com/i/180365355/my-biggest-takeaways-from-this-conversation
Where to find Alexander Embiricos:
• LinkedIn: https://www.linkedin.com/in/embirico
Where to find Lenny:
• Newsletter: https://www.lennysnewsletter.com
• LinkedIn: https://www.linkedin.com/in/lennyrachitsky/
In this episode, we cover:
(00:00) Introduction to Alexander Embiricos
(05:13) The speed and ambition at OpenAI
(11:34) Codex: OpenAI’s coding agent
(15:43) Codex’s explosive growth
(24:59) The future of AI and coding agents
(33:11) The impact of AI on engineering
(44:08) How Codex has impacted the way PMs operate
(45:40) Throwaway code and ubiquitous coding
(47:10) Shipping the Sora Android app
(49:01) Building the Atlas browser
(53:34) Codex’s impact on productivity
(55:35) Measuring progress on Codex
(58:09) Why they are building a web browser
(01:01:58) Non-engineering use cases for Codex
(01:02:53) Codex’s capabilities
(01:04:49) Tips for getting started with Codex
(01:05:37) Skills to lean into in the AI age
(01:10:36) How far are we from a human version of AI?
(01:13:31) Hiring and team growth at Codex
(01:15:47) Lightning round and final thoughts
Referenced:
• OpenAI: https://openai.com
• Codex: https://openai.com/codex
• Inside ChatGPT: The fastest-growing product in history | Nick Turley (Head of ChatGPT at OpenAI): https://www.lennysnewsletter.com/p/inside-chatgpt-nick-turley
• Dropbox: http://dropbox.com
• Datadog: https://www.datadoghq.com
• Andrej Karpathy on X: https://x.com/karpathy
• The rise of Cursor: The $300M ARR AI tool that engineers can’t stop using | Michael Truell (co-founder and CEO): https://www.lennysnewsletter.com/p/the-rise-of-cursor-michael-truell
• How Block is becoming the most AI-native enterprise in the world | Dhanji R. Prasanna: https://www.lennysnewsletter.com/p/how-block-is-becoming-the-most-ai-native
• Lessons on building product sense, navigating AI, optimizing the first mile, and making it through the messy middle | Scott Belsky (Adobe, Behance): https://www.lennysnewsletter.com/p/lessons-on-building-product-sense
• Sora Android app: https://play.google.com/store/apps/details?id=com.openai.sora&hl=en_US&pli=1
• The OpenAI Podcast—ChatGPT Atlas and the next era of web browsing: https://www.youtube.com/watch?v=WdbgNC80PMw&list=PLOXw6I10VTv9GAOCZjUAAkSVyW2cDXs4u&index=2
• How to measure AI developer productivity in 2025 | Nicole Forsgren: https://www.lennysnewsletter.com/p/how-to-measure-ai-developer-productivity
• Compiling: https://3d.xkcd.com/303
• Jujutsu Kaisen on Netflix: https://www.netflix.com/title/81278456
• Tesla: https://www.tesla.com
• Radical Candor: From theory to practice with author Kim Scott: https://www.lennysnewsletter.com/p/radical-candor-from-theory-to-practice
• Andreas Embirikos: https://en.wikipedia.org/wiki/Andreas_Embirikos
• George Embiricos: https://en.wikipedia.org/wiki/George_Embiricos: https://en.wikipedia.org/wiki/George_Embiricos
...Resources continued at: https://www.lennysnewsletter.com/p/why-humans-are-ais-biggest-bottleneck
_Production and marketing by https://penname.co/._
_For inquiries about sponsoring the podcast, email podcast@lennyrachitsky.com._
Lenny may be an investor in the companies discussed.
0
Comments
Want to join the conversation?
Loading comments...