Lenny's Podcast• Dec 14, 2025• 1:58:29Interview

Inside OpenAI： 2026 is the year of agents, AI’s biggest bottleneck, and why compute isn’t the issue

From Lenny's Podcast

Nick Turley•VP of Product, OpenAI

Executive Summary

Alexander Embirikos, Product Lead for Codex at OpenAI, discusses the vision for Codex as a proactive 'software engineering teammate' rather than just a coding tool.
He details the product's explosive 20x growth, driven by a shift to a more intuitive, locally-integrated user experience.
Embirikos provides concrete examples of Codex's impact, such as enabling a small team to build and ship the #1 ranked Sora Android app in just 28 days, and highlights the belief that all future AI agents will need to be coding agents to effectively interact with computers.
The conversation also explores OpenAI's unique, high-speed, bottoms-up operating culture and the evolving role of engineers in an AI-accelerated world.

8 quotes

Concerns Raised

The user experience of reviewing large volumes of AI-generated code can be less enjoyable and is a new bottleneck.
Non-coding AI agents are still in their early stages and lack the effectiveness of coding agents.
Initial product versions can be 'too far in the future' for the general market, requiring a more gradual on-ramp for users.

Opportunities Identified

Massive acceleration of software development cycles, as demonstrated by the Sora and Atlas projects.
The emergence of a 'super assistant' or 'teammate' paradigm for AI, where agents are proactive and integrated into workflows.
Building vertical AI startups that deeply understand and serve specific, currently underserved customer segments.
Blurring of traditional roles (PM, design, engineering) as AI tools empower individuals to perform a wider range of tasks.

Key Themes

Codex as a Software Engineering Teammate

The core vision for Codex is not just as a code completion tool, but as a proactive, integrated teammate that participates across the entire software development lifecycle, from ideation and planning to deployment and maintenance. This involves moving beyond reactive prompting to a state where the agent can anticipate needs and work autonomously.

This signals a shift in the human-computer interaction paradigm for developers, moving from tool-user to collaborator.

Extreme Development Acceleration

The use of Codex within OpenAI has led to a dramatic reduction in development time. The Sora Android app was built and shipped by a team of 2-3 engineers in 28 days, and tasks on the Atlas browser that once took 2-3 engineers weeks now take one engineer a single week.

This demonstrates a step-change in engineering productivity, suggesting that the primary bottlenecks are shifting from implementation to ideation, planning, and review.

The Primacy of Coding Agents

A core belief at OpenAI is that the most effective way for an AI to interact with a computer is by writing code. This implies that any powerful, general-purpose 'super assistant' will need to have a coding agent at its core, positioning Codex as a foundational technology for future AI products.

This suggests that the development of agentic AI will be heavily reliant on advancements in code generation and execution capabilities.

OpenAI's High-Velocity Operating Model

OpenAI's rapid progress is attributed to its high talent density, an 'incredibly bottoms-up' culture, and an empirical 'ready, fire, aim' approach to product development. This allows the company to quickly test ideas and learn, which is essential in a field where capabilities are evolving rapidly.

This provides a model for how organizations can structure themselves to innovate in fast-moving, uncertain technological domains.

The Shifting Role of the Software Engineer

With AI handling more of the code generation, the role of the engineer is evolving. The focus is shifting towards higher-level tasks like planning, architectural design, and reviewing AI-generated code. This creates a new challenge in making the code review process efficient and enjoyable.

Professionals must adapt their skills to focus on system design and validation, as the manual task of writing code becomes increasingly automated.

Get started free

Topics

OpenAI Codex AI Agents Software Development Developer Tools Sora Atlas Browser GPT-5.1 Codex Max Product Management AGI Code Review Proactive AI Spec-driven Development Vertical AI

Processed Feb 24, 2026 yt-dlp + mlx-whisper + Gemini