OpenAI Codex lead on the new shape of product work | Andrew Ambrosino
Summary
OpenAI's Andrew Ambrosino, Product and Engineering Lead for the Codex app, discusses how AI is fundamentally reshaping product development. The Codex app, used weekly by nearly 100% of OpenAI employees and boasting 5 million weekly active users with 6x growth since January, exemplifies this shift. Ambrosino highlights that software implementation is now cheap, making "taste" and curation the most expensive and critical aspects of product work. He notes that current AI models struggle with design due to the subjective nature of human taste and the challenge of grading design outputs. While the traditional, linear design process is obsolete, understanding the stage of the design process remains crucial. Ambrosino describes a trend of role collapse, where individual contributions are defined by the average of their work, blurring traditional boundaries. Planning for AI products is challenging, requiring short-term precision and long-term flexibility, as product success is heavily tied to evolving model capabilities. The Codex app is envisioned as a "home base" for general knowledge work, seamlessly integrating with other applications like Excel and Premiere Pro, rather than solely a developer tool.
Key takeaway
For AI Product Managers or Directors of AI/ML navigating evolving team structures, recognize that AI has inverted the cost of software development. Implementation is now cheap, making "taste" and curation the most valuable skills. You should prioritize cultivating high-agency, adaptable individuals who can discern signal from noise amidst abundant prototypes. Embrace fluid roles and continuous prototyping, but avoid completely abandoning established product disciplines, as specialized knowledge remains crucial for effective outcomes. Focus on building products that can evolve with model intelligence, even if they aren't fully viable today.
Key insights
AI shifts product focus from costly implementation to subjective "taste" and curation, demanding adaptable roles and processes.
Principles
- Implementation is cheap; taste and curation are paramount.
- Product success hinges on model intelligence, not just shape.
- Roles are fluid, defined by average work, not strict boundaries.
Method
Detail short-term plans, keep long-term hazy. Prototype all ideas, then let them "bake" until models catch up. Automate personal workflows and coach the AI.
In practice
- Automate daily briefs from Slack channels using AI agents.
- Develop features that are currently too ambitious, awaiting model advancements.
- Integrate AI apps with specialized desktop tools via extensions or computer use.
Topics
- OpenAI Codex
- AI Product Development
- Product Management
- Software Engineering
- Design Process
- AI Agents
Best for: Product Manager, AI Product Manager, Director of AI/ML, AI Engineer
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by Lenny's Newsletter.