OpenAI Codex lead on the new shape of product work | Andrew Ambrosino
TL;DR
OpenAI Codex lead Andrew Ambrosino explains how AI has inverted product development, making implementation so abundant that taste and curation—not coding—are now the primary bottlenecks, while Codex scales to 5 million weekly users and 90% internal adoption at OpenAI.
🔄 The Inversion of Product Work 3 insights
Implementation is now abundant
At OpenAI, implementation is no longer the expensive part, with 90 uncoordinated teams often prototyping the same feature simultaneously due to AI lowering barriers to entry.
Taste becomes the bottleneck
The critical skill has shifted from building to curation, determining which of many possible implementations to pursue and how they fit into broader product systems.
Codex achieves massive adoption
Codex usage grew 6x since January to over 5 million weekly active users, with 90% of OpenAI employees—including non-engineers—using it weekly.
📝 Documents vs. Prototypes 3 insights
PRDs are not dead
Despite claims that documents are obsolete, they remain essential for clarifying vague conceptual areas where prototypes would create premature anchoring.
The primal mark risk
High-fidelity prototypes now risk misleading stakeholders because polished visuals no longer signal derisked assumptions or strategic clarity.
Match medium to message
Teams must deliberately choose between documents and prototypes based on whether they need to explore ideas or stress-test interactions.
🎨 Why AI Struggles with Design 3 insights
Taste is hard to grade
Design lags behind coding because it lacks objective correctness metrics and requires human taste as part of the training feedback loop.
Novelty vs. patterns
While software engineering benefits from known patterns, good design requires cultural novelty and randomness that models currently overindex away from.
The abstraction gap
Current technology struggles with the semantic abstraction layer connecting visual design to codebase architecture, such as maintaining consistent interaction patterns across rebrands.
Bottom Line
Develop taste and curation skills to select and frame the right ideas, as the ability to build becomes commoditized and abundant.
More from Lenny's Podcast
View all
Building the most AI-pilled engineering team in the world | Fiona Fung (Anthropic)
Fiona Fung, leader of Claude Code and Co-work at Anthropic, reveals how her engineers now ship 8x more code than in 2021, fundamentally shifting the engineering bottleneck from writing to verification and requiring new AI-native management techniques to maintain quality at scale.
The hidden pattern behind successful products | Mark Pincus (FarmVille, Words with Friends, & more)
Mark Pincus shares his "Proven Better New" framework for building hit consumer products, arguing that founders should copy proven elements, make incremental improvements that 10/10 users love, and treat truly novel features as high-risk experiments—contrary to the instinct to prioritize innovation over familiarity.
Father of the iPod and iPhone on building taste, judgment, and creativity in the AI era
Tony Fadell shares lessons from building the iPod and iPhone, arguing that creating category-defining products requires resisting AI-driven cognitive surrender, embracing opinion-based decision-making for 1.0 versions, and micromanaging critical details while maintaining ruthless focus on customer pain points and storytelling.
A rational conversation on where AI is actually going | Benedict Evans
AI represents a transformative shift comparable to the internet or mobile, but we remain in its infancy (like 1997), with adoption spread unevenly across industries and the real challenge being integration and judgment rather than mere task automation.