AI inverts product dev: Taste beats implementation as Codex usage explodes

Jun 28, 2026 · Lenny's Podcast
🎧 PodShort 70 min squeezed to 2 AI SprinklerAS AI / ML
Episode artwork
Andrew Ambrosino
Product and Engineering Lead for the Codex App at OpenAI
Lenny's Podcast
70 min squeezed to 2
Full episode from Lenny's Podcast
Quotable Moments

The implementation is actually not the expensive part anymore. It's dare I say taste.

Andrew AmbrosinoProduct and Engineering Lead for the Codex App at OpenAI
01:00

I think design's a little bit harder to grade... because the human aspect of taste is like part of the feedback mechanism you need.

Andrew AmbrosinoProduct and Engineering Lead for the Codex App at OpenAI
13:09

I am very confident that the Codex app that we released in February, if that had been ready in November, it would have absolutely failed in the market. And the only difference was the models between November and February.

Andrew AmbrosinoProduct and Engineering Lead for the Codex App at OpenAI
33:34
Key Insights
  • 90% of people at OpenAI use Codex, not just engineers, highlighting its widespread internal adoption.
  • The implementation phase of software development is no longer the expensive part; instead, 'taste' and the curation process have become the critical and costly elements.
  • AI is changing the shape of product work by making implementation so cheap that the traditional product development process is inverted, focusing instead on design and curation.
  • Design is more challenging for AI than coding because evaluating design involves subjective human taste, which is difficult to quantify and use as a feedback mechanism for AI.
  • Product roles at OpenAI are collapsing, leading to significantly more overlap between engineering, design, and product responsibilities, with individuals defined by the 'average of where they're working.'
  • Eliminating dedicated product roles is a mistake, as product management is a distinct discipline with valuable best practices that would be lost.
  • The most valuable individuals in the current AI-driven landscape are those who possess high agency and 'taste,' capable of guiding an idea from its inception to a high-quality finished product.
  • The Codex app, as released in February, would have 'absolutely failed' if it had been ready and released in November, solely due to the rapid advancements in underlying AI models during those few months, not changes to the product's shape.
Metrics Mentioned
  • Codex usage has grown 6X (since January.)
  • Over 5 million weekly active users (for Codex.)
  • Nearly 100% of OpenAI employees use Codex weekly (internal usage.)
  • 90% of people at OpenAI use Codex (widespread internal adoption, not just engineers.)
  • Planning for 9 months out is 'false precision' (in the current fast-evolving AI landscape, precise long-term planning is futile.)

RevBots.ai View:

  • AI Sprinkler teams bolt on AI tools but miss the cultural shift: taste curation is the new moat.
  • SaaS Hoarders clinging to rigid product processes will drown in technical debt as AI evolves.
  • ARM orgs treat AI models as living systems; they prototype on tomorrow's capabilities today.
  • Revenue teams must adopt Codex-like tools or risk being outpaced by AI-native competitors.
🎧Full Episode:Lenny's Podcast →