Andrew Ambrosino leads development of the Codex desktop app at OpenAI — and his ambition for it is unambiguous: he wants it to be "the best desktop app that has ever existed, full stop."

Nearly 100% of OpenAI employees — not just engineers — use Codex on a weekly basis. That kind of internal adoption is a signal worth paying attention to.

AI Has Flipped the Product Process

Ambrosino's central thesis is that AI hasn't just accelerated product development — it has inverted it. The traditional sequence of discovery, design, and build no longer holds in teams where anyone can ship functional software quickly.

When building velocity increases dramatically, the constraint shifts from execution to judgment. That's where a new kind of professional skill becomes critical.

"Taste" as a Core Competency

One of the most discussed ideas from Ambrosino is the rising importance of taste — not just aesthetic preference, but a sharper, more operational definition.

"Taste" is not just taste in aesthetics.

In an AI-first environment, the ability to evaluate outputs, make sharp editorial calls, and know what "good" looks like before a model can articulate it is becoming the most valuable capability a product person can have. It's the skill that doesn't automate away easily.

Why Timing the Launch Mattered

Ambrosino believes the Codex app would have failed if launched in November — and only succeeded because they waited until February. The gap wasn't about features; it was about the underlying model capabilities catching up to the product experience they were trying to deliver.

This is a useful reminder that product-market fit, in AI tools, is partially a function of model maturity, not just UX decisions.

How OpenAI's PMs Operate: Zone Defense

With everyone at the company now able to build functional software, the traditional PM role has shifted. Ambrosino describes a "zone defense" model where product managers focus less on managing execution and more on owning outcomes across a fluid, cross-functional surface.

Key characteristics of this model:

  • PMs cover broad areas of responsibility rather than fixed features
  • Role boundaries are deliberately blurred — but not eliminated
  • Removing roles entirely is, in Ambrosino's view, a significant mistake
  • Judgment, prioritization, and taste fill the vacuum left by reduced process overhead

Codex in Ambrosino's Own Workflow

Ambrosino is a practitioner, not just a builder of the tool. He uses Codex to run his own workflows, treating it as a development partner rather than an autocomplete assistant.

His background — spanning engineering, design, product management, and founding companies — gives him an unusually broad lens for thinking about where agentic tools are most and least useful.

The Bigger Vision: A Home Base for AI Work

Beyond the current desktop app, Ambrosino sketches out a vision for Codex as a coordination layer — a home base that connects work happening across ChatGPT, Codex, and the tools people already use.

This positions Codex not as a standalone coding assistant but as something closer to an operating environment for knowledge work. Whether OpenAI can execute on that vision — while Anthropic and others push hard on their own developer tools — will be one of the defining product battles of the next two years.