OpenAI Codex lead on the new shape of product work | Andrew Ambrosino
OpenAI Codex lead on the new shape of product work | Andrew Ambrosino
OpenAI Codex lead on the new shape of product work | Andrew Ambrosino
Andrew Ambrosino, who leads development of the Codex desktop app at OpenAI, discusses how AI is fundamentally reshaping product development. He argues that the most valuable skill in an AI-first workplace is no longer implementation, but 'taste'—the ability to curate, direct, and make strategic decisions. The conversation explores how this shift is collapsing traditional roles, changing team structures, and creating new challenges for planning and design.
Ambrosino explains that AI has inverted the product process: implementation is now cheap, so the focus moves from de-risking with documents to curating many attempts with taste. He defines taste as systems thinking and strategic direction, not just aesthetics. While AI excels at code, it struggles with design due to a lack of clear feedback loops and the need for cultural context. On the Codex team, roles are collapsing—designers and PMs write code—but Ambrosino warns against eliminating specialties entirely, as that discards valuable best practices. He describes a 'zone defense' model for product management, where people spread out to cover gaps. Planning roadmaps is difficult in a fast-moving AI environment, where success often depends on model timing. Ambrosino shares how he uses Codex to automate his own workflows, from managing releases to setting up daily briefs, and envisions Codex as a home base that coordinates work across other tools. He emphasizes that the key skill is having the taste and agency to take an idea from concept to completion.
00:00
00:00
Taste, not implementation, is now the most valuable skill.
02:33
02:33
Implementation is cheap, curation is expensive
09:18
09:18
Taste is emerging as the most valuable skill across all fields.
10:26
10:26
Taste involves deciding what to build and how it fits into a larger system
12:06
12:06
Design is harder to grade than code
16:18
16:18
The process was always flawed.
21:35
21:35
Roles are defined by the average of their work, not strict boundaries.
23:41
23:41
Eliminating roles discards valuable best practices.
29:55
29:55
Hire product-minded engineers to avoid full team reviews.
30:13
30:13
The most valuable person is someone who can take an idea from concept to completion with high taste and agency.
31:38
31:38
Product success depends on model timing, not feature shape.
35:17
35:17
A feature's success depends on timing and model intelligence.
38:13
38:13
Bottom-up exploration enables disruptive innovations
39:19
39:19
The key question is supervised vs. unsupervised development.
46:52
46:52
Codex used computer use to automate the setup by clicking through the interface
51:32
51:32
Designing for both beginners and power users is a challenge.
52:06
52:06
Codex's value spans both developer and general knowledge work.
57:20
57:20
Be more ambitious with AI and just try things.
1:00:30
1:00:30
Micro-failures shaped the product
1:01:52
1:01:52
The Big Orange Splot has a great anti-HOA message.
1:09:35
1:09:35
Found product-market fit for his role.
