OpenAI Codex lead on the new shape of product work | Andrew Ambrosino

Andrew Ambrosino discusses how AI, particularly models like GPT, has completely flipped traditional product development processes. Instead of lengthy research and prototyping phases before implementation, the new era emphasizes rapid exploration via AI-generated prototypes and the importance of taste and curation in selecting the best options among many. He highlights that implementation is now cheap and fast, making the curation and taste the most valuable skills. The shift allows teams to experiment more freely and dynamically without the heavy resource constraints of the past.

Ambrosino emphasizes the importance of 'taste'—good judgment and aesthetic sensibility—in the AI-first workplace. Taste involves making nuanced decisions about design, features, and context, which AI currently struggles with due to its difficulty in understanding human culture and subtlety. He explains that taste involves systems thinking, context awareness, and a sense of novelty and appropriateness, all crucial for producing effective and appealing products in an AI-enhanced environment.

Despite advances, AI models still lag in producing high-quality design because design evaluation is subjective and complex. Ambrosino notes practical reasons such as the difficulty of grading design quality and the lack of direct training focus on design as a research priority. However, he believes future models will improve, though the deeper challenges around culture, abstraction, and creative novelty remain. He also discusses how design processes are adapting—moving away from rigid frameworks toward more experimental, prototype-driven, and iterative approaches.

Ambrosino describes a collapsing of traditional roles in AI-powered organizations, especially on his team at OpenAI, where boundaries between engineers, designers, and product managers are blurring. Roles are seen as averages of what team members do, emphasizing capacity and taste over strict discipline boundaries. He advocates for a 'zone defense' model, where team members specialize in areas aligned with their strengths to guide chaos and coordinate innovative efforts, rather than rigid role assignments.

The conversation explores the idea that AI enables teams to pursue more ambitious projects, even ones that feel ahead of market readiness. Ambrosino discusses how continuous iteration, rapid prototyping, and AI-driven exploration allow products to evolve through multiple versions—sometimes re-releasing the same core idea better suited to current AI capabilities. He highlights explorations around autonomous development, loops, and AI-automated workflows as the newest frontiers for building in this era.

Ambrosino shares how he personally leverages Codex and AI tools to automate his workflows—from managing Slack channels, automating research, to setting up complex integrations like linking with Excel or building extensions for video editing. He describes a future where AI-driven apps act as home bases for work, orchestrating across multiple tools and tasks seamlessly, and emphasizes the importance of experimenting with these capabilities to discover new possibilities.

The podcast discusses the evolution of Codex from a developer tool to a general knowledge work platform. Ambrosino envisions a unified environment that acts as a central hub for diverse workflows, connecting with third-party apps and automating complex tasks. He talks about building primitives for extensibility and the ongoing work to balance powerful automation with user control, aiming for a flexible, intelligent desktop environment where AI can handle both routine and complex tasks.

Ambrosino candidly shares that he experienced numerous micro-failures over his career, including iterations that didn't succeed and lessons learned from them. He stresses that persistence, curiosity, and willingness to try new approaches without fear of failure are critical. This also underpins the culture at OpenAI, where continuous testing, feedback, and iteration drive progress.

1. AI dramatically shifts the traditional product development process toward rapid experimentation and curation. 2. 'Taste'—judgment and aesthetic sensibility—becomes an increasingly valuable skill in an AI-centric environment. 3. Design remains a challenge for AI, especially in producing novel, culturally nuanced solutions, but future improvements are likely. 4. Traditional boundaries and roles are collapsing; teams are more fluid and overlapping in responsibilities. 5. Teams are encouraged to be more ambitious, constantly experimenting, and iterating through multiple versions. 6. Automation and AI integration into workflows are now practical, enabling more complex and personalized tasks. 7. Continuous failure and learning are vital for innovation, with persistence crucial for success.