In fluid, AI-powered teams, job descriptions are obsolete. A person's role is defined by the center of gravity of their contributions—whether they skew towards code, product specs, or design. This allows for more overlap and agency, moving away from rigid "lanes."
While AI makes prototyping easy, it's not always the right first step. A prototype can create a "primal mark" that biases the team towards a specific execution. For clarifying a vague problem space, a document may be better to avoid anchoring to a visual solution too early.
Traditionally, implementation was expensive, so teams de-risked ideas with docs. With AI, building is cheap, so teams now create numerous prototypes first and then curate them. The process is now "build then decide," not "decide then build," with curation and taste becoming the most expensive part.
The Codex team resists optimizing their own workflows. Instead, they use the product to perform those tasks, even when it's not the best tool. This painful dogfooding loop forces them to make the product better at solving real-world process problems, turning internal pain into user value.
AI's impact isn't eliminating roles like engineering or design, but rather the artificial barriers to entry. Previously, proficiency in a specific tool or syntax acted as a gatekeeper. AI lowers this barrier, shifting focus from tool mastery to core skills like problem-solving and taste.
AI models excel at coding because correctness is easy to evaluate. Design is harder because "good" is subjective and tied to human taste, making it difficult to create a training feedback loop. Furthermore, design values novelty and cultural context, whereas software engineering prefers established, reliable patterns.
With engineering democratized, ideas and prototypes emerge from everywhere. A PM's role shifts from a top-down feature owner to a 'zone defender' who spreads out with other PMs. They identify gaps, provide curation, and ensure alignment across chaotic, bottoms-up innovation.
The OpenAI Codex app would have "absolutely failed" if launched three months earlier. The only difference was the underlying model's capability. This reveals a new product risk: a perfectly designed product can fail simply because the AI isn't smart enough yet, requiring teams to relaunch ideas as models improve.