OpenAI's CPO, Kevin Wheel, emphasizes that AI models are improving at an exponential rate, with new, more capable versions being released every two to three months. This rapid advancement means that products on the edge of current capabilities will soon become highly effective, a concept he frames as 'the AI model you're using today is the worst you will ever use.'
The skills required for product management are changing. Instead of traditional spec writing, the crucial new skills are writing 'evals' to programmatically test a model's performance on specific tasks and understanding how to fine-tune base models with proprietary data.
OpenAI operates with a 'PM light' structure (~25 PMs) to empower its product-focused engineering team. Their philosophy is built on 'iterative deployment'—shipping early to learn from real-world usage—and 'model maximalism,' which prioritizes improving the core model over building complex product features around its current limitations.
OpenAI sees itself as a platform provider, creating foundational models for 3 million developers to build upon. The company explicitly states it cannot and will not build applications for every industry, creating a vast greenfield opportunity for startups to build vertical-specific AI solutions.
Contrary to the belief that chat is a temporary UI for AI, the speaker argues it's a versatile and powerful interface. Because natural language is how humans communicate across a wide spectrum of complexity and intelligence, chat provides a universal and flexible medium for interacting with increasingly capable AI models.
Keep pulling the thread on Kevin Weil.