AI Product Development Requires Dual Approach

Nick Turley, head of ChatGPT at OpenAI, shares a strategic framework for building AI products that balances model capabilities with user needs. This approach helped grow ChatGPT from an experimental project to 700 million weekly active users.

The Dual-Track Product Development Framework

Track 1: Working Backwards from Model Capabilities

• Focus on identifying and productizing the unique capabilities of your AI model
  • "You have to practice working backwards from the model capabilities... looking at what tech we have available and what is the most awesome way to productize it"
  • "If you apply some sort of PM framework to that, I think you would do something horribly wrong"
  • Example: When GPT-5 showed strong front-end coding abilities, it required reprioritizing product roadmaps
• Identify "magical" capabilities that would be impossible without AI
  • Voice features weren't built because customers demanded it, but because they discovered a way to make models work with "anything in, anything out"
• Ship capabilities before they're fully polished to discover their potential
  • "You won't know what to polish until after you ship"
  • "This is a pattern with AI where you really have to ship to understand what is even possible"

Track 2: Classic Product Management

• Listen to customers and identify common needs across segments
  • "When you look at end users, there's actually an immense amount of overlap in terms of what they want"
  • Core primitives like projects, history, search, sharing and collaboration are universal needs
• Build infrastructure for enterprise requirements
  • "You gotta do HIPAA, you gotta do SOC 2... if you wanna be a serious player"
• Systematically improve the model on use cases people care about
  • "You need to look at what people are trying to do... and systematically improve on those use cases"
  • This requires talking to users, analyzing data, and iterating based on feedback

Key Implementation Principles

Maximize Velocity

• Create a culture of urgency with a "maximally accelerated" mindset
  • "If this was the most important thing and you wanted to truly maximally accelerate it, what would you do?"
  • This helps distinguish critical path items from what can happen later
• Set the "resting heartbeat" for your team at a fast pace
  • "I always felt like part of my role here is to set the pace and the resting heartbeat"
• Ship quickly, even with imperfections
  • ChatGPT went from decision to ship in just 10 days
  • "The model had a bunch of shortcomings... but it was so cool to be able to iterate on the model"

Treat the Model as a Product

• "There really is no distinction between the model and the product - the model is the product"
• Iterate on the model like you would a product feature
  • Move away from "hardware-like" releases to continuous improvement
  • "My dream is that we could just ship daily or even hourly like in software land"
• Focus on improving "vibes" and personality, not just capabilities
  • "The vibes are good" is a key metric for model quality
  • "It just feels a little bit more alive, a bit more human"

Learn from Real-World Usage

• Use real-world failures to improve models
  • "You need real failure cases to make these things better"
  • "Benchmarks are increasingly saturated so really you need real world scenarios"
• Develop evals based on actual user behavior
  • "It's not some technical magic... it's really just about articulating success in a way that is maximally useful for training models"
• Balance speed with safety through rigorous processes
  • "For frontier models there actually needs to be a rigorous process where you red team, work on the system card, get external input"

Team Structure for AI Products

• Build interdisciplinary teams that combine research, engineering, design and product
  • "If a feature doesn't get 2x better as the model gets 2x smarter, it's probably not a feature we should be shipping"
• Hire for curiosity over experience
  • "Curiosity is an attribute that matters so much more than your ML knowledge"
  • "If you were to filter for people who've done it before, you would have a very narrow filter of very lucky people"
• Structure teams like jazz bands, not orchestras
  • "I don't believe in everyone having this set part that they have to play and me telling people when to play"
  • "Great product development is like [jazz] in the sense that ideas could come from anywhere"