Product-Led AI: How to Turn LLM Geekery Into Customer Outcomes

Engineers are excited about LLMs, agentic AI, RAG, and new vendors. They spend time exploring tools and playing with new toys, often without a real problem in mind and sometimes without a realistic path to deploy anything inside the organization. The engineers may not be talking to Legal, Security, Risk, or anyone else who controls whether this can actually reach users.

Can you think of other times when this much energy was spent on technical innovation without connecting it to a real problem? Of course you can: mobile, cloud, big data, and almost every major technology wave. I cannot think of a significant technology wave that did not have a cart-before-the-horse phase.

That is tech-led behavior. Product-led AI would look different. A starting point:

• Align around the customer outcomes we’re hoping GenAI will serve—e.g., improving the Developer Experience, the Shopper Experience, or the Fraud Detection Officer Experience.

• Empower the team to deliver outcomes. If Legal is a major issue for deploying AI, include someone who understands that perspective and is ideally empowered to make some decisions. If Security or Risk is the issue, include Cybersecurity. Localize dependencies so the team can try real things fast.

• Explore the problem and opportunity space. What are the biggest friction points or bottlenecks in the current experience or process?

How are you approaching your GenAI efforts? What would it look like to apply a product mindset and operating model to developing AI capabilities, even when they have nothing to do with your external products?