Blog
-
How to Run Customer Interviews for a Capability Your Users Have Never Experienced
Standard discovery interviews ask about past behavior. That breaks when the capability doesn’t exist yet. Here are three ways to…
-
Traits before tools: rethinking when to insert AI into your ways of working
AI amplifies how you already work. Barry O’Reilly’s traits→tasks→tools model explains the AI ROI gap—and how flipping the order closes…
-
How to Collect Product Feedback When Your AI Gives Every User a Different Answer
Traditional product feedback assumes everyone uses the same product. AI features break that assumption. Here’s an instrumentable way to fix…
-
How to prioritize your backlog when effort is no longer the constraint
How to use learning and reversibility to prioritize work AI can now build cheaply and quickly
-
How to write OKRs for an AI product
If your AI product delivers different outputs to each user and prompt, here is how you measure the success of…
-
How product management can fix your AI integration problems
Your AI integration problem is a product management problem: why pilots stall on old workflows, and the product approach that…
-
Training your PMs on AI tools is the easy part
AI tool training makes teams faster, not more innovative. Without permission to experiment and fail, you just get a faster…
-
The Amazon AI-tokens problem isn’t an Amazon problem.
Measuring token consumption means exactly zero to your customers. So why are companies like Amazon forcing their employees to do…
-
What “done” means when you’re shipping AI features
AI broke the old definition of “done.” You’re no longer shipping vending machines—you’re shipping probabilistic behavior distributions.
-
Karpathy said vibe coding is obsolete. What he described instead is product management.
Andrej Karpathy called “agentic engineering” the future of building. Strip the engineering vocabulary and he just described product management.