Graham McNicoll

AI has made it ridiculously easy to build and ship features. The problem is that AI doesn't actually solve any of the problems of product development. It hasn't told you whether that feature is going to work, or whether it did work. So the AI product development world actually makes experimentation and A/B testing even more relevant than it ever was. If you can ship a lot of features, it's really important to figure out you're not making your product worse with them. Quite often a new feature cannibalizes a part of your product that works really well, and causes a reduction in revenue. Usage of the new thing is up, but you're cannibalizing revenue or other parts of the business that turn into long-term value. So as you ship and release at ever-increasing velocities, measure the impact of what you're shipping, and make sure it doesn't violate any guardrails or parts of your business that are working. That's what we do at GrowthBook. We ship at tremendous velocity, but we make sure we measure the important parts. At GrowthBook, we use our (open source) skills for agentic experimentation to effortlessly turn any code change into an AB test.