Recommendations in the AI Era

A question I get asked a lot If Rendezvu exists to power the future of commerce with expert recommendations, what happens when everyone just asks an AI what to buy? I used to get a little defensive about this question. Now I think it's a fun one to answer. I agree with the premise. Shopping is moving into the chat window (or some extension of that). A year from now, more people will ask a model which trail shoe to buy. I do not think that is a threat to what we are building. It's exactly what we're building for. Context is the whole game I have written before that context is king. I meant it about communication inside a company. It turns out to be even more true about these machines. Anyone who has spent real time working with AI tools learns the same lesson eventually. The model is rarely the bottleneck now. What you give it is. Hand it a vague prompt and you get a vague answer. Hand it the right documents, the right constraints, and the right history, and it feels like working with someone who has been on your team for years. Same model. Different context. That gap is not a model problem. It is a data problem. The best reasoning engine in the world, asked to pick a shoe for your first 50 miler, is still reasoning over marketing copy, aggregated star ratings, and whatever a stranger typed into a forum in 2019 who spends most of their time writing articles rather than being outside. I do not expect that gap to close on its own. The labs are going to keep making the engines better. Someone has to make the fuel better. That's where we come in. What we're actually doing When you ask a model what to buy today, you are mostly getting a weighted average of random opinions from people you don't know. But that's not where shopping is headed, especially in the outdoor industry. I want to be clear that I love YouTube and newsletters and blogs. It has taught me over half of what I know about running and AI and is a serious wealth of knowledge. But there is a difference between a thread started by a username and a recommendation from your friend, coach, guide or someone you trust. That is the entire premise of Rendezvu. Before we existed, that guide's knowledge lived in a conversation 1:1 conversation. Nobody wrote it down. Nobody could cite it. A model certainly could not find it. So we gave people a place to write it down. Not reviews. Not sponsored posts (well, some). The actual kit they use in the wild, with their own notes on why, in their own words, tied to their real name and their real record. A person who is in the field, testing the gear, and standing behind the call, publicly. What we are seeing I am careful about sharing numbers, but a few patterns are worth putting in writing because they changed how I think about this. The first is depth. The average person on Rendezvu is not recommending three things. They are recommending multiple kits, somewhere north of twenty products. Nobody asked them to write that much. Experts want to explain their choices. Give them a place to do it and they will. The second is breadth. Those recommendations now span well over a thousand brands. That surprised me. I expected the big names to dominate (which to some extent they do), but a huge share of the kit people actually trust comes from small companies that no algorithm would surface on its own. More and more people are wanting to feel closely tied to a brands identity, and the smaller brands have the ability to get very close to their customer. The third is intent. When someone lands on a host's profile, a meaningful share of them open a gear list, and a meaningful share of those click through to buy. The ratio between the top and the bottom of that funnel is tighter than anything in traditional affiliate. People who arrive through a trusted person behave like they already made the decision and are just confirming it. None of this is a claim that we have won. It is early and the numbers are still small. But the shape of the data is promising. Trusted people, writing in their own voice, about gear they actually use, produce recommendations that convert. That is the theory being confirmed. Why this compounds I try to think in decades. On a one year horizon, AI looks like a threat to anyone in the recommendation business. The models get better every quarter and they are free. On a ten year horizon it looks like the opposite. Every model will be excellent. Excellence will be table stakes. What will separate a great answer from a mediocre one is what the model was allowed to read before it answered. That is why I believe the moat is the data and not the interface. Interfaces get copied in a weekend now. A decade of real people documenting what they actually carry, why they switched, what broke, and what they would buy again does not get copied. It has to be lived. And every month that goes by, the gap between a dataset built from lived experience and one scraped from the open web gets a little wider. I am not betting against AI. I am betting on Anthropic and the rest of them to keep making the engines better. I am betting that knowledge from people who have earned it stays scarce, and that the systems of the future will pay a premium for it. Our job is to be where that knowledge lives. What we are building toward Rendezvu is not competing with AI shopping. Rendezvu is what AI shopping is going to need. We are taking the knowledge that used to disappear at the trailhead and putting it into the digital world, attached to real people, in a form that both a human and a machine can trust. We are making sure the people who hold that knowledge get paid when it is used. And we are giving brands a way to find and back the people who believe in them too.