Exa raised $225M to build search from scratch for AI agents. Two days later Google announced 1 billion AI Mode users with queries doubling every quarter.

Google's moat was never the best algorithm. It was distribution - the default, the habit, the verb.

Exa's bet: agents do not have habits. They call an API. Developers choose on merit.

Does Google's moat transfer to the agent layer? Genuinely do not know. Curious what people think.

Every major e-commerce player is trying to crack buying inside an LLM. Four completely different bets.

Walmart tried ChatGPT checkout. 3x worse conversion. Pulled it.

Alibaba built the AI inside their own stack. 200M AI-native orders in Spring Festival.

Amazon retired a 300M-user product because the brand was the constraint.

Google built a protocol to preserve merchant trust inside AI.

Four bets. One of them is right. https://matthewmamet.com/blog/agentic-commerce-trust-problem/

The dangerous version of a fast-moving CEO is not the one who makes bad decisions. It is the one who moves fast enough that no one disagrees with them anymore.

Speed that is right 80 percent of the time trains teams to assume it is right 100. The culture tips from alignment to deference.

The best fast-moving leaders built deliberate friction into their own process. Not committees. One person whose explicit job was to push back, with the standing to do it.

I have had the same conversation twice in two months with fractional operators at different stages.

Both were carrying a client who paid the least and demanded the most. Both finally fired them.

In both cases the trigger was not the client getting worse. It was their other pipeline getting better.

When you have one client, dysfunction is a cash flow problem. When you have three, it is a decision.

The pipeline is not just revenue. It is a positioning instrument.

Alibaba is moving Taobao to conversational AI shopping. Browse, compare, buy, returns - all through AI chat.

The number everyone will talk about is scale. The problem nobody is talking about is trust infrastructure.

Marketplaces run on trust. Review systems, dispute resolution, return policies are why the transaction happens - not UX features.

At TripAdvisor we assumed 200 million reviews transferred to a new context. They did not. Different interface, different trust contract.

Steve Kaufer ran TripAdvisor's product review personally. My first session: he asked which OTAs were the largest buyers on meta. I didn't know. His reply: "Aren't you supposed to know that?"

I spent the rest of the two-hour meeting writing a HiveQL query from the back row. When the room cleared I walked over and gave him the answer. All I got was a wry smile.

Full format + calibration rules at matthewmamet.com/product-review-meeting/