A Radisson hotel search in Chrome makes 568 requests. Only 4 of them return the prices. Hyper Solutions rebuilt it as plain HTTP calls in Go, no browser. Across 100 UK sessions on Massive, 490 of 500 searches returned rates at a median 1.5s each.

Most teams treat robots.txt as either law or noise. Courts in 2025 said it is neither. In the US it's a request. In the EU it can be the opt-out that blocks AI training.

Seltz (@seltz_ai) returns web search results to AI agents in about 300ms, as full-text documents from its own index. Five endpoints, one per job: Search for raw documents, Answer for one cited answer, Fetch for URLs you have, Monitor for what's new, Agent for deep research.

Are residential proxies legal? Yes. The ones in the headlines took devices without asking. What gets networks shut down is taking devices without consent. Sourcing, use, and local law decide

Agentic commerce could handle up to $1 trillion in U.S. retail by 2030 (McKinsey). It runs on 4 open protocols: ACP, UCP, AP2 and TAP. We checked 36 well-known stores for the file an AI shopping agent reads first. 11 had one. Only 2 of 22 big retailers did.

Gaffa runs a real cloud browser behind one REST call. Send a URL and a list of actions, get back markdown, a PDF, a screenshot, or JSON. For residential IPs, Gaffa works with several proxy providers, and Massive is one of them.

Myth: an AI agent doesn't need a location. It already has one. Sites read it from the IP, and the page changes with it: search results, AI answers, prices. If you don't choose it, your server's data center does.

Daily AI agent requests on Cloudflare grew more than 1,700% in a year, by Cloudflare's count. At Birthday Week two weeks ago, it added ways to charge that traffic. A thread on what is new, and what was already there: