A human being only has the time and mental space to read 2 Google search results before they make their decision. https://cxl.com/guides/click-through-rate/benchmarks/. The rest of the results get <50% of the views. This is the "long tail" of unread results.
The discovery layer can be improved because agents have the ability and time to parse 1 million words of content.
If you notice the way that AI does web search (or "deep research"), it's much more willing to do ten searches in a row. Then it browses 100 links. Then, for each one of those links, it does a full site scrape. I don't know a single person who would be down to research anything this deeply, no matter how much they were paid.
What this also means is that finding individual suppliers of information and individual sellers of products becomes easier.
The challenge for the future of AI discovery is this: can the AI actually evaluate the rest of those results quickly? For example, if I have a very specific type of database operation that I'm doing a lot of and I need to pick a new database that is more efficient for this process, I probably can't evaluate 1,000 databases in a short amount of time today. This is changing. Dev tools are the first to start allowing agents to sign up, and I think other types of products available online are going to follow.
As more online tools begin to allow agents to sign up for their services, the long tail of products and services that get evaluated gets longer and fatter. The equivalent of the 3rd link down on Google search goes from getting looked at 6% of the time to getting looked at 95% of the time. The 2nd page goes from ~1% of traffic to 90% as well. This is a major shift.
So then the question is, what impact does this have on the rest of finding and buying stuff on the internet, or as they say, "discovery"?
Intuitively, my mind goes to product purchases becoming more evenly distributed amongst more players.