What is agentrefs.net, and why does it exist

It’s 2026, and I’ve just finished a big contract building a custom AI agent framework for social media message automation. It went off without a hitch. I delivered on the software requirements, on time and on budget.

Afterwards they needed help with the marketing integrations into their SaaS: bottom-of-funnel tooling and offers, to track conversions from Meta ads and organic search. So I spent every day building analytics, watching traffic and search volume, and tailoring landing pages to search intent. That was when I started properly researching SEO and AEO, and how to get more traffic to a site for less money.

What I found instead

I’d like to say it shocked me. It didn’t. From my own experience I already knew it was true: the entire SEO game is changing right now, and has been for a couple of years, in a way that demands a complete paradigm shift.

Bot traffic isn’t all AI agents, and we have no reliable way to separate the two. But a ratio that keeps climbing can only mean one thing. Human behaviour is changing.

And that breaks the arrangement that made the web successful in the first place. It was never written down anywhere, but everyone publishing on the open web was working to the same three-step bargain.

Every step of that depends on the one before it. Google’s AI Overview quietly removed the second one. It answers the question on the results page, and the reader never arrives. Pew Research put a number on how often they do.

Here’s the part that matters, though. Consumers and business owners are still buying at the same rate. Nobody stopped spending. They have only changed the way they decide, and the tools they use to get there. Agents are doing more of the deciding. They gather the information, weigh it, and put an option in front of the buyer before they even realise what the alternatives were.

Once that landed, I invested properly in my own sites for the first time, writing content that would let an agent make the decision better than any human could.

Because if your content doesn’t convince the agent that you’re the best option, you will never get to convince the customer. They won’t see your product at all, even if you rank in Google, because their agent will already have chosen to present something else.

Which leaves two options. Stay the same, keep publishing for the ten blue links, and watch the traffic that funded the work get answered upstream by someone else’s summary. Or accept the new terms: build better products, and write the kind of content that models read, quote and rank, while the war is still being fought.

Web search will die

It’s inevitable, and I can see it in my own life. I almost never search the web for something I need. The whole sequence now looks like this:

1. 1I ask GPT to go and do the search
2. 2It comes back with a compiled report
3. 3I read the top three it named
4. 4I double check them, and compare
5. 5I pull the trigger

Nowhere in that sequence is there a page of ten blue links. That’s the world we have to optimise for now. It is the world we already live in, and businesses need to adapt to it.

So this site is the experiment

Can I, from first principles and study, build a content engine and a process that consistently gets myself and other people results in this new agentic world? I think so. agentrefs is where I find out in public, and I’ll be posting regularly as I go: the tests, the failures and the numbers, not just the parts that worked.