An abandoned Tumblr blog sent me down a rabbit hole recently. It had been rediscovered on TikTok and was being passed around by people who *might* just have been in primary school when its owner last posted. Somewhere between looking through it and wondering what had happened to that whole way of being online, I remembered being 17 and building my own site on GeoCities.
For anyone who missed that particular slice of internet history, GeoCities gave you a free homepage in a ‘neighbourhood’ you chose according to what you wanted to put on it. Music sites lived on SunsetStrip; science fiction fans could move into Area51. Your site had an address, and the sites around it belonged to other people who were interested in similar things. There was somewhere to go after you’d finished reading. You could have a wander. For the uninitiated, visiting the archived neighbourhoods will give you a sense of how it worked.
At 17, I was more interested in having a website, than in worrying about how someone might find it (oh the pre-algo glory days! the freedom!). And looking back, I think there’s something in the way GeoCities worked that we’ve spent a long time trying to replace, with varying degrees of success.
And, somewhat unexpectedly, thinking about AI agents has brought me back to this idea.
NB: This is very much a thought experiment. I wholeheartedly reserve the right to change my mind, particularly once someone who knows more about building this sort of thing points out the holes. But I’ve got far enough with the idea that I’d like to put it somewhere other people can pull it apart. So this is your invitation, dear reader, to pull apart my vision below.
What happens when the visitor is an agent?
I’ve been following the work around agent experience, or AX, pretty closely, which asks how websites and services should work when an AI agent is using them on someone’s behalf. Netlify, the web hosting company, has been advocating the idea and bringing people together around it. The aim is to make it easier for agents to use digital services to carry out the tasks we’ve given them. The AX project explains the approach here.
An agent booking a hotel needs accurate information about availability and a reliable way to make the booking. The lovely photograph across the homepage does less for it, than it does for you or I. A website can serve both of us, of course, but if I outsource my travel booking to my agent, I may never visit the site myself.
That has consequences for the people whose businesses depend on those visits. Search has supported a fairly familiar arrangement: a service (e.g. Google) reads your website, helps people find it, and sends some of them your way. Those people might see an advert or buy something. If an AI service can answer their question using your work, there’s less reason for them to come to you.
Cloudflare (which provides infrastructure for websites) has been making this argument publicly. In July 2025 it announced changes to block AI crawlers by default unless permitted, alongside plans to help publishers charge AI crawlers for access. Its argument is that AI companies are taking content while sending very little traffic back to the people creating that content. Cloudflare’s announcement sets out their reasoning here.
I don’t think this means people will stop visiting websites, or that advertising will disappear. But it does make me wonder what happens to the parts of the web we’ve built around the goal of ‘getting a human to click through to this page’. If more of that activity happens between machines - if the utility of a webpage now needs to be optimised for agents, what do we want the places we still visit ourselves to be?
I miss having somewhere to wander
There’s an appealing version of this where the machinery built to attract clicks becomes less useful, and people get back to writing things they care about. I would quite like that version - no more clickbait headlines, no more hot takes purely for the sake of engagement... It’s exactly the sort of internet that suits someone writing about how much she misses personal websites.
The issue with this internet utopia though, is money. Today, advertising often helps pay for work that matters, including work its readers couldn’t afford to fund themselves. Losing that income could leave us with fewer people able to spend time writing. A web populated entirely by people who can afford to do it as a hobby would have some fairly obvious gaps, and paid subscriptions can only stretch so far.
But the way we’ve funded publishing has also shaped what gets published. Anyone who’s worked in marketing will recognise the hours spent adjusting headlines and checking which version brought in more traffic. I’ve spent plenty of my career thinking about how to get people to read something. I don’t begrudge anyone doing the same. I do wonder how much writing ends up shaped by the requirements of ‘being found’.
People still build personal sites, of course. Neocities gives people space to make their own websites, while Bear Blog offers a deliberately simple way to publish. I’ve tried different blogging platforms over the years - cool projects from indie hackers, like Feather and Nicheless, that make writing simple or joyful. The wish to have a small corner of the internet that feels like yours clearly hasn’t gone away. Finding those corners, and then finding another one you might like (AKA ‘discovery’) is why we’ve seen people gravitate to algorithm-driven aggregators.
Another blast from the past
RSS ought to help. It stands for Really Simple Syndication, and it lets a website, podcast or whatever, make new posts available through a feed to anyone that wants to follow it. A reader app brings together the feeds you subscribe to, so you can follow lots of sites without visiting each one to check whether anything’s changed.
I’ve tried RSS readers. And, well... I like the idea of them much more than I like the experience of using them. The unread counts make me feel as though I’ve acquired another inbox, and reading everything through the same interface loses some of what made each site interesting to visit. More practically, I still have to find the sites I want to subscribe to in the first place.
Social feeds are very good at giving us something to look at immediately. We’re quite literally fed a never-ending supply of content. An empty RSS feed reader requires homework. And while there are readers with discovery features, I haven’t found one that gives me the feeling of following-a-link-from-someone’s-blog-and-ending-up-somewhere-surprising.
Bluesky is doing interesting work here. Its custom feeds let people choose different ways of finding posts, including feeds built by other members of the community. But in its January 2026 plans, the team still identified finding people to follow and improving ‘discovery’ as things it needed to work on.
So I’ve been wondering whether we could borrow something from GeoCities. Keep the freedom to build a site wherever and however we like, and give those sites a way to belong somewhere, together.
Moving into a neighbourhood
So let’s say you’ve built yourself a website to share whatever-the-eff you want to share. My idea starts with adding a small public file to your website, saying which ‘neighbourhood’ you’ve chosen, and where your RSS feed lives. A person writing about food could put their site in the food neighbourhood. Their website would stay exactly where it was, with its own design and domain name, but it would become part of a collective of sites writing about food.
I want the writer to choose that neighbourhood. We already have plenty of software making assumptions about what we’re interested in and where we belong. There seems to me to be something useful in making that decision ourselves, even if the category we choose is broad or we later decide to move.
Software (like crawlers) could then read those files and build an index of the sites taking part. Within each neighbourhood, it could look at the writing and suggest which sites have something in common. A food writer might end up near someone growing unusual vegetables, who in turn connects to someone writing about the history of their allotment. I could start with something familiar and follow my interest a little further each time.
For the technically curious, the declaration could sit at a well-known URI (a predictable address where software knows to look for a particular file), and the index could use embeddings (which represent aspects of a text’s meaning as numbers so software can compare them; which is the same fundamental technology that tools like ChatGPT use). The well-known URI convention already exists. In ordinary use, I’d want the experience to be as simple as choosing a neighbourhood and finding out who else was there - wandering the virtual streets, as it were.
The software’s idea of similarity would be imperfect, admittedly. Two people can write about the same subject and disagree about almost everything. Deciding what makes a useful neighbour would need a lot more thought than I’ve given it here but I think it’s an interesting place to start.
There are two choices I’d want to build into this ‘internet neighbourhood’ from the start.
The first is that anyone could build an index from the public declarations. Writers wouldn’t need an account with a particular discovery service to take part, and readers could switch services without asking every writer they follow to move with them. I’m a big believer in the need for a decentralised web but recognise that much of web3 use at the moment isn’t particularly easy to navigate to the average internet browser.
I’d also make publishing the list of participating sites a condition for indexers joining the system, so someone starting a new service wouldn’t have to find everyone from scratch.
Writers should also be able to declare that they wanted particular indexers to leave them out. Making sure services respected that choice would still require work; putting a preference in a public file doesn’t enforce it. But I’d want ‘joining’ and ‘leaving’ to remain decisions the writer could make.
The second choice is the one I most crave on a personal level - some mechanism that lets me decide how far from my usual reading to wander. Some evenings I want more of a subject I’m already absorbed in, but other times I’m bored of my own interests and would quite like to encounter someone else’s. A reader could let me adjust that ‘distance’, before showing me suggestions of what neighbourhoods or sites to explore.
Would we actually share a neighbourhood?
This is where I start arguing with myself. If every indexing service arranges the sites differently, and every reader adjusts what they see, how much would we have in common? I could call it a neighbourhood and still end up with something that behaves very much like a personalised feed.
I think my starting point would be a shared set of named neighbourhoods, with public membership that everyone could browse. The recommendations could vary depending on our preferences, but we’d both be able to look around the same neighbourhood and see everyone who’d chosen to put their site there. I might take a different route through the food neighbourhood from you, while still being able to visit the same collection of sites.
This thought experiment leaves plenty unresolved. Someone has to decide how the categories change, and what happens when people disagree about them. Someone will 100% try to fill a neighbourhood with spam. Making the address list available to everyone would help a competing service get started, but one company could still become so convenient that almost everyone used it. I’d want to make leaving possible, and then see whether people actually felt able to do it.
There’s also a possible connection to how writers get paid. Really Simple Licensing (RSL) is an open standard for publishing terms that machines can read, including terms for attribution and payment. A neighbourhood declaration could point to those terms, so a service finding a site could also find the conditions its writer had set for machine use. The RSL specification describes what’s possible and I think it fits nicely with what I am imagining.
That would still leave the considerable business of getting companies to agree to those terms and pay. But I like the possibility that a writer could make their work easier for people to discover while also saying how machines may use it. (There’s another article in that, but I’m trying quite hard to keep this one to a single idea here!).
Who looks after the neighbourhood?
There was something else in GeoCities that I haven’t accounted for: GeoCities’ Community Leaders were volunteers who helped people with their sites and took an interest in what was happening in their neighbourhoods. You can still find traces of those groups in their archived pages.
I’ve spent most of this piece thinking about how software might help people find each other, but those volunteers make me wonder how much of what I miss depended on someone doing the unglamorous work of making a place feel welcoming. They might recommend a site they’d enjoyed or help a newcomer who couldn’t get something to work, and over time people would get to know them. I’d be interested to know how much difference that made to whether someone kept returning to a specific neighbourhood.
There are people doing that work in online communities now. It takes time, and often the people who care most end up doing far more than their share. I don’t have a good answer for how this version would support them, or whether it could work without them. I suspect a collection of well-organised websites would still feel fairly empty if nobody seemed to be around.
I’ve written up this idea as an open spec on GitHub. It sits firmly in the drawer marked interesting, alongside the RSS reader I’d quite like someone to make for me. I’m not building it, but I wanted the idea somewhere people could respond to it in more detail than a comments section allows.
And I know a proposed file format gets us very little on its own. There would need to be a reader worth using and an index with enough interesting sites to make opening it worthwhile. The first few people would need a reason to turn up before there was much to find. Those are substantial problems, and nostalgia alone won’t solve them.
But the thing I keep coming back to is how much I used to enjoy coming across someone else’s little corner of the internet. I’d like more of that again. And don’t come at me with any Instagram ‘communities’ nonsense - algorithm based platforms incentivise engagement and homogeneity rather than true expression. But, if agents take on more of the things we currently go online to get done, perhaps we could put some thought into the places we’d visit just because we wanted to be there.
Would a good map be enough to bring us back? Or would we need to know someone was there to welcome us? And is anyone out there doing something like this that I should check out?