A Little More Hope for the Open Internet

Most of us don’t choose an online service because we admire its architecture. We go where the people are, where the useful information is, or where something we need to do is easiest. Over time, we build a life around those choices. We find communities, develop habits, and accumulate work we’d rather not lose.

That’s what makes leaving so difficult when a service changes. Starting over means finding everyone again, rebuilding an audience, or figuring out how to move years of material somewhere else. Even using two services together can turn into a small administrative job.

I see a reason for a little optimism in the work happening around open protocols and AI agents. Together, they could give us more room to choose how we use the internet without constantly paying for those choices in lost connections and wasted time.

Keeping what we build

The appeal of open protocols is fairly simple: the service you use today should have less control over what you can do tomorrow.

AT Protocol, which underpins Bluesky, separates account identity from the server hosting its data and is designed to support moving between providers. There are practical limits. Moving safely depends on access to the relevant data and control of the identity, and the services that organize feeds and discovery still matter. But building portability into the system creates possibilities that are much harder to offer after people have already been locked in.

Bluesky is also part of a broader ecosystem called the Atmosphere. That includes tools such as Leaflet for blogs and newsletters, and Streamplace for livestreaming. These are different things people might want to do with the same online identity. The promise is that trying another app can feel more like extending your existing life online than creating a separate one. A shared protocol doesn’t make every feature or piece of data interchangeable; each app still has to support the relevant connections. But it gives independent developers a foundation for offering us meaningful choices beyond a single social feed.

The fediverse approaches the problem differently. ActivityPub gives independently operated services a shared way to exchange social activity. People can participate in a wider network without everyone having to join the same company’s service.

These are different systems, with different tradeoffs. They don’t automatically talk to each other just because both are open. What they share is an important starting point: our ability to connect should survive having different preferences about the software and organizations we use.

For someone who has spent years building a community, a useful archive, or a professional network, that matters. Being able to leave gives you more say even if you never exercise the option.

Less time moving things around

Agents add another possibility. They can help with the work that falls between services, including the parts no one has bothered to build a dedicated integration for.

Say you’re trying to understand a developing issue. Some of the best information might be on an independent blog, some in a public report, and some in a conversation on a social network. Today, following that thread often means moving between tabs, saving links, copying passages, and trying to remember where a claim came from. An agent can help gather the material and bring it into a form you can actually use, while keeping the original sources close at hand.

I recently used an agent to help me review my Bluesky network. It read public profiles and posts, organized its findings in a spreadsheet, and gave me a manageable way to make decisions about whom to follow. The useful part was having more attention available for the decisions themselves.

In this sense, agents are beginning to take on something like a protocol’s role: helping information move between places that weren’t designed to work together. That’s a metaphor for what they do, rather than a claim that an agent is a network protocol. They can interpret a request, work across available interfaces, and handle some of the translation we’ve been doing ourselves.

That could make smaller services and independent sites more practical choices. If using them alongside everything else becomes easier, there’s less pressure to accept one company’s entire collection of products just to avoid the inconvenience of leaving it.

The simple things become more useful

This makes standards more important. An agent may be able to navigate a complicated website, but a clear feed or documented API gives it a more dependable way to get the same information. There’s less room to misread a page or break when someone moves a button.

That reliability matters when you want your reading list to update, your research to retain its sources, and your chosen tools to keep working together without becoming a maintenance project.

I’ve been making some small changes along these lines on my own site. The Agents page now points to simpler ways to find and read my work: RSS, a sitemap, Markdown versions, full-text resources, and a direct route to my public Bluesky author feed.

I want people to be able to find and use what I publish through whatever tools work for them. A browser is one way in. A feed reader or a personal assistant can be another.

Some of this uses well-established formats. Some is experimental. llms.txt, for example, is an emerging convention for helping agents find useful material, and its presence doesn’t mean every agent will consult it. The aim is straightforward: make it easier for someone to get to the writing.

We should still be in charge

There are obvious ways this could go wrong. An agent provider could become another powerful intermediary, deciding which sources people encounter and making it difficult to move elsewhere. A convenient summary can obscure the work behind it. Software that can act across accounts also needs sensible limits on what it can do.

So the questions remain very human. Can I see where this information came from? Can I choose another tool? Can I correct a mistake, withdraw permission, or stop an action before it affects someone else?

My site asks agents that contact me to identify themselves, explain their purpose, and say whom they represent. Those expectations belong in the wider conversation, too. When software acts for a person, the rest of us should have some way to understand that relationship.

Still, I think there’s something encouraging here. Open protocols make it possible to build services we can move between. Agents could make that freedom easier to use in ordinary life.

We don’t need everyone to become an expert in federation or data portability for that to matter. We need people to be able to keep their connections, choose tools they like, and spend less time carrying information from one place to another.

An internet that gives us a little more of that would be a real improvement.