This time, the walled garden talks back

In a 2022 presentation for Yorba, I divided the internet into eras: dial-up, search, social. The fourth column said "Reset." This was where we would take our data back and renegotiate the terms.

Such optimism. Such a nice purple.

I have since changed that column to "AI." I used AI to redraw the graphic, too. My objections to the future apparently do not extend to doing my own layout, cause you know, time.

For 30 years, the internet has offered us a succession of helpful guides. First a portal, then a search engine, then a feed, and now an assistant. Each makes an overwhelming world easier to navigate. Each gives us a reason to stay. Eventually, leaving means abandoning something we've come to depend on.

Now the newest guide wants permission to act on our behalf.

Before we grant it, we should ask whether we can fire it without losing the life it has learned to manage.

In the 1990s, the guide came on a disc. AOL offered a manageable entrance to an enormous network, provided you could tolerate connecting through what sounded like a fax machine being interrogated. It was the original walled garden, and for a while, most people were happy to stay inside.

Then search became the guide. It helped us explore the wider web. Google could find the obscure page you wanted. With AdWords, launched in 2000, it also built a business selling advertisers access to people looking for things. We got directions. Our intentions acquired a price.

Then the feed took over. Social media made the internet personal. Friends, photographs, businesses and audiences gathered inside a few platforms. Leaving became complicated because the people we liked could not be persuaded to migrate simply because we had developed principles.

Meanwhile, the platforms collected information far beyond what people knowingly posted. A 2024 Federal Trade Commission report found extensive surveillance across major social and video services, including information about nonusers.

You could avoid the party and still end up in the photographs.

These eras overlap and compound. Search did not disappear when social media arrived. Neither will disappear into AI. We keep making another bargain without getting the previous installments of ourselves back.

AI offers a more intimate guide. An assistant can know the unfinished thought, the private doubt, or the reason behind the purchase. Given access to enough accounts, it can turn that knowledge into action. This time, the walled garden talks back.

The usefulness is real. So is the cost of rebuilding that relationship somewhere else.

And already, the companies offering to navigate the world for us are arguing over whose assistants may enter.

In September, Amazon blocked Meta's Muse assistant from shopping on customers' behalf. Amazon raised concerns about identification, credentials and consent. Meta says Muse stores credentials securely without seeing them.

Meanwhile, Amazon's own "Buy for Me" feature purchases goods from outside retailers. Amazon says those retailers can choose whether to participate.

The consent arrangements matter. But so does the commercial ambition: become the assistant that travels everywhere while retaining control over who enters your store.

The garden now wants a foreign policy.

It begins to resemble digital feudalism. The kingdoms negotiate access. We furnish the purchases, attention and personal information that make the territory worth fighting over.

They aren't arguing about whether you can leave. They're arguing about who gets to keep you.

A rival platform's agent is not necessarily our liberator. It may simply offer a more efficient transfer to another estate. An assistant can serve its provider's interests while speaking very warmly about mine.

There are legitimate reasons to block bots. Websites must be able to stop fraud, protect accounts and resist unwanted mass scraping. A user's request cannot excuse every action an agent takes.

But an assistant carrying out a specific instruction raises a different question from a crawler harvesting material for a model: how much say should I have over the tools I use to manage my own affairs?

The web's founders answered generously. In 1993, CERN released its web software into the public domain. Anyone could build on it. Openness was a deliberate choice at the foundation of the thing.

We need rules that make that choice meaningful. Agents should identify themselves, operate under limited permissions and leave records people can inspect. Access restrictions should address demonstrable risks, with scrutiny when platforms favor their own agents over comparable alternatives.

And leaving must become practical. Lawmakers should require usable ways to transfer the personal information and preferences that make an assistant useful. A download nobody else can use is a souvenir, not an exit.

Investors, too, should fund tools whose usefulness survives our decision to switch providers. Making people difficult to lose is an attractive business model. It is a miserable design principle for their lives.

I have a commercial interest in this argument. My company promises people more control over their digital relationships. We should be judged by the same standard. So should the founder who let AI redraw his slide.

The internet keeps asking us to make ourselves comfortable.

Before we unpack another piece of ourselves, I would like to see the door.

Less, please.

Chris