AI agents mean the death of the web

Last week both Microsoft and Google held major developer conferences with one common theme: Tech giants have moved on from talking about chatbots and language models to focusing predominantly on AI agents.

AI agents are to some extent a shiny new thing that tech companies are dangling in front of our eyes, perhaps out of necessity because generative AI still hasn’t quite caught up with the hype. But it’s also a technology that can create real benefits, especially for automation in businesses if you see it as a natural evolution of technologies such as RPA. 

But just like when chatbots arrived, a lot of the marketing is about AI agents making everything easier for individual users, ordinary people, and above all, a lot of two conferences ended up being about the web and the things people do there.

Google seems to be going all in on its “AI Mode,” which it purports to be the future of web searching. In short, Google’s AI agents will Google for you, present you with information they think you want, or perform tasks for you like buying tickets or booking restaurant reservations. This is the first step toward tying everything together into a ubiquitous AI assistant that will handle all your planning and perform a ton of tasks on your behalf.

Microsoft, for its part, wants to lay the foundation for what it calls “the agentic web,” which is a bit of the same thing but more focused on infrastructure. The company is implementing protocols such as Model Context Protocol (MCP) and Agent2Agent, which enable AI agents to communicate with one another, and it has launched the NLWeb protocol, which will enable any website to have its own AI functionality and become compatible with any agent that wants to use it.

The goals here are obvious: Web browsing and online tasks are not something that you should have to do; everything should be taken care of for you by AI agents.

An AI-mediated vision of the future

The problems are also obvious, but the tech giants don’t seem to want to acknowledge them. One problem that will not be solved is that generative AI models still cannot be trusted. They are good at many things and can be disciplined for specific tasks through training on very specific data, but for general knowledge the error rate is far too high. Unfortunately, it seems a bit like the suppliers of these models think that’s how it should be; so now we move on.

Another “problem,” if you want to call it that, is the question of whether people even want this. The more new tech companies present, the clearer it becomes that development is driven by what is possible, not by what is in demand. My impression is that people in general are already getting tired of AI being pushed everywhere, but what we have seen so far is apparently not even close to what is to come. 

I personally find it quite fun to plan my vacations and would never dream of letting an AI decide what pants I should wear or what gift my mother should get for Mother’s Day (even if it’s flowers). For example, I saw someone writing enthusiastically about an AI agent being able to plan his daughter’s birthday party. Isn’t that what you, like, live for?

Anyway, perhaps the biggest problem when you look at the web and the digital economy is that these solutions somewhat assume that all websites, all content, all services and all products are available to agents just as they are available to regular users today. But then you completely ignore why the content and services are there in the first place. 

For example, a newspaper is online to reach its readers, sell subscriptions or advertisements, and create relationships between the newspaper and the hopefully growing readership. In the same way, other selling companies are online to sell their products but also to build brand loyalty and long-term relationships with customers to create repeat sales.

Now, instead, the content and products will be consumed by AI agents in the first tier, to potentially reach customers in the second tier. The direct relationship between company and customer disappears. Control over how to commercialize your content or services is handed over to someone else. The incentives to even have a presence on the web are drastically reduced.

It can be pointed out that this is already a reality, mainly in terms of Google Search and the large marketplaces. But companies’ dependence on being visible on Google is double-edged to say the least. Google has certainly enabled a spread and visibility that for many companies has meant everything for their business. At the same time, there is hardly any company that really wants to be dependent on Google — direct customer relationships are always preferable.

Now, however, that dependence is becoming even greater, and of a different kind. Over the years, companies have learned how to spend marketing money on things like SEO so that customers will see you. Now that marketing will be done towards AI agents instead? Because apparently it is the AI agents who will make purchasing decisions for customers? Exciting.

Existential crisis ahead

The enormous Google dependence is also why it will be almost impossible to “say no” to this. On the content side, Google has already informed publishers that they can certainly deny the AI ​​bots free access to all content on newspaper sites. But that would mean that Google would stop indexing the sites altogether — saying no to AI means saying no to being visible on Google. This in turn means that only those media companies with enough of their own muscle will have the resources to even exist.

Similarly, companies that sell products and services will be forced to accept doing business with AI agents, because otherwise the AI ​​agents will simply make “their” purchases from a competitor instead.

If you look into the oracle, it’s easy to see gigantic antitrust investigations here, because how do you ensure free competition when deals are made by AI agents in an AI-controlled market? It will undoubtedly be interesting to know.

You may like the visions that are being painted, or not, but it is probably fair to say that the “agentic web” means the death of the web as we know it today. Because what Google and Microsoft see ahead is something completely different — not just on the surface but fundamentally.

Paradoxically, this also leaves an opening to bring the web back to life — the old web, the one we had before first algorithms and then AI models took over. 

A web that was built on human-created and -selected content, on personal recommendations, on direct relationships between companies and customers. Where digital was the connection, the glue, the enabler, and not the product itself.

Of course, that would require more from users than the fully automated AI world promises, and it’s unlikely to happen. But if you can put on your nostalgic glasses and dream a little, wouldn’t that be pretty wonderful?