The Great U-Turn: Why AI Walled Gardens Will Break the Legacy Web
As major players like Anthropic and OpenAI build walled gardens and ban open-source clients, heavily subsidized Chinese models are poised to dominate the API market. Meanwhile, the internet faces a massive architectural U-turn. After years of building defenses to block bots, companies must now completely redesign their data layers to serve AI agents. The shift from SEO to Answer Engine Optimization (AEO) will leave legacy platforms behind, creating a massive opportunity for new players to build agent-first interfaces from the ground up.
There is an interesting change happening in the open-source and AI space. Everyone knows about OpenClaw, right? And we know that Anthropic is pretty much banning API accounts that they identify as OpenClaw users. They don't like their services being used by unpredictable open-source applications - in this case, OpenClaw.
Similar moves are happening with Gemini and ChatGPT, because everyone wants users to stay on their own platforms. You can see Claude Code going after OpenClaw's features, with dispatch and channels. The enterprise is once again trying to shut it all down, to build those locked-in platforms and walled gardens - something we have seen before with APIs. They want to preserve the status quo.
In this case, I think there will be a bigger push towards open-source models that will always work with your own instance of OpenClaw, or whatever comes next. At the same time, I believe the Chinese players won't miss their chance to capture even more market share. The moment you start using their models in OpenClaw, they first get access to a lot of data. On top of that, they get market penetration, and they can subsidize it in the very beginning as they did with pretty much every product they have been building. So these cheap models will be there for you to use on the API level, not just via local hosting.
Just like the Xiaomi's recent model and many other examples, they will be dirt cheap. This will be their foot in the door. It might be a massive opportunity for Chinese players to distribute their models, harvest more data, and create better products that are simply cheaper. And since people don't really care that much about cybersecurity anymore, the data will flow. It will flow to China simply because the major players like ChatGPT, Claude, and Gemini are trying to create walled gardens.
This also resurfaces the very same question we had a couple of years back when ChatGPT was released. We built the entire web to block the bots. We had scrapers, proxy services, fingerprinting, and shifting UIs to prevent you from taking screenshots. We had many things that were built to obfuscate the backend infrastructure from everything on the surface level.
And now, these companies have to make a rapid and dangerous U-turn and rework their entire database layer to resurface much more data for the agents. If we assume these agents will do grocery shopping for us, travel bookings, maintenance work, and so on, many of those applications will have to figure out how to monetize the whole thing. The primary interface will now belong to AI agents. The very bots that we were supposed to be tracking and blocking are now the key to success. The companies that open up and figure out how to monetize this new paradigm will win the game.
Bigger players who won't be able to shift fast and some of them might lose. For example, Answer Engine Optimization - the new term replacing SEO - will rely heavily on websites optimized for LLMs and crawlers, with extremely easy and fast access to data. We can expect new e-commerce players, new websites, and new SaaS ideas coming to the market that will be overly optimized for LLMs. They might even be boring in a way, or they might have separate interfaces on the API level, on the MCP level, that will be simplified and fast to load, with a completely different set of data and commands available.
For example, API endpoints won't be that atomic anymore. They will have more telling names, with a bigger set of descriptions. Some API endpoints won't land in the MCP, and some data will be made available in a more convenient way for LLM crawlers and agents to access. So there will be another layer built specifically for the agents.
This is a chance for fresh e-commerce players, for pretty much every website and everything on the web these days. It is an opportunity for new players to build from the ground up - from scratch - a totally new set of interfaces. To come up with a new vision, a new approach, and a new standard to the web that will account for the new type of user: a dummy agent that requires fast, clear, and highly available data. This data shouldn't even be visually present on the page anymore. There is already this agent-to-agent protocol proposed by Google, and there will be more of those.
Soon enough, there will be separate interfaces created exclusively for AI agents. There will be a separate discipline of cybersecurity experts. The shift will be massive. The investment required to get this data open again is massive. It will take months and years for some companies to get this in order. And this means that whoever gets it fast will win this race again.