As the boundaries between apps melt away
There’s a good chance the Internet will be communicating in natural language vectors in the future because that’s how LLMs talk. It’s equally likely most of the communication will be AIs talking to AIs. In such world, APIs will gradually be only used for efficiency. Eventually most databases disappear as it doesn’t matter whether data is structured or unstructured. The application-centric model will come to an end. I’m sure there will be many steps in between but it helps to think in extremes to make the edges visible.
You will likely have your own AI-client or a number of clients that you use. The client presents you the data regardless of where it resides assuming you have access rights to it. There won’t be a distinction whether it’s coming from your own files, your friends’ AI, sensors around and in your body, or from the wider Internet. It just seamlessly flows in your client, parsing the data for you to interact with or to consume. The boundaries between apps will melt away.
Most data around us will just flow and be AI-mediated to our benefit. Model Context Protocol (MCP) by Anthropic is the canary in the coal mine. What I’m the most curious about is how the social web will emerge in the new application layer once the MCP matures and the guard rails get built.
Will we collaborate with our friends and their AIs? How does teams of people and AIs work together? Who’s instructing who? Are the private data sources shared among groups of trusted people for collaboration? Where will the value creation shift?
Open source, open boundaries
The classic worry is that there will be a handful of companies who own the leading LLMs and become the new gate keepers to the Internet we experience much like Meta (Facebook), Apple, Amazon and Google are dominating their closed silos of today’s Internet. Despite the worries, I’m hopeful that the open Internet will prevail. It seems that it’s harder to gate your data when its usefulness comes from being able combine it with the machine intelligence in multiple contexts. The value comes from understanding data across your life, not just for example in a siloed feed of photos. Open Source language models (or open weights) has reminded us how powerful Open Source can be in shaping the way we use our tools. DeepSeek, Llama and many others are forcing the industry to stay open. Apple’s closed infrastructure around iPhone might be the biggest worry in the horizon shaping up how most of us will use AI in the future.