
llms.txt and MCP feel like the beginning of a new kind of protocol stack — one designed not for humans, but for AI.
Just like HTTP and TCP/IP made the web readable and writable for people, these new standards are doing the same for agents. They're giving AIs a shared language to understand what tools are available, what they’re allowed to do, and how to carry out actions on our behalf.
As more of these conventions emerge and gain adoption, the network doesn’t just grow — it evolves. We unlock a layer of interoperability where agents can talk to each other, delegate tasks, and orchestrate workflows across APIs and services. That’s when things get really interesting.
We're not just shipping apps anymore.
We’re laying the groundwork for an AI-native internet.
As the stack matures, we’ll need new protocols for key infrastructure:
Authentication, payments, capability negotiation, and trust.
Because when agents act on your behalf, security and alignment aren’t features — they’re fundamentals.

Why good retrieval still produces bad answers, and where most systems actually fail
AI
Relevance, diversity, reranking, and why most systems return the wrong things for the right reasons
AI
Chunking, dimensions, and the knobs that decide whether retrieval works
AI