Web APIs Are Broken, So How Do We Fix Them?

Web APIs are increasingly inadequate for modern AI-driven applications due to tight coupling between clients and servers. Originally designed for deterministic systems, traditional APIs struggle with autonomous AI agents that operate probabilistically and require runtime context and tool discovery. Three potential solutions are explored: leveraging reusable media types via AI protocols like A2A and MCP so agents can self-discover formats; adopting an Agentic API specification that replaces CRUD with intent-driven methods like SUMMARIZE and CHAIN; and embracing HATEOAS, which enables loosely coupled, self-descriptive APIs with embedded hypermedia links. The consensus is that future APIs must be loosely coupled, semantically rich, and dynamically discoverable to support the estimated 1.3 billion AI agents expected by 2028.