Status: Done

At MCP Dev Summit North America 2026 (April 2-3, NYC) I gave a talk on what I call the tool abstraction problem. Before MCP existed, my team and I at Arcade had already shipped over 1,000 tools - first as native Arcade tools with our own Open Execution Protocol, and eventually adopting MCP.

The lesson: the hard part isn’t writing the code, it’s finding the right abstraction. Most MCP tools today are thin wrappers around APIs - GET /users/{id} becomes get_user(id). But LLMs reason about tasks (“find the customer who complained last week”), not endpoints. Build tools around tasks and intents, obsess over description quality, and you get dramatically better selection accuracy and fewer runtime failures. It’s the discipline I think of as the machine experience - UX, but for the model.

The talk was also released as a standalone video on the Agentic AI Foundation channel:

The standalone release of the Tool Abstraction Problem talk on the AAIF channel.

For more about Arcade check out the website at https://www.arcade.dev/ and the GitHub https://github.com/arcadeai/