{"@attributes":{"version":"2.0"},"channel":{"title":"Ond\u0159ej \u00dalehla \u2013 notes","description":"Notes on building and measuring AI agents, from Ond\u0159ej \u00dalehla.","link":"https:\/\/ulehla.dev\/","language":"en","item":[{"title":"MCP tools cost 1.8\u00d7 more than the estimate everybody used","link":"https:\/\/ulehla.dev\/notes\/mcp-tools-cost-1-8x-more\/","guid":"https:\/\/ulehla.dev\/notes\/mcp-tools-cost-1-8x-more\/","description":"I counted what the tool definitions of every server in the MCP registry cost in Claude. The common o200k estimate was low by 1.8\u00d7, and the gap only showed up once I asked for the exact number.","pubDate":"Fri, 16 Oct 2026 00:00:00 GMT"},{"title":"The tool result is the big number","link":"https:\/\/ulehla.dev\/notes\/the-tool-result-is-the-big-number\/","guid":"https:\/\/ulehla.dev\/notes\/the-tool-result-is-the-big-number\/","description":"Code mode comparisons count tool definitions. In my runs the definitions took about 2 percent of the window, and one tool call took all of it. What a call returns decides whether a task runs at all.","pubDate":"Tue, 22 Sep 2026 00:00:00 GMT"},{"title":"Measuring agents like flights","link":"https:\/\/ulehla.dev\/notes\/measuring-agents\/","guid":"https:\/\/ulehla.dev\/notes\/measuring-agents\/","description":"A demo that works once tells you nothing. What pass rate, flakiness and cost per success taught me about building AI agents \u2013 and why none of it works without a clean room per run.","pubDate":"Thu, 23 Jul 2026 00:00:00 GMT"},{"title":"Thirteen tools through a sandbox, and the race that hid them","link":"https:\/\/ulehla.dev\/notes\/mcp-server-in-a-throwaway-vm\/","guid":"https:\/\/ulehla.dev\/notes\/mcp-server-in-a-throwaway-vm\/","description":"A scanner can warn you about an untrusted MCP server. It cannot stop the code once it runs. What I learned from moving the server into a throwaway VM, and from the one bug that only a real cloud runtime could show me.","pubDate":"Mon, 13 Jul 2026 00:00:00 GMT"}]}}