Notes on building and measuring AI agents, from Ondřej Úlehla.https://ulehla.dev/enhttps://ulehla.dev/notes/mcp-tools-cost-1-8x-more/https://ulehla.dev/notes/mcp-tools-cost-1-8x-more/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×, and the gap only showed up once I asked for the exact number.Fri, 16 Oct 2026 00:00:00 GMThttps://ulehla.dev/notes/the-tool-result-is-the-big-number/https://ulehla.dev/notes/the-tool-result-is-the-big-number/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.Tue, 22 Sep 2026 00:00:00 GMThttps://ulehla.dev/notes/measuring-agents/https://ulehla.dev/notes/measuring-agents/A demo that works once tells you nothing. What pass rate, flakiness and cost per success taught me about building AI agents – and why none of it works without a clean room per run.Thu, 23 Jul 2026 00:00:00 GMThttps://ulehla.dev/notes/mcp-server-in-a-throwaway-vm/https://ulehla.dev/notes/mcp-server-in-a-throwaway-vm/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.Mon, 13 Jul 2026 00:00:00 GMT