Posts

I built an MCP server that searches flights across flexible dates and multiple airports, then used it for an actual trip I was planning. Three tool calls in, it had produced a correct answer, and along the way had wasted most of a day’s API quota and told the agent several things that weren’t true. None of the four problems was a crash. Every one of them was a shape problem: the output was well-formed, plausible, and misleading. That’s the failure mode that matters when your consumer is a language model, because an agent can’t smell that a number is wrong. It acts on whatever you hand it — and when the number is wrong in an expensive direction, it acts again. ...

WhatsApp Coexistence is the feature that lets a business keep using the WhatsApp Business App on a phone while the same number is also connected to the Cloud API through a Business Solution Provider (BSP). Messages mirror in both directions, and the API side receives a webhook event — smb_message_echoes — every time a human replies from the app. That echo event is gold if you’re building a bot that shares a number with real staff: the bot can detect that a human has stepped into the conversation and shut up. I was building exactly that — an order-taking bot for a client whose employees answer the same WhatsApp number all day — and the handoff logic hinged entirely on those echoes. ...

I spend my time building three kinds of things: computer-vision systems that run on cameras bolted above truck lanes, automation pipelines for small businesses that live entirely inside WhatsApp and spreadsheets, and open-source tools for AI agents. All three produce the same byproduct: hard-won lessons that only exist in private design docs and postmortems. A validation split that quietly flattered every metric. A “camera focus problem” that turned out to be a missing wire in a config. A power issue fixed by a shorter USB cable. ...