In Monday's film the transcript heard "Master Glentorian", and someone had to catch it. Every stage in that film is done by a worker, and the question I get most is what a worker actually is. So here is one, opened: the Transcriber, the worker that turns my recordings into words. 𝗪𝗵𝗮𝘁'𝘀 𝗶𝗻𝘀𝗶𝗱𝗲 It's a folder. These six files are what make it a worker: CLAUDE.md: who it is and its rules, in plain English. The first line reads "Transcriber takes your recordings and writes down what you said." sources.md: the folders where my recordings already live, one line each. vocabulary.md: the names I don't want misspelled, so they come out right every time. transcripts/: where the words land, ready for the writing worker. STATUS.md: where it left off. It reads this first and writes it last. skills/transcript-check.md: the check it runs before handing anything on. Rule one is "Nothing added." The rest is plumbing: a small script that does the transcribing, a setup file, a context note, and folders for recordings and output. No app and no dashboard. You can open every one of those six in a text editor. 𝗪𝗵𝘆 𝗼𝗻𝗲 𝗳𝗼𝗹𝗱𝗲𝗿, 𝗻𝗼𝘁 𝗮 𝗰𝗿𝗼𝘄𝗱 Each worker does one job, hands a file to the next and stops. Between them, I check. The film calls it four steps: Got. Made. Checked. Your OK. A crowd of agents talking to each other makes a better demo. I chose folders because when something goes wrong, I want to open one file and see where. The idea underneath is Jake Van Clief's ICM: the folder is the instructions, so the work is written down as it happens. 𝗪𝗵𝗮𝘁 𝘄𝗲𝗻𝘁 𝘄𝗿𝗼𝗻𝗴 The first Transcriber didn't take recordings at all. It started from text you had already pasted in. Then the next idea was to drag your recordings into its folder. Both were backwards: your files already have a home, and moving them around is the chore we were meant to remove. Now it's one line in sources.md, and it reads them where they already are.