A tech digest, Monday, Wednesday, and Friday.
Automata reads 34 feeds three mornings a week and writes one page about what happened. It exists because keeping up with software and AI had turned into skimming a dozen sources badly rather than reading one thing well.
Plural of automaton, from the Greek αυτοματον — acting of itself. It carries three senses and all three fit: the self-operating machines of the eighteenth century, the finite and cellular automata of computer science, and the plainer modern reading of something that runs without a hand on it. This page was assembled by a process that woke on a schedule, read its sources, and wrote what you are looking at.
A scheduled job fetches every source, drops anything older than seven days or already sent, and hands the rest to Google's Gemini with one instruction: pick what genuinely matters to a working engineer, merge the same story arriving from several places into one entry, and say plainly what happened.
The model is told what not to do at some length. No “rapidly evolving landscape”, no “delve”, no “key takeaways”, no emoji, no bolding half a sentence, no lists of exactly three parallel items. Each summary is one factual sentence. A “why it matters” line has to name a concrete consequence or it gets cut, because a vague one is worse than none.
Dates come from the source feeds rather than the model. Asking a language model to report a date invites a plausible wrong one, which would defeat the point of showing it.
The email is the curated cut, trimmed to fit the hundred kilobytes Gmail allows before it clips a message and hides the rest. This site is the complete edition: every item in the window, a why-it-matters line on each curated story rather than only the leads, filtering by source and category, and every past edition at its own permanent URL.
The email rotates through three typographic treatments so it never goes visually stale, and arrives at 8am Eastern on Monday, Wednesday, and Friday — correct year-round, which takes more care than it sounds like, since scheduled jobs run on UTC and ignore daylight saving.
LinkedIn is absent. No API permits reading a personal feed, and scraping one would mean handling session cookies and breaking their terms. The engineering blogs cover most of what circulates there.
Anthropic publishes no RSS feed, so a community-maintained mirror stands in and sometimes lags. A handful of company engineering blogs were dropped for having no working feed at all.
When the curation step fails — a bad key, a quota, an outage — the digest falls back to ranking by source and engagement and says so at the top. A plainer edition is acceptable; silence is not.
Python, no framework. The pages are plain HTML written to disk; the filtering is about twenty lines of vanilla JavaScript. Set in Georgia. Built with Claude Code.