Vibes per Token is my build-in-public journal. Almost every night I write up what I actually shipped that day, what broke, and what it cost, from real work on production AI, agents and automation. No hype, no course to sell. Two channels, English and Russian, plus a few-times-a-week digest of AI news that is genuinely worth your time.
Two things, most weeks.
A short field note from the day: the thing I built, the bug that ate an hour, the decision I would make differently. Written the way I would tell a colleague, not a press release.
The handful of AI and automation stories actually worth reading, filtered and checked so you skip the noise. Sent when there is something to say, not on a schedule for its own sake.
Pick your language and open the channel.
Build-in-public, dry and to the point. The daily journal and the digest, in English.
The same journal, written natively in Russian and a little livelier. Not a translation.
Same voice, longer form, on the blog.
Not the hype. The honest arc of using an AI assistant every day, from five days in to three months in, two hundred hours in, a year in. The wins that were real, the ones I imagined, and the dip nobody warns you about.
Read the post →While I sleep, routine work gets done. Sounds great until something goes wrong at 3am with nobody watching. Here's the boring part that makes overnight automation actually safe to rely on.
Read the post →Most AI demos look impressive. Most AI agents fall over in production. Here's what I learned building Tree AI, a personal multi-agent system I've run every day for over a year, about the gap between the two.
Read the post →An automated daily report I rely on was confidently wrong for five days, and nothing flagged it. The bug wasn't a crash. It was worse. Here's why silent failures are the dangerous ones, and the ten-minute habit that catches them.
Read the post →