Ivan Plankin

AI agents do the work. I set the rules and own the result.

I design systems where agents write code, run Telegram channels, check each other's work and fix what breaks. How they are built and what counts as done is my call.

Over the last 7 days the agents closed 985 tasks and worked 168 of 168 hours.

The rules they work by

Right now

updated 60 s ago
  1. 1queued
  2. 9in progress
  3. 5in review
  4. 54done today

In progress

  • Writingassigned by merunning 3 min2 steps
  • Repairassigned by merunning 7 min22 steps
  • Repairassigned by merunning 8 min16 steps
  • Repairassigned by merunning 22 min63 steps
  • and 5 more in progress

Recently finished

  • Repairstarted by the system4 min ago
  • Repairstarted by the system1 h 47 min ago

A step is one agent action: a command, a file edit or a check.

Rules

I write the working rules for my AI agents. These six are the quickest way to understand me.

  • A system has to work when I'm not around.

    If it needs me to remind it, it was designed badly.

  • When money, reputation or something irreversible is at stake, a person decides.

    The machine decides the rest on its own and shows what each decision rests on.

  • Done means checked where it is used.

    The result is what appears on the screen of the person it was made for. A finished build and green tests only get you closer to it.

  • A silent failure is worse than a loud one.

    If something is skipped or not done, it has to leave a trace that someone will notice.

  • A bug gets fixed together with its cause.

    One bug usually points to a whole class of similar ones. Fix only that one and it comes back somewhere else.

  • Whoever checks the work did not do it.

    A reviewer looks for mistakes and does not fix them. Otherwise they start defending their own solution.

Projects

What I designed and launched. Some of the code is open on GitHub.

Contact

Telegram is the easiest way to reach me.

Or leave a message here