I run several software projects with AI agents, on a system I built so that someone without a coding background can ship real work. I am not an engineer. I work in Level 2 technical support.
I started working seriously with AI less than a year ago. Before that I used it the way most people do: research for videos, images, small things. In March 2026 I set my other creative projects aside and went all in on using it for serious work.
I did not come from software. I studied a little HTML and CSS years ago, nothing more. What I have instead is a system: it records every decision, it keeps several projects moving in parallel, and nothing goes out into the world without my sign off.
I use Lavish, Kun Chen's editor for HTML artifacts, almost every day. You open a page in it, click any part of it, and leave a note there for your AI.
The notes were text only. An image is worth more than a thousand words, so I asked for image attachments. Then I built it, and Kun reviewed it and merged it into his own project.
- The request: issue #103, opened 2026-06-24
- The work: pull request #188
- Merged on 2026-08-14 by the maintainer, @kunchenguid
I file precise bug and security reports against tools other engineers use every day.
- NousResearch/hermes-agent #90449: a stream watchdog kills long reasoning responses before the first token arrives.
- pingdotgg/t3code #4901: the Android app disconnects right after getConfig. Fixed.
- pingdotgg/t3code #5742: the mobile thread list goes stale over a churning remote connection.
- kunchenguid/firstmate #1637: a required CI gate certifies itself from a payload the pull request author controls.
- kunchenguid/no-mistakes #652: after an intentional history rewrite, a branch is left permanently unrunnable.
The system does the work I cannot.
Imperare sibi maximum imperium est.