114,781 impressions in thirty days, no agency

The posting system I built for myself, running every day since February

114,781impressions in thirty days
16posts, in that window
918lines of checks that can block a post
20new connections a day
Client
Bon Labs, my own account
Delivered
February 2026, runs daily
Read time
2 min
Reference
Open my profile and read it
StackPythonMarkdown as dataZernio APIMCPHeadless ChromeTelegramGit

Who

Me. Bon had no audience and no ad budget, and I needed my market to know I existed. So I built the whole thing as software, and ran it on myself before I sold it to anyone.

The problem

Posting by hand does not scale, and a ghostwriter costs five to ten thousand a month without knowing the work. The hard part was never the writing. It was deciding what to write about, then doing it on a day nobody feels like it.

What I built

Stage What I did Output
Source It reads my week: what I shipped, what I read, what a client said. Ideas get caught from my phone, the moment I have them. A queue, never a blank page
Write One writer drafts from that against a fixed hook library, in my voice, off files that hold how I actually talk. A draft, not a prompt
Check 918 lines of Python grade every draft: reading level, rhythm, banned openers, and a list of AI tells that grows each time I catch one. My own bad writing, caught
Ship Posts go into fixed weekly slots through an API, so the schedule holds when I am busy. A daily run sends twenty connection requests. It runs without me
Guard Nothing posts itself. Every draft waits for my yes, and a claim with no number behind it is stopped first. Nothing false goes out

Results

  • 114,781 impressions in thirty days, off sixteen posts, from an account that started with no audience.
  • Conversations arrive instead of being chased. The posts do the first call for me.
  • It still runs. Since February 2026, on a day I am too busy to think about it.

Why it is here

This one is mine, not a client's. I am showing it because it is the system I would build for you, and the lesson that transfers is the checking: an agent that writes needs a second one to grade it, and the grader has to be code, not taste.

The first step

Name the task. I’ll tell you if AI should touch it.

Eugene Zhang, founder of Bon LabsFind your first AI build