my AI called itself a terrible employee
my AI wrote the first draft of this email. its hot take was: "AI agents make terrible employees."
my AI wrote the first draft of this email.
its hot take was: “AI agents make terrible employees.”
I said no.
because in the first 15 days of October, my AI team:
- made 372 ads for 3 ecom brands
- built 98 of them in Meta, ready for my click
- closed 44 tasks
- logged 1,445 changes
and as most of you know, I’m a one-man show with no employees.
last week I made SENPAI my AI CEO
and honestly? being a one-man show with AI is amazing right now
so here’s my real hot take:
YOUR FIRST HIRE SHOULD BE AN AI CEO.
this is my org chart
me, the board. I set the goals. I say yes or no on 5 things: money, switching ads on or off, a client’s store, deleting client data, and any message that leaves the building.
SENPAI (Claude Code), the CEO. it plans the work, gives each task to the right agent, runs the server and checks every result. When an agent gets stuck, it takes over.
LUKA (Codex), the CMO. it reads the ad data, finds what sells and plans the next tests.
BOSSMODE (Hermes), the operator. it makes the ads, builds the campaigns and does the heavy lifting.
they share one brain: one folder of notes, one memory, 100+ skills.
they all live in my Discord. every task gets its own thread. one page shows me every task, who owns it and what waits on me.
here’s how one ad workflow works:
- LUKA reads what performed in ad accounts and writes 4 different ad concepts.
- BOSSMODE makes 4 ads per concept, with the brand’s real product photos. (AI still draws products wrong but I get 16 versions of ads I can use…)
- Fresh agents check every ad against the real product
- An “art director” agent cuts the weak ones.
- I say yes or no to each ad on one page then the client sees only what I kept and approves creatives.
- After the client’s yes, the team (AI) builds the ads in Meta, paused. I flip the switch.
my approval rate on the first look of ads went from 42% to 84%
it helps me outside the ads too:
- it reads my client calls the same day and turns what we decided into tasks.
- it read 980 of my old Looms and pulled out 3,219 nuggets I can search.
- every morning it works out what a purchase cost in each country for mutliple brands, so we judge every ad by profit.
does it make mistakes? sure.
it sent 49 ads to Meta at once. Meta rejected 22.
this week it deleted its own memory during a cleanup. a copy from 10 minutes earlier saved it 😅
every mistake becomes a written rule. now ads go to Meta in batches and it has to read back if we throttle the API
the 3 agents cost me about $640 a month: 2 Claude plans, 1 ChatGPT plan, 1 server (plus the image and data tools they use)
want to try this? steal these 3 rules:
- give every agent one job title.
- write your red lines down.
- never let the agent that made something check it’s own work.
and if you want people around you who already hit the walls you’re about to hit…
my friend Naum did 70+ free 1-on-1 calls with people building businesses in the past year
then he took what worked across all of them and put it in one place. it’s called Builders.
- 214 vetted ideas with the research done
- a strategy wizard: problem, offer, traffic, funnel, MVP
- 8 AI agents you run inside Claude or Cursor
- rooms for Claude Code, payments and VAT, GTM, company and legal
- 1000+ builders inside.
founding seats are $90/year for the first 100. less than your gym membership.
Talk soon,
Kris
P.S. what job would you give an AI first? hit reply. I want to know 😅



