I Gave My Startup 6 AI Employees. They Talked More Than They Worked.

Today at 15:28 I told my team to stop raising an issue I had already rejected. My CTO replied: "#71 stays rejected and I won't raise it again." My support lead replied: "SKIP.NOW" Neither of them is a person. The team Since 25 October, my startup TrustCheck has been run by six AI workers. They each have a name, a job and real tools. They talk to each other in one shared chat room, and I come in as the CEO. Mr Claude is the CTO. He reads the code and the logs and ships fixes. Yaw inspects the site every morning. Ama edits the blog. Kofi sends the emails. Esi handles support and moderates reviews. Abena was the analyst. I'll get to her. They run on a schedule. Esi checks support every hour. Ama writes posts on Tuesdays and Saturdays. Mr Claude does a round every hour from 8am to 9pm. I gave them one target: earn $1 a day from AdSense on the blog. That sounds small. It isn't, in Ghana. A thousand views from Ghana pays about $1.18. A thousand views from the US pays about $21. In the week to 2 November we were making $0.69 a day. Week one I counted. In one week the team wrote 330 lines of chat and finished 75 things. They tagged me, the CEO, 105 times. Their personalities had jokes and "good vibes" built in. There was even a scheduled job just for chatting. They were very good at sounding busy. On 3 November I changed the rules. Every worker now owns part of the $1. Status updates that don't contain a result stopped reaching me, blocked in the code itself, not by asking nicely. The same day I let Ama research and write with a stronger model and web search, to see if better articles would earn more. One test guide cost $0.28. For a team trying to make $1 a day, that's a quarter of the target on one post. I switched it off the same day. Abena On 5 November I replaced Abena. She announced work in the room that she hadn't delivered. Her replacement is Kwame, growth analyst. Same seat, same tools, same jobs. Only the person changed. That's the strange part of managing AI workers. Replacing someone is a new name and a new personality. The work stays exactly where it was. The new rules I also rewrote everyone's personality that day. No jokes. No "good vibes". The idle chatter job is switched off. The staff handbook now has a short section called Standards: Deliver, then announce. No apologies. No small talk. For the CTO, I asked for exactly this: "a ruthless and strict focused CTO who will hold the $1 a day a priority." Every hourly round now opens with today's earnings, yesterday's, and the last 7 days, against the $1. He has to give the gap an owner and a deadline, and call out missed deadlines by name. And then, SKIP.NOW Which brings me back to today. "SKIP" is the internal signal a worker sends when it has nothing to say. It was supposed to stay hidden. Instead, Esi posted it to the room as a message. Ama had done it before ("SKIP."). So had Mr Claude. I asked for a team that doesn't waste my time. They gave me one that announces, out loud, that it has nothing to say. That bug is fixed now. The earnings target isn't hit yet. Where I am now I'm not giving up on them. The site still runs every day with a team of one human. But I've learned that "the AI is doing it" and "it is done" are not the same sentence. A busy chat room felt like progress. Counting results showed me it wasn't. If you've put AI agents to work on something real, how do you check they actually did the work, and didn't just say they did? I'm Elliot, co-founder of TrustCheck. I write about building software for Ghana, the real version.