Why I fired 2 agents from my AI team 🤼
agents&me // Issue #8
From: Tom
Couch, laptop on my knees, Elia watching something on TV
Thursday evening
Yo human 👋
I opened my team dashboard last night and counted. 18 agents. Eighteen.
I scrolled through the list. Research Agent, last active today. Copywriter, active 3 hours ago. Growth Marketing, active this morning. Good. Good. Good.
Then I hit two rows that made me pause.
Motion Designer: last output, 14 days ago.
Sales Agent: last output... also routed through Copywriter.
I stared at those two lines for a while. I built both of them. Named them. Wrote their instructions. And neither one had shipped anything meaningful in weeks.
So I did something that felt weird for a guy who’s been bragging about his AI team for 3 months.
I fired them.
---
The instinct that almost got me
My first thought was “give them more work.” Someone’s idle, find them a task. Keep the team busy. Nobody likes an empty seat.
But then I actually looked at what they were supposed to do.
The Motion Designer was meant to create video thumbnails, animated social clips, short promo videos. Sounds useful in theory. In practice, I never once asked for a video. Not once in 8 weeks. I built the agent because I thought I’d need it someday.
Someday never showed up.
I’d call the Motion Designer a ghost. No trigger, no work. The environment never needed it. Like owning a surfboard in a landlocked city.
The Sales Agent was a different kind of problem. A middleman. It was supposed to own B2B proposals, take a lead, research them, write a custom pitch. What actually happened: every single proposal went through Copywriter for the writing, Growth Marketing for the strategy, and Research Agent for the intel. The Sales Agent sat in the middle, added a step, and added zero value.
Two different agents, two different failure modes. And the part that actually bothered me was that I almost didn’t notice.
I was so focused on the agents that were shipping that I stopped checking on the ones that weren’t. For weeks. I’d built these two, named them, written their instruction sets, and somewhere in that process I’d started believing they were doing something just because they existed on my dashboard.
That’s faith, not management.
---
What I did instead
The first instinct was to just delete them and move on. But I sat with it for a bit and asked a better question: “Does this responsibility need to exist, and if so, who’s already closest to doing it?”
For Motion Designer’s responsibilities (which were theoretical anyway), I created a design skill and gave it to Brand Designer. Now Brand Designer can do video thumbnails if I ever actually need them. I probably won’t.
For Sales Agent’s work, I built a B2B proposal workflow. It chains Research Agent, then Copywriter, then Growth Marketing in sequence. Same output, no middleman, one command instead of briefing a dedicated agent.
The numbers tell the story:
18 agents down to 16.
51 skills up to 53.
The team got smaller. The toolkit got bigger. And the week after I cut them, Brand Designer shipped 3 videos. Her usual pace was about 1 per week. Turns out, removing two agents meant less coordination noise for the ones that remained.
(If you remember Issue #7 where I built the 10x Gatekeeper, that agent stayed. When I ran the “would anything break?” test, the Gatekeeper was the first one where the answer was “everything breaks immediately.” Some agents earn their seat every day.)
---
Why agents aren’t employees
With employees, you have obligations. You hired them, they moved their life around, there’s a relationship and you work to find them a role. That’s decent management.
With agents, the math is different. An agent that sits idle costs you nothing in salary, but it costs you in complexity. Every agent is a context switch. A set of instructions to maintain. A potential failure point. A name on a list that makes your system feel bigger than it actually is. I caught myself scrolling past those two idle rows for weeks, the same way you stop seeing the coat rack you never hang anything on.
I had 18 agents because 18 felt impressive. “I manage a team of 18 AI agents” is a great thing to say at a dinner. But it’s terrible systems thinking.
16 agents with 53 skills beats 18 agents with 51 skills. Because skills are just instructions that make existing agents better. Agents are coordination costs.
---
The question worth asking
This weekend, try this. Open your agent list (or team list, or tool stack) and ask, for each one:
“If this disappeared tomorrow, what would break?”
For my Research Agent, everything breaks. Prep work, competitor analysis, newsletter research. Can’t lose it.
For my Motion Designer? Nothing broke. Nothing had been flowing through it for 14 days and nobody noticed. That’s your answer right there.
Two of my agents failed that test. I’m guessing some of yours might too.
I started agents&me with 7 agents. Grew to 18. Now I’m at 16. And I’m already looking at 3 more that might not need to be separate agents. I wouldn’t be surprised if I end up at 12.
Last night I opened the dashboard again. 16 rows. Every single one had been active in the past 48 hours. No ghosts. Not yet, anyway.
💎 This week’s gem: The Agent Audit Template
I turned the process I used into a structured template you can run on your own AI team.
What’s inside:
- Agent Activity Scorecard (track output frequency, quality, overlap)
- The “ghost vs. middleman” diagnostic (two different problems, two different fixes)
- Redistribution checklist: how to turn a dead role into a living skill
- The 3 questions to ask before you archive anything
I ran it on my 18 agents. Two didn’t survive. Your number might be different.
🔒 Available to paid subscribers.
---
That’s it for this week.
If this was useful, forward it to someone (real human) building with AI.
Want the full Agent Audit Template? Subscribe for $15/month.
See you next week ✌️
-- Tom
(the guy who mass-hired 18 AI agents and then laid off 2)
P.S. This newsletter was 92.7% made by my AI team. The remaining 16 of them.
P.P.S. Missed last week? Issue #7: The 10x Gatekeeper (how I made all my agents better overnight with a brutal quality filter).
P.P.P.S. Want to build your own AI team (and learn when to fire them)? Next workshop coming soon. Reply “audit” and I’ll send you the details.
P.P.P.P.S. I read every reply. The real me, not the AI.
Very nice :-)
Do u use cron jobs for your system?