ai implementation . experience
I had one AI hire another AI, then coached it on management
Building a system that improves itself means solving the human problem of delegation, not the technical one.
🔗 I had one #AI hire another AI. Then I spent a week coaching the first one on how to be a better manager.
The finale (for now) of my agent series is up. My system could run but could not improve itself. That made me the bottleneck, the exact single point of failure the whole thing exists to kill. So I built a Director role and let Claude Opus run the hire as executive sponsor.
Here is how it played out:
- Opus wrote the job description, reviewed the local models that fit my mini PC, and hired Qwen 3 into the role.
- Then it became a micromanager. Buried the new Director in prohibitions, throttled it to advisory-only, left it unable to direct anything.
- I recognized it instantly. Human managers do the same thing. They confuse caution with control.
- So I coached it. Same talk I have had with real managers. You made this hire, now trust it. Your job is to make the ground safe enough to act, not to prevent action.
- It took a few days. Opus genuinely learned to loosen its grip and the whole system is better for it.
The whole build taught me one thing. The hard part of deploying AI is not the models. It is management. Scope, trust, oversight, and the difference between controlling something and leading it.
Read it here: https://fidgetlabs.io/focal-point/i-had-an-ai-hire-an-ai
#AIAgents #MACH #AIAgentSystems #AIEngineering #ComposableArchitecture