I’m building Mindpod with AI agents, and I hold them to the same three questions I just asked you to put to your own systems. Honest status: 35 agents are designed; five are in alpha, which means they may rehearse, read, or propose a change, never make one. Every consequential action, a send, a publish, a change to a setting, stops at a human approval gate. Mine.
What that has taught me is that the bottleneck isn’t the agents. It’s the quality of my approvals. If I rubber-stamp, the gate is theater, so every request has to arrive with the answers attached: what it wants to do, why, what the bad week looks like, and how we’d undo it. An agent that stops and asks beats one that guesses right most of the time. That isn’t an AI invention; it’s plain change control, pointed at a new kind of tool. I’m building the company with them, gated, logged, supervised, and writing down what happens, misses included.
If this reached you by email, hit reply and tell me your role and the one system you suspect has nobody’s name on it. I read every reply. On the web? The address below works the same.
Jaras FunderburgFounder, Mindpod Technologies