I built an AI agent whose only job is to tell me what I don’t want to hear.
It reads messages I’ve sent and drafts I’ve written, then gives me feedback like a trusted advisor would: direct, but not cruel.
Five filters.
One uncomfortable mirror.
Here’s what showed up in the first week.
I undersell my expertise in writing.
I hedge. I soften. The agent notices when I use I think maybe we could consider…
in a thread where I’m the subject-matter expert.
I over-explain when I’ve already won.
Someone agrees with my recommendation, and I respond with a three-paragraph justification of why I was right. The agent caught it immediately: They already said yes. A two-line thank you is the power move. You’re turning a win into a campaign.
I manage my lows, but not my highs.
I’ve gotten decent at staying composed when things go wrong. But when things go well—when I’m proven right—that’s when I’m most likely to damage a professional relationship. The agent flags it: This is where the ‘I told you so’ energy lives. Manage this moment carefully.
I confuse being right with being effective.
Being effective is political and relational. It requires leaving room for other people to arrive at the conclusion themselves. The agent’s feedback: You’re optimizing for accuracy when you should be optimizing for influence.
I didn’t need an AI to tell me my growth areas. I already knew them.
What I needed was something that would catch me every single time—not just when I’m paying attention, but especially when I’m not.