The Number Was Real. It Just Wasn't Mine.

I use AI agents to help build my job search materials. One of them did some research and put a number on my resume.

The number was real. It came from a press release about PowerFlex, a company I worked at.

But the press release was dated after my time there. The number described a company I wasn't there for. It wasn't wrong about PowerFlex. It was wrong about me.

I only caught it because I asked a second AI agent to cross-check the resume. That's where it fell apart. I think at least one application went out with that number on it. That one never came back to me, and I'll never know whether it mattered.

Here's what I took from that. The research didn't fail. The agent did what research agents do: it found something plausible and wrote it down. What failed was that nothing stood between its output and my name. Nobody asked, "Is this true for you?"

So I changed where truth lives. In the tool I'm building, Sovereign Drive, anything an agent proposes is unconfirmed until I approve it. Documents get built only from facts I've confirmed. An agent can suggest. It can't assert.

What I can actually stand behind at PowerFlex is smaller, and it's mine: we reduced expensive truck rolls. That was always the point.

Between your AI's output and your name on it, who confirms it's true?