Are AI Agents Compromised by Design?

Yes.

     *The OODA loop—for observe, orient, decide, act*

Very logical. Current AI doesn't do logic very well.

What it does instead is probability. At best, any results are a quasi-match from it's training data. At worst, it is something invented to sound plausible.

In any case, it is less than inspired thinking.