Let the models fight
Their bias might produce your best answer.
If you ask an AI model to compare its plan with another one, it will most of the time find reasons why its own approach is better.
If you mention which model created the plan, the behaviour becomes… interesting.
When reviewing output from one of its siblings, it will usually be more forgiving.
But tell it a competitor model created the plan, and suddenly there is a missing assumption, an overlooked risk, or a better way to structure the work.
I find this kind of bias funny. Some would consider it outrageous. Either way, it is useful.
Instead of treating one model like an oracle, treat several models like opinionated reviewers.
Let one create the plan. Give that plan to another for review. Then return the critique to the first model and let it defend or improve its work.
One catches the risk. Another simplifies the solution. A third notices the question nobody asked.
The models can argue about who had the better plan.
You only care about the result.
Let them fight. Keep the improvements.
Not sure which AI recommendation to trust? Let’s talk.