I ran an experiment this week that's a good example of what I actually build: put two AI models, Claude and Codex, in a structured debate against each other, no script, ten rounds, until they either converged or one gave up.

The question I gave them was about startup ideas, but that's not really the point. What's interesting is watching two models argue, concede points, catch each other rationalizing weak answers, and converge on a shared framework neither started with. That's a different kind of AI output than a single prompt-response, and it's the kind of multi-model orchestration work I build for clients.

Full writeup on Medium:

https://medium.com/@rawaldelhi/5-apps-you-could-build-today-that-could-pay-you-in-30-days-c5118099597a

If you're trying to get more out of running multiple AI models together instead of just picking one, that's exactly the kind of problem I work on.