AI routing systems now have to show their work — explaining why they pick cheap models over good ones

Researchers built a tool that forces AI systems to explain which model they're using for each task and why. Instead of silently picking the cheapest option that might work, the system now has to show its reasoning — whether it's using a specialized model because it's genuinely better, or because it's cheaper and the developer didn't notice the quality gap.

Right now, when you build an AI system that farms different tasks to different models, you have no idea if you're being smart or just cutting costs in places that matter. A developer thinks they're optimizing for efficiency but they might be systematically degrading quality in tasks they can't measure. This tool forces the decision into the light — you can see the trade-offs you're making and decide if they're actually acceptable.