AI can now test whether other AI's optimization models actually work

Researchers built a system where AI agents automatically check whether other AI systems have created correct optimization models from natural language descriptions. Until now, there was no reliable way to validate these AI-generated models — they could sound right but fail in practice.

When you ask an AI to turn a business problem into a mathematical optimization model, you have no way to know if the model is actually correct without manually reviewing it line by line. This system catches errors automatically by generating tests, running them, and mutating the model to see if the tests still pass — the same technique human software engineers use. What matters is the throughput: if AI can generate optimization models and AI can validate them without human intervention, companies can move from 'can't use AI for this task' to 'AI handles the entire pipeline, humans spot-check occasionally.'