Researchers use AI agents with different reasoning levels to simulate supply chains, revealing how cognitive mismatches between tiers create inefficiency

Researchers at an unnamed institution ran computer simulations of multi-stage supply chains using two different LLMs (AI systems that understand and generate text)—DeepSeek and GPT—deliberately paired them at different tiers so some agents were 'smarter' than others, then measured how well the chain coordinated.

Unlike prior studies that assumed all decision-makers think alike, this experiment showed that when suppliers, distributors, and retailers have different levels of reasoning sophistication, they make myopic (short-term focused) and self-interested choices that cascade down the chain, amplifying inefficiency rather than spreading it evenly.

Supply chain managers and operations leaders now have evidence that cognitive mismatch—not just information lag or poor incentives—drives the bullwhip effect (where small customer demand changes explode into huge upstream swings). This suggests that hiring, training, or AI-tool selection decisions that create uneven decision quality across your vendor network may be costing you more than previously understood.