Google DeepMind researchers Alex Imas, Andrew Koh, and co-founder Shane Legg published a study arguing that multi-agent collaboration capability—not individual AI performance—is the critical threshold for triggering abrupt economic shifts. The paper outlines four stages of AI development: minimal AGI, supercompetent AI, supercollaborative AI, and ASI. According to the authors, current AI systems excel in individual task performance but remain weak in collaborative capabilities. Improvements in standalone AI will only gradually affect wages and prices. However, once multi-agent systems cross what the researchers term the "super-collaboration" threshold, market concentration, employment patterns, and wage structures will undergo sudden, discontinuous changes. The study notes that labor market disruptions will precede corporate restructuring. The researchers call for establishing new benchmarks to measure multi-agent organizational capability, emphasizing that existing evaluation frameworks fail to capture this dimension of AI progress.