The Performance Plateau Illusion

The ironies of automation that Lisanne Bainbridge identified in industrial process control acquire a specific and dangerous form in AI-augmented knowledge work: the performance plateau illusion. In the early months of AI adoption, the system performs brilliantly. The expert's judgment, still at full strength from years of manual practice, amplified by AI's execution capability, produces output that is genuinely superior to what either could achieve alone. Organizations measure this improvement and calibrate their expectations, commitments, and competitive positions accordingly. Quarterly targets are adjusted upward. Client promises are made based on the new productivity baseline. Hiring models, compensation structures, and project timelines are redesigned around AI-augmented capability levels. These adjustments are rational—the AI-augmented performance is real, and planning should reflect it. But the performance the organization measured was the performance of the system during what Bainbridge would call the honeymoon: the period when the human's contribution, still built from years of pre-AI practice, remains at its pre-AI peak. As the compounding degradation proceeds, the human's contribution