Build For Judgment, Not Just Usage

Recent Wharton research from Steven Shaw and Gideon Nave offers an important warning. Their work on “cognitive surrender” shows that people often adopt AI outputs with too little scrutiny. When the AI was accurate, performance improved. When it was wrong, performance dropped, while confidence still rose. That is a risky combination in any enterprise setting. It means AI can improve work, but it can also make weak thinking feel more certain.

This is one reason experienced users outperform novices. They are less likely to treat AI as an answer machine and more likely to use it as a thinking partner. They iterate. They challenge. They compare. They bring harder problems to the interaction. That is where leaders should focus their enablement efforts—not just on access but on the behaviors that make collaboration productive.

Create The Conditions For Knowledge To Flow

Advanced augmentation depends on human context: edge cases, institutional memory, customer nuance and hard-won pattern recognition. But workers will not consistently share that expertise if they believe the system is capturing it in ways that weaken their position. Research from Harvard and MIT shows that when workers understand that workplace data can be used to train AI systems that replicate their expertise, they become more likely to withhold documentation and resist monitoring.

That matters because augmentation gets stronger as more context enters the system. If people start protecting what they know, the organization may still see activity, but the quality of the collaboration begins to thin out. The AI has less of the expertise that makes it useful, and leaders have less visibility into why progress is stalling. The issue is not simply culture. It is whether the organization can keep the knowledge flywheel turning.

Measure Augmentation On The Right Terms

This is where many organizations still stall. Automation gets more attention because it is easier to understand and easier to model. Its value often shows up neatly in labor savings or cycle-time reduction. Augmentation is harder. Its returns show up in better judgment, faster iteration, broader problem-solving and the ability to take on work that previously sat outside a team’s capabilities.

That makes augmentation easier to underinvest in. If leaders only look for the clean math of substitution, they will miss the larger opportunity: capability expansion. OpenAI’s finding that many workers are using AI to do tasks they could not do before is important for exactly this reason. Part of the return is not just efficiency. It is new capacity.

But that value does not reveal itself automatically. Organizations have to decide where stronger human-AI collaboration should change the outcome, not just speed up the task. They have to define what better looks like in advance, and they have to track whether time saved is being converted into higher-quality decisions, better service, faster learning or more ambitious work. Otherwise, augmentation remains conceptually appealing but economically hard to defend.

The next gap in AI will not be between companies that bought licenses and companies that did not. It will be between companies whose people learned how to think with AI and companies whose people learned only how to hand work to it.

Two or three years from now, that difference will be hard to hide. One group will be moving faster because it has built stronger judgment, richer knowledge flows and new organizational capacity. The other will still be counting activity and wondering why the returns feel smaller than expected.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?