As the AI budget of the enterprise continues to expand, discussions around whether “utilisation equals value” are warming. Cognition CEO Scott Wu believes that internal appraisals should not be built around token consumption, and that companies should see if AI actually brings about output growth.

Objection to the token ranking

In a podcast, Wu indicated that some companies ranked engineers with token consumption, a practice that had been diverted from focus. He looked more at how much the staff had delivered and whether there had been a clear return than on who had used more models.

Cognition is the developer of AI programming agent Devin. Wu argues that the company measures the impact by looking at whether engineering capacity is up, rather than simply tracking the size of a model. If the speed of delivery is significantly increased, it is easier to justify the related calculus and model costs.

It used to be on top of a big factory.

Wu’s statement appears after companies such as Meta and Amazon have been exposed to internal incentives and rankings to promote the use of AI. The mechanism was intended to encourage teams to look for applications, but later some staff members started calling AI more frequently to promote their ranking.

The Financial Times had previously reported that some employees had even allowed robots to handle tasks that lacked real value, and the companies concerned had subsequently cancelled such internal tracking. It was also mentioned that a senior vice president of the Amazon had asked staff not to use AI for the use of AI.

Total expenditure continues to expand after cost decreases

Although token prices have fallen significantly compared to 2023, total corporate AI expenditure has not been synchronized. The article mentions that Uber spent only four months on the AI budget in 2026, and then set the monthly token spending cap at $1,500.

  • 42% of staff interviewed said they could save about 8 hours per week
  • 66% indicates that the company has barely explained how to use time savings
  • About half did not shift time to more strategic work

In its 2026 Global Workforce AI report, Boston Consulting stated that AI had not fully translated into productivity gains and that the problem was not only in the tool itself, but also in management that had not clearly stated objectives, applied jobs and expected returns. Wu also argued that enterprises should assess AI inputs in terms of income, efficiency and cost savings.