When assessing AI input output, it is moving from “how much has been spent” to “what has been saved”. Boris Cherny, head of Claude Code, an Anthropic programming tool, states that token consumption or dashboard usage alone is only a reflection of activity and does not directly indicate whether AI offers returns.
Not just use.
Cherny, in her X platform address on the path to the adoption of AI by an enterprise, stated that the next step would be to measure the real benefits when employees had incorporated the AI tool into their daily processes. In his view, many companies would look at the data first, but such indicators were closer to activity records rather than investment returns.
He raised the more meaningful question as to whether, in the absence of AI, the work would have required time to be done by the engineering team; if the answer was yes, the enterprise would have been able to further estimate the amount of manual time that would have been spent and the corresponding human cost. This part of the savings is closer to the direct return of AI.
The return is on the time saved.
According to him, instead of focusing on the costs of a model call per se, an enterprise should see AI as an alternative to part of the engineering work. This measure is particularly direct in the case of code generation, repair and maintenance tasks, as enterprises can generally estimate more clearly the development resources that would otherwise be required.
Cherny also mentioned that the greater benefits were not just “quicker” to do the same, but to move more repairs and maintenance backstage and release the team for new and constructive work. It is only when the team is no longer occupied by a large number of repetitive tasks that it is possible for the enterprise to start moving forward with projects that were previously under-resourced and difficult to schedule.
Enterprises start to recalculate AI
This statement also reflects a changing focus on AI spending. In the preceding period, there was a general industry-wide trend around Token ' s use of scale to measure AI progress, but as inputs expanded, enterprises began to focus more on cost control and actual output.
In the recent past, the issue has been openly addressed by many executives of technology and finance companies. Jamie Dimon, CEO of Morgan Chase, said earlier that the AI costs of the enterprise were rising rapidly, so that the company would be as rational as it was in managing other resources to assess inputs. The CEO of OpenAI, Sam Altman, also stated that lower spending and higher value had become a topic frequently raised by business clients.
From this trend, the focus of enterprise procurement and deployment of AI tools is shifting from pursuing higher calls to proving whether these tools actually reduce manual inputs and expand the range of work teams can do.