JPMorgan Chase is moving artificial intelligence from an optional productivity tool toward an expected workplace skill. Managers are tracking how roughly 65,000 engineers and technologists use tools such as ChatGPT and Claude Code, and that behaviour may influence performance reviews.
Internal systems reportedly classify employees as light or heavy users. The bank’s objective appears straightforward: avoid the uneven adoption that has weakened many enterprise software rollouts and make AI part of ordinary work in coding, document review and routine tasks.
The difficult question is whether usage itself is a meaningful measure of performance.
Adoption metrics can change behaviour quickly
Large organisations often provide new tools without changing incentives. Enthusiasts experiment, cautious teams retain familiar processes and executives struggle to connect licence costs with business results.
Linking AI use to management attention creates a stronger signal. Employees understand that literacy is becoming part of the job, much as spreadsheets and development environments became baseline tools in earlier periods.
This can accelerate learning across teams. Frequent users discover effective patterns, share examples and help establish common expectations. Managers gain visibility into where adoption is high or low instead of relying on surveys and anecdotes.
But frequency is not quality. An employee can generate many prompts without improving accuracy, speed or customer outcomes. Another may use AI selectively for a difficult task and create greater value with fewer interactions.
If the metric becomes a target, staff may feel pressure to use AI where it adds little. The bank therefore needs to distinguish thoughtful application from activity performed to satisfy a dashboard.
Productivity still requires verified output
Generative tools can draft code, summarise information and reduce repetitive work. They can also return incomplete or incorrect answers. In a regulated bank, employees remain responsible for checking what enters production or influences a client-facing process.
The right performance question is not whether a worker used AI. It is whether the tool helped produce better work under the required controls. Review criteria could consider time saved, error rates, quality of verification and the employee’s judgment about when not to automate.
JPMorgan already applies AI in areas such as fraud detection and risk analysis, where governance is well understood. Extending tools across a much larger employee population introduces different risks: sensitive information can enter the wrong service, generated code can contain defects and polished summaries can conceal omissions.
Training should therefore cover data boundaries, source checking and escalation—not only prompting. Heavy use without these habits may increase operational exposure rather than efficiency.
AI literacy becomes a workforce decision
The bank’s approach may influence hiring, training and promotion. Candidates could increasingly be assessed on their ability to collaborate with AI, inspect outputs and redesign a workflow around the tool.
It may also reset expectations about workload. If one task takes less time, leaders may ask employees to deliver more. That can create genuine capacity, but it can also turn productivity into permanent acceleration without allowing time for deeper work or control.
Managers need evidence that the gain survives beyond the individual step. An AI draft that saves ten minutes but creates twenty minutes of review elsewhere is not a productivity improvement.
JPMorgan’s scale makes this a consequential experiment. A common adoption standard across 65,000 technologists can reveal how quickly AI becomes normal in a large regulated workforce—and where measurement distorts behaviour.
Other financial institutions will watch whether the policy produces better output, faster delivery and stronger skills. The lesson should not be that every employee must become a heavy user. It should be that companies need a more mature definition of effective use.
AI literacy is becoming part of work. Measuring it responsibly means rewarding judgment and results, not simply counting how often someone asks a machine for help.
