Amazon employees use artificial intelligence only for reporting, not for real results

How Generative AI Turns into a “Performance Indicator” for Employees

With the growing popularity of generative artificial intelligence (AI) technologies, companies are increasingly demanding that their employees demonstrate practical achievements in working with these systems. In response to such pressure, a new phenomenon has emerged: employees assign meaningless tasks to AI solely to improve their usage statistics.

The Financial Times Describes the Situation at Amazon

In an article by the Financial Times it is described how workers at the e‑commerce giant are increasingly using the MeshClaw software solution. This agent connects to employees’ work tools and can carry out tasks on their behalf. Some of them use it to automate virtually useless actions in order to raise their personal AI adoption efficiency rating.

The reason is mounting pressure from management: now more than 80 % of developers are required to use AI every week. Each employee’s usage is tracked separately, and those who fall behind the target metrics must explain this to leadership. Although managers formally should not evaluate employees based on token‑usage data, in recent months there have been suspicions that this is indeed happening.

In such an environment some Amazon workers face false incentives: they simulate high activity for personal gain rather than for the company’s benefit. This is especially important for those tech giants that invest huge sums in AI infrastructure and want to demonstrate real value from implementation to investors and leadership.

How Amazon Manages AI Usage Data

Amazon plans to spend about $200 billion on capital expenditures this year. To ensure that technologies are actually being applied, the company is tightening control over AI usage. Initially group statistics on employee AI use were published, but now only individual results are accessible to the specialists themselves and their direct supervisors. Managers are formally prohibited from using these data for performance evaluations.

The Situation in Other Tech Companies

The Financial Times notes that similar abuses are also observed at Meta Platforms. MeshClaw, developed by more than 30 Amazon specialists, can upload code to production versions, sort emails, and interact with applications such as Slack. Many employees express concerns about the security of such solutions: AI can generate errors, and control over its actions remains weak.

Thus, rising demands for generative AI usage lead to the creation of “performance indicators” that sometimes become goals in themselves rather than means to achieve real business results.