20:12Machine-translated from Azerbaijani by AIRead in Azerbaijani
Amazon Asks Engineers Not to Artificially Inflate AI Use
Amazon released updated artificial intelligence (AI) guidelines for engineers on October 27, 2026. The company asked employees not to artificially increase their use of AI to appear more productive. This practice, known as “tokenmaxxing,” involves inflating the number of text units, or tokens, processed by AI systems. The document had also been shared internally at the company in September.
The guidelines state that spending more tokens does not mean better results. AI use, token consumption, code changes, and code reviews can create a misleading impression when evaluated separately from the work performed. Making these metrics the main target increases unnecessary AI use. As a result, computing costs rise and software becomes more complex. Amazon will prioritize evaluating employees’ work based on long-term outcomes. Managers may review usage metrics carefully and in conjunction with other results. The company also recommends comparing the results of teams that use AI the least and the most. If the current tools do not benefit a particular task, not using them will not be considered a failure. In May, the company shut down an internal leaderboard showing employees’ token usage. Some employees were unnecessarily using an internal AI tool called MeshClaw, which could insert code into the system and operate in Slack, to boost their metrics. The company did not impose token limits on employees. In light of this trend, Dictionary.com named “maxxing” its 2026 Word of the Year. These guidelines do not mean that Amazon is moving away from artificial intelligence. CEO Andy Jassy believes AI will change the speed and scale of day-to-day work. According to him, automation could reduce the number of people employed by the company in the coming years, but other positions may also be created. Amazon expects AI systems to play a central role in the future of software engineering. Other companies are experiencing a similar cost problem. By April, Uber engineers had spent the entire annual AI budget allocated for Claude Code. Despite this, high token usage had not translated into useful new features for customers. Accenture acknowledged that it is difficult to estimate the cost of systems based on AI agents in advance. The company is developing an advisory tool called Token IQ for clients. OpenAI CEO Sam Altman also said token costs had become a major problem for companies.
The guidelines state that spending more tokens does not mean better results. AI use, token consumption, code changes, and code reviews can create a misleading impression when evaluated separately from the work performed. Making these metrics the main target increases unnecessary AI use. As a result, computing costs rise and software becomes more complex. Amazon will prioritize evaluating employees’ work based on long-term outcomes. Managers may review usage metrics carefully and in conjunction with other results. The company also recommends comparing the results of teams that use AI the least and the most. If the current tools do not benefit a particular task, not using them will not be considered a failure. In May, the company shut down an internal leaderboard showing employees’ token usage. Some employees were unnecessarily using an internal AI tool called MeshClaw, which could insert code into the system and operate in Slack, to boost their metrics. The company did not impose token limits on employees. In light of this trend, Dictionary.com named “maxxing” its 2026 Word of the Year. These guidelines do not mean that Amazon is moving away from artificial intelligence. CEO Andy Jassy believes AI will change the speed and scale of day-to-day work. According to him, automation could reduce the number of people employed by the company in the coming years, but other positions may also be created. Amazon expects AI systems to play a central role in the future of software engineering. Other companies are experiencing a similar cost problem. By April, Uber engineers had spent the entire annual AI budget allocated for Claude Code. Despite this, high token usage had not translated into useful new features for customers. Accenture acknowledged that it is difficult to estimate the cost of systems based on AI agents in advance. The company is developing an advisory tool called Token IQ for clients. OpenAI CEO Sam Altman also said token costs had become a major problem for companies.