Infobae reported that Amazon published a memo by Terk Terkowitz on October 27, 2026, after it circulated internally in September. The memo reportedly calls token consumption, AI-tool adoption and code-change counts flawed proxies for engineering work, while allowing their cautious use as diagnostic signals. Terkowitz is a vice president and distinguished engineer in Amazon’s Stores division.

Activity counts are not engineering outcomes

Tokens are pieces of text an AI model processes. A token total measures model activity; an adoption count shows whether people use AI tools; a code-change count tallies edits. None, by itself, says whether the work delivered a useful result.

According to Infobae’s account, the memo calls these indicators fundamentally flawed because they are disconnected from what matters. It also reportedly says they correlate with important outcomes and can help managers examine teams with higher or lower AI use. That makes them possible clues, not direct measures of engineering results.

The memo leaves room for work AI does not help

The reported memo treats identifying work that does not benefit from current AI tools as a valid finding, not a failure. It also reportedly says managers should have used AI tools on real work and be able to name a task where they helped and one where they did not.

That approach focuses on whether a tool fits the task, rather than whether an engineer can raise a usage count. The account describes careful diagnostic use of the indicators, not a blanket prohibition on AI.

An earlier weekly target and performance reviews

A May 12, 2026, Tom’s Hardware report carried by Yahoo Finance described an earlier target for more than 80% of Amazon developers to use AI tools weekly, along with token-consumption tracking. The report attributed to Amazon a statement that usage statistics would not affect performance evaluations. Separately, it reported that employees believed managers monitored the data.

The company statement concerned performance reviews; the employees’ accounts described their perception of monitoring. The reported weekly target is historical, not a stated current target.

Amazon Connect tracks customer-service agents

Amazon Connect’s AI-agent analytics concern customer service, not Amazon engineers’ use of AI tools. AWS says its service can track AI-agent-led interactions, handoffs, conversation turns and average handle time in Amazon Connect analytics.