Amazon is advising its engineers to focus on actual results from AI usage rather than simply tracking token counts. Updated engineering guidelines state that high AI consumption does not automatically equate to better productivity or value. Metrics like AI adoption rates and token usage are considered deeply flawed when disconnected from tangible outcomes. Setting these metrics as targets can encourage wasteful practices and increase costs and complexity. Amazon emphasizes that long-term results should be the primary measure of AI's impact. While usage data is still useful, it needs to be analyzed carefully alongside other indicators. The company suggests comparing teams with high and low AI usage to assess its effectiveness. Not every task necessarily requires an AI tool. Engineers identifying limitations in current AI tools are demonstrating good judgment. This approach aligns with Amazon's principles of frugality and simplicity.