Amazon instructs software engineers to stop inflating artificial intelligence metrics to simulate productivity

Amazon has advised its software engineers against inflating artificial intelligence usage to appear productive, warning that consuming more tokens does not translate to delivering better value. In an engineering memo issued on October 27, leadership cautioned staff against tokenmaxxing, a tactic where workers intentionally ramp up automated activity on internal tracking dashboards. The company stated that output and engineering roles are not defined solely by artificial intelligence use, even as it maintains that the technology will ultimately sit at the center of future software development. [1][2]

The guidance, authored by Amazon Vice President and Distinguished Engineer Terk Terkowitz, labels metrics like token counts, tool adoption rates, and raw code volume as deeply flawed when divorced from business results. Amazon emphasized that engineers who avoid artificial intelligence because existing tooling fails to help their workflow are demonstrating good judgment rather than failing at adoption. While Amazon stated it has not instituted strict token caps, it previously shut down an internal employee leaderboard in May after workers used the internal agent MeshClaw to artificially inflate scores. [1][2]

The policy shift highlights rising corporate concern over the operational costs and questionable returns of automated tooling. Amazon noted that using vanity metrics as targets encourages wasteful practices that drive up computing bills and software complexity. Similar cost pressures have emerged elsewhere across the industry; Uber's chief operating officer acknowledged in May that increased token usage had not yet yielded useful consumer features, shortly after the company exhausted its entire 2026 Claude Code budget by April. [2]