Amazon sends chills through its workforce with mixed message memo about AI use

Amazon engineering leader Terk Terkowitz’s guidance on software development in the AI era arrives amid widespread anxiety about job security and reports of another 250 job cuts. The memo urges engineers to prioritize customer outcomes and professional judgment rather than delegate decisions to AI. It acknowledges that today’s tools do not improve every task and says documenting that limitation is a legitimate finding, not a personal failure. It also recommends judging adoption by whether teams’ delivery improves, rather than simply counting tool use. That advice matters against the backdrop of Amazon’s abandoned KiroRank leaderboard, which rewarded token consumption and encouraged employees to generate pointless work; similar usage-focused incentives have appeared elsewhere.

The central concern is that the memo sends conflicting signals: it calls adoption metrics flawed but still usable, while warning that engineers who do not adopt AI are falling behind. This leaves managers room to treat non-use as reasonable in one case and evidence of poor performance in another. The key test is whether engineers who show that AI made work worse will be heard - and whether that judgment can withstand promotion and performance reviews. The memo’s call to measure useful results is sound, but incentives often reward what is easiest to count. That contradiction is especially striking at Amazon: token-heavy activity is wasteful inside the company, yet AWS sells AI services on a per-token basis.