AI use is not a performance goal: Amazon changes how it rates coders
On 27 October Amazon published a guideline calling token consumption and AI-adoption rates flawed measures of performance; a day later it cut about 250 jobs in its devices unit.
In a guideline for software engineers, Amazon calls three common metrics, token consumption, adoption of AI tools and volume of code changes, deeply flawed; what matters, it says, are long-term outcomes. On 28 October Amazon cut about 250 jobs in its devices unit, mostly in software development, according to Reuters.
What happened
Many companies judge their AI roll-out by how much staff use the tools. Amazon has seen where that leads: in May it shut down an internal leaderboard ranking employees by their use of Kiro, its in-house tool, after staff gamed their numbers. The new guideline, written by Terk Terkowitz, a vice-president, and first circulated internally in September, judges engineers by their impact on customers, the business and the team. Those whose work does not benefit from AI should document why; that is a finding, not a failing. Amazon does not drop the metrics entirely: it says they correlate with good results and may still inform reviews with care. It publishes no evidence for that correlation. Amazon confirmed the devices cuts as a small number of roles without giving a figure. Since October 2025 it has cut about 30,000 corporate jobs, and it plans capital spending of about $220bn in 2026, largely on AI.
What follows
Assessment: the lesson for managers is that a usage target quickly becomes an end in itself; consumption gets measured rather than benefit, and costs rise. For employees the message is mixed. The guideline eases pressure but still allows usage to count in reviews, and it arrives in a week of fresh cuts among engineers. A direct link between the guideline and the cuts is not established; Amazon attributes them to a review of its structure. German companies rolling out AI assistants widely face the same question: how success is measured before the metric starts to drive behaviour.
What changes in practice
- Software engineers
- At Amazon, results officially count, not token consumption. Those not using AI must explain why.
- Managers
- AI usage targets can be gamed and raise costs; Amazon shut down an internal leaderboard for that reason.
- Amazon employees
- About 250 jobs went in the devices unit, mostly in software development, according to Reuters.
What to watch
- Whether Amazon keeps counting AI use in performance reviews
- Further cuts in software teams at large tech groups
- Metrics German companies use to measure AI adoption
Sources
- Amazon: Software engineer role guideline and AI (27.10.2026)
- CNBC: Amazon warns engineers about tokenmaxxing (27.10.2026)
- PPC Land: Amazon calls AI usage metrics 'deeply flawed' but keeps using them
- Business Today: Amazon slashes 250 jobs after issuing new AI rulebook (29.10.2026)
- American Bazaar: Amazon cuts 250 jobs in devices unit (28.10.2026)