Amazon Urges Engineering Leaders to Investigate Slow AI Adoption Among Teams

Amazon Focuses on AI Adoption Metrics and Software Development Guidelines

Amazon has asked its engineering leaders to closely examine teams that are slow to adopt artificial intelligence. According to a memo published on the company's website and shared by Terk Terkowitz, VP and Distinguished Engineer for Worldwide Amazon Stores, the company recommended paying special attention to the 10th percentile of AI adoption to understand why certain groups are not embracing the technology effectively.

The guidance accompanies updated role guidelines for software development engineers (SDEs), created by the company's Distinguished and Senior Principal Engineers. Leadership is also directed to review the 90th percentile, where teams use AI heavily, to evaluate whether their delivery output has increased correspondingly.

Shift in Guidelines and Acknowledgment of Flawed Metrics

For two decades, Amazon traditionally excluded specific technologies from its SDE guidelines, even as major shifts like cloud computing and microservices transformed the industry. However, senior engineers concluded that AI is changing software development more quickly and fundamentally than previous technological shifts.

To measure AI utilization, Amazon has used input metrics such as token usage and tool adoption, alongside throughput indicators like code changes and reviews. The memo explicitly describes these metrics as deeply flawed because they are disconnected from business outcomes and warns that treating them as rigid targets could increase costs and complexity. The memo notes that excessive token consumption conflicts with internal frugality principles, while measuring lines of code can clash with the principle of inventing and simplifying.

Practical Implications and Verification

Despite acknowledging these metric limitations, Amazon states that current data generally tracks real output, with high-adoption frontier teams delivering more while non-adopting teams fall behind.

The memo provides several balanced directives for staff and management:

• For Non-Adopting Engineers: Employees whose work does not benefit from current AI tools are told that this outcome represents a finding rather than a personal failure. They are encouraged to document their reasoning and share it with peers and managers.
• For Managers: Engineering managers are expected to use AI on their own production work and be able to articulate both a task where AI helped and one where it failed.
• Customer Proximity: The company emphasizes that engineers and managers closest to actual products and customers derive the greatest benefit from agentic software development.

*How to Verify:* Readers and industry observers can verify the original announcement by reviewing Amazon's official corporate communications and published engineering blog updates on their corporate website.