A finding, not a failing (Amazon)
Amazon has rewritten the role guideline for its software engineers, and one instruction in the memo behind it deserves to be quoted in full. "If your work does not benefit from current AI tooling, that is a finding, not a failing."
The memo, from Terk Terkowitz, a VP and Distinguished Engineer, asks engineers who reach that conclusion to document why, share it with their peers and bring it to their manager, and it describes the engineer who can explain where AI does not yet help as demonstrating exactly the judgement the company is asking for. It is equally candid about measurement, calling token usage and adoption rates deeply flawed input metrics and warning that chasing token volume sits at odds with frugality.
Yet the same document recommends watching the tenth percentile of AI adoption to find teams who are not embracing the tools, and cautions that individuals who do not adopt are falling behind. Negative findings are welcomed, and opt-outs are counted. Both things are written down, and the engineers reading it will draw their own conclusions about which one shapes their career.
That tension is the measurement problem inside any serious AI rollout. What reaches leadership depends on what the metrics make safe to say. When adoption is the number that shapes careers, negative findings move to private channels and the programme gets managed on filtered evidence. Amazon deserves credit for naming that failure mode in the very document that creates the incentive.
None of this is an argument against measuring adoption. Amazon reports that these input metrics correlate well with the outcomes that matter today, and someone has to watch the tail of the distribution. The harder question is whether the engineer who reports the negative finding ends up inside the percentile being watched, and whether the person reviewing the data can tell the difference between resistance and information.
The test of an AI programme is not the success stories it collects. It is what happens to the first credible person who says the tool did not help, and whether that sentence reaches the top unchanged.
#AI #EngineeringLeadership #Management
Sources:
- Amazon, "An update from Amazon engineering leadership on the role of the developer in the age of AI", memo by Terk Terkowitz, VP/Distinguished Engineer, Worldwide Amazon Stores: https://lnkd.in/gtCDG25Z