Alvis Ng argues that AI coding tools become risky when management mandates their use instead of letting engineers adopt them based on trust and experience. It uses Amazon as a warning case, claiming that a forced rollout of the Kiro AI coding assistant led to high adoption metrics but serious production failures and outages.

He says the real issue is not resistance to AI, since engineers often want AI tools, but resistance to being told which tool to use and being measured on compliance. The article contrasts “calibrated trust,” built slowly through hands-on experience, with “calibrated compliance,” which can produce fast adoption but weak understanding and unsafe deployments.

Its core message is that companies should measure and build trust in AI-assisted engineering rather than simply tracking usage dashboards.

Medium:
https://medium.com/@iamalvisng/your-ai-coding-mandate-will-backfire-amazon-proved-it-e20973399f62