Graham McNicoll

Amazon got a mandatory meeting. Most engineering orgs will get a quieter version of the same problem first. Four major outages in a single week. Reporting on an internal memo described a pattern of unsafe practices: AI-generated code reaching production without adequate guardrails, causing availability failures across ecommerce infrastructure. This is one of the largest engineering organizations in the world. Four major incidents in a week. I am not surprised. I use AI coding tools every day and I have seen how much faster code reaches production now. That speed is real and it is valuable. What has not kept pace is the infrastructure to contain what happens when something goes wrong. This story is going to repeat across a lot of engineering orgs. It may already have at yours, in smaller ways that have not forced a mandatory meeting yet. A feature flag on every AI-generated change is a circuit breaker. The feature ships. Users do not see it until you are ready. If something breaks, you turn the flag off. No rollback. No emergency deploy. The flag is off and the damage stops before your team even opens the incident channel. Without that, you are shipping more code than ever with no ability to limit the blast radius when something breaks. At Amazon's scale, that means customers unable to check out and a mandatory all-hands to explain how it happened. At yours, it might be quieter. The problem is the same. The teams who come through this transition in good shape are the ones who maintain engineering discipline: maintaining quality control, and use feature flags to release and roll back. If your team is shipping AI-generated code directly to production, send me a DM.

Only one snapshot so far — the engagement-over-time curve appears once the daily scrape has captured this post at least twice.