← sources

The Microsoft incident proved what I always thought about AI

source: reddit (reddit:developersIndia)published 2026-11-14 07:52ingested 2026-11-15 02:34

retained text

I am a senior engineer. I have worked on big projects in multiple US European and Korean companies from scratch before and after AI.

AI is extremely good at writing code, developing applications, tests, docs and pretty much all things. But that’s a very low layer in software engineering.

The real game is to understand the system well, because in my experience no matter how good or well a system is made, at some point an impossible edge case bug will come and break the system for how incredibly complex these are, also the reason why IT field pays dramatically more than other sectors.

And AI is pretty bad at it, as it lacks context about complex systems which usually comprises a combination of other complex systems. Like in the case of Microsoft it was DB, infra, app, network entangled together for the whole thing to work. And failure could have been at any of the thousands of checkpoints in the whole system.

For even to rollback you need to point out what to rollback. And it took a whole 15 hours for Microsoft to figure it out and probably cost the company the combined salaries of all employees laid off.

It’s the senior people who built the apps have core knowledge about how things work, like for something I build it may take within an hour for me to figure out but for someone who has not worked on it, may take a day even with AI. 

Because AI at best should still be an assistant to an engineer not the developer itself. Unfortunately companies will only realise it after the damage has been already done.

document fetched 2026-11-15 00:23 · canonical https://reddit.com/r/developersIndia/comments/1xvwqni/the_microsoft_incident_proved_what_i_always