“More than a quarter of new code at Google is generated by AI” [source]
How does this make you feel from a testing perspective?
Drop some hot takes in the comments!
- Excited
- Cautiously optimistic
- Worried
- Skeptical
0
voters
I think this distinction is important:
“Google is building a bunch of AI products, and it’s using AI quite a bit as part of building those products, too. “More than a quarter of all new code at Google is generated by AI, then reviewed and accepted by engineers.”
As far as I’ve seen in the news, Google’s AI products are usually experimental, or of lower value. And people accept the issues that arise as they blame it on AI anyway (regardless if code was written by AI or a human)
I dare them to write with AI the code for the products they generate a lot of money from 
They say that the code is reviewed by developers. What I understand then is that if a developer reviews the code of AI, then there’s a person less to review/think of the code. I’d be a bit worried about the engineering profession, the people and that the products are seen and thought of by one less bright mind.
Btw…in order to do this kind of stuff, they now invest in Nuclear Power:
To think globally about it, who does it help or what does it benefit?
Let’s use AI to generate some unit and functional tests; KPI numbers will look great. But as we know, this won’t help them actually deliver better, more secure, higher-quality features or products. Google moves at a snail’s pace, pouring time and resources into minor changes and fixes - and even major issues can take ages to resolve. Now, they’re using AI to make things even more complicated and slower. For some internal goals and graphs, exec goals, and bonuses, these achievements might look impressive. But from the user and engineering perspective, it’s not exciting.
Yes, they can pull this off at a high cost and without obvious failures, but we know they’re too big to fail. They have some wild and brilliant ideas and solutions, but they only work at Google’s massive scale, they don’t scale down. It’s great that, for example, AI can analyze a billion log lines per second, but will that mean faster fixes or secure, high-quality releases? The same with code, maybe the proper refactoring, removing legacy code/systems, fixing tech debt, implementing proper testing and QA, proper planning and management, etc could remove a quarter of the old code and reduce significantly the amount of new code but it’s boring old-fashioned doubting time-consuming not fancy and trendy engineering task, no one cares about this stuff and again, on their scale and state it would be very challenging to “fix” everything but using AI is much cooler.
PS: on my own small scale, I’ve seen how AI generates code - it works, but the number of lines, “extra” features, and added complexity often go overboard. Many tasks could be solved with half the AI-generated code in much simpler, more elegant ways.
I’ll give them a 60% threshold. Then i’ll start switching.