My guess is 25% of their developers use AI coding assist. Because as a developer who uses AI almost every day, I can promise you only the most pedestrian code can be written by AI. As autocomplete, it saves me some time typing. But actually writing code from scratch, no way.
Yesterday, I asked it to write some particular code for me to do with multi-threading, and it constantly wrote things wrong, like initializing the access of a database client with the user of a request, which would mean every single user would have the access of the first user, not their own.
I reviewed some code earlier this week that did the same thing with the GlobalExceptionHandler that I suspect was also written by AI. These are sort of insidious in that when you write tests to make sure the code works, the tests will pass just fine.
You have to have a skilled developer to identify those issues because the code looks good, and just about any test an individual developer will throw at it will pass. That bug would have gone to production if I hadn’t caught it. And that’s on top of code that just uses the wrong class or confuses common methods from two completely different classes.
And I couldn’t even get a job at Google when they interviewed me, twice. So you can’t tell me 25% of their code is AI-generated. It’s useful, and a time saver. But it’s not capable of generating reliable code on its own. Certainly not yet, and I believe not ever in this form of AI (maybe AGI if it ever comes about).
That’s the kind of thing I suspect as well. Thank you for sharing your insight/experience. It’s always interesting and valuable to hear others experiences.
My guess is 25% of their developers use AI coding assist. Because as a developer who uses AI almost every day, I can promise you only the most pedestrian code can be written by AI. As autocomplete, it saves me some time typing. But actually writing code from scratch, no way.
Yesterday, I asked it to write some particular code for me to do with multi-threading, and it constantly wrote things wrong, like initializing the access of a database client with the user of a request, which would mean every single user would have the access of the first user, not their own.
I reviewed some code earlier this week that did the same thing with the GlobalExceptionHandler that I suspect was also written by AI. These are sort of insidious in that when you write tests to make sure the code works, the tests will pass just fine.
You have to have a skilled developer to identify those issues because the code looks good, and just about any test an individual developer will throw at it will pass. That bug would have gone to production if I hadn’t caught it. And that’s on top of code that just uses the wrong class or confuses common methods from two completely different classes.
And I couldn’t even get a job at Google when they interviewed me, twice. So you can’t tell me 25% of their code is AI-generated. It’s useful, and a time saver. But it’s not capable of generating reliable code on its own. Certainly not yet, and I believe not ever in this form of AI (maybe AGI if it ever comes about).
That’s the kind of thing I suspect as well. Thank you for sharing your insight/experience. It’s always interesting and valuable to hear others experiences.