AI is coming where I work. My employer is cautious and only gingerly rolls it out, but it’s coming.

As a long-time employee, I’m responsible for some of the most critical bits of software in our products - APIs and libraries that are used absolutely everywhere. They’re not particularly complex, but I’ve spent a lot of time over the years to ensure that they are tested to death, and they’re 100% reliable, bomb-proof code. You know, the sort of code you can include in your own code, and if the final product has bugs, you can be completely certain the bug is in your code, not in mine.

I’m one of the few employees who are OCD enough to have spent the countless hours required to make this code as stable as it is. It didn’t seem like a good use of my time at the beginning. But as the years rolled by and my code got used more and more, became more and more foundational to our products, and proved to be totally problem-free, my company and I reaped the rewards: bug-free stuff that can be relied upon.

And so as a result, I’m pretty much the only one working on it. I do offer my changes for review now (it wasn’t like that before), but mostly it’s just to follow good practices: unless I messed up in that code - and I rarely do - what I say goes, and nobody touches it without my approval.

I never planned it that way, it just came to be like that over the years. I’m the de-facto gatekeeper on those bits of code the entire company depends on.

On Monday, one of the new employees who use AI a lot for non-critical side projects submitted a PR to one of the core libraries I maintain. It wasn’t a trick: he freely admitted it was AI-generated, and even commented that he expected me to reject it outright with a smiley. Which I promptly did - with a smiley also 🙂

I had previously indicated that I would NEVER let AI touch that code, because the last thing I need is mediocrity and uncertainty in that stuff. So no great surprise to him. I told him I understood what he needed to do and I would implement it myself.

So I did: it took me two days and it’s done right: it’s coded cleanly and it’s tested every which way intelligently, with proper understanding of what needed to be implemented and tested, and with consideration for backward compatibility with older firmware versions and idiosyncrasies of our test equipment with very slow processors.

For shits and giggles, I took a peek at the AI stuff: yeah, it worked, but it totally missed the deep architecture of the library: it added 50 variables and methods to a class instead of instantiating another class that already existed, that had all the necessary bits. It reinvented the wheel in all sorts of clumsy ways, copied code from other methods just because it was done a certain way there, without understanding that it didn’t apply elsewhere… Yeah, functional. But do that 10 times and you end up with a pile of shit full of subtle bugs. And I wasn’t about to let that happen to my code.

Today, my boss called me in his office for one of those uh-oh…, closed-door one-on-ones:

“I heard you rejected Jason’s code on Monday.”
“Yeah, it was AI and it wasn’t good enough.”
“But Jason said it worked.”
“Yeah, it worked. But it wasn’t good enough. Working isn’t high enough a bar for this code. You want perfect there, not just working.”
“But we lost two days on the new feature…”
“Yeah, but I warned everybody that I wouldn’t let AI touch any of this, and nobody objected. If this is a problem, then we need to have a serious discussion about it. Is this a problem?”
“No. I just wanted to tell you you did the right thing.”

Admittedly, I could only impose no-AI because I’m senior enough, and because the company isn’t exactly thrilled to let the tidal wave of mediocrity lick its boots in the first place. But it feels good to be backed by management on this.