When Code Got Cheap, Reviewing Got Expensive

**Viktor Farcic** (0:00)
People who maintain project and are not good enough with agents to fight agents. And this is the key, right? If you think that your manual review process is going to work in today, you're terribly wrong. And I feel that's where most of the contention comes from. Without AI, 6,000 lines of code is very different than with AI.
When you say with AI, I mean AI is a maintainer, is a reviewer. I feel that many maintainers are actually falling behind, and that could be the issue.

**Darin Pope** (0:40)
This is DevOps Paradox, episode number 361
When code got cheap, reviewing got expensive.
Welcome to DevOps Paradox. This is a podcast about random stuff in which we, Darren and Viktor, pretend we know what we're talking about. Most of the time, we mask our ignorance by putting the word DevOps everywhere we can, and mix it with random buzzwords like Kubernetes, serverless, CI-CD, team productivity, islands of happiness, and other fancy expressions that make us sound like we know what we're doing. Occasionally, we invite guests who do know something, but we do not do that often since they might make us look incompetent. The truth is out there, and there is no way we are going to find it.
Yes, it's Darren reading this text and feeling embarrassed that Viktor made me do it. Here are your hosts, Darren Pope and Viktor Farsic. Imagine you're an open-source maintainer. You get this PRN from somebody you've never heard. It's about 6,000 lines. All the tests pass, all your tests pass, all the tests they added pass. You look at it, you can see that it was all generated, not assisted, generated. What are you going to do with that PR?

**Viktor Farcic** (2:02)
How do you know that it's generated and not assisted?

**Darin Pope** (2:06)
Because I've been seeing all these generated PRs come in, that look like a bunch of crap. That's how I know.

**Viktor Farcic** (2:11)
Yeah, but I've seen assisted code that looks like crap.
AI assisted code, just to be clear. I've seen non-AI assisted code that looks like crap. It's just that people rarely are capable of creating so many lines of crap for a PR. That's the big difference. But when it comes to 6,000 lines, I would actually be very grateful if somebody would tell me, how do I distinguish AI assisted from AI not assisted or autonomous AI code? I cannot distinguish those.

**Darin Pope** (2:40)
Again, this was a scenario. It may or may not be true.
Today's conversation is sort of based around what somebody wrote up as an open source maintainer. And it's basically saying, I just don't want your PRs anymore. I just don't.

**Viktor Farcic** (2:55)
I think it's crap, to be honest. So I feel that the responsibility of maintainers in open source projects is to guide and help and bring people in to contribute. That's the first responsibility over anything else. So whomever thinks that you're a maintainer and your job is to primarily write code, I think you're in the wrong business. It's not business, but you're in the wrong place. And I think it's beautiful, really beautiful, that people now, with help of agents or agentic AI are more willing to contribute because it's easier, because it's less daunting and because it demands less of your time and so on and so forth. It's beautiful and I'm fully behind it. And I think that we should support people creating those PRs, even if that means that we need to spend more time on it. Pick any of the projects I worked in or working on.
If 100% of my time would be sifting through massive number of PRs created by others, I think that's beautiful and I would consider it mission accomplished. With or without AI, does not matter, really. Now, if you put AI into, let's say, Exclusive AI, AI helps me review code like I was never able to review.
I can easily find inconsistency in code itself, and I can focus on architecture, and I can focus on the feature itself, and that's beautiful as well. I love it because before, I could not spend enough time on those things. I could not spend enough time on architecture. I could not spend enough time on features themselves, because by the time I review the code itself, to find neat peaks and silliness of the code itself, I'm so exhausted and so much time passed that I cannot actually spend time on things that really matter. So both things are great. CodeRabbit, Greptile, whichever one you're using, CodeRabbit, they can eliminate the obvious issues.
I don't have to look at those. Or code the issues. I don't have to look at those. And I have more time thinking about the things that matter for real on that project. I love it. And I'm pretty sure that I'm in a minority of people on both accounts.