// AI Slop vs The New Engineering

The world is divided. Don’t ask me to give exact percentage differentials, but a load of people love using AI and a lot of people hate it. That divide isn’t any less prevalent in Software Engineering. Let me preface this by saying I’m a heavy user, in a company that is actively pushing agentic first coding. This post will probably be biased.

That being said, there are still people in the company who are unsure about the future of our profession and are pretty much crapping themselves. Whatever happens, happens. What I do know is that devs who are heavily using AI will be light years ahead of devs that aren’t.

I know this because it’s been true for me. Recency and personal bias are cropping up here, of course, but I was never this efficient as an engineer without AI. We’re lucky in my company. We’ve been given access to a huge variety of tools. This post isn’t about rating them, so I won’t, but I will say how I use them and what for.

For the post’s sake, let’s go over engineering before, in the glory days. You picked up a ticket. You worked on said ticket. You debugged, Googled, read documentation, fiddled and prodded until the work was done and tested sufficiently. You’d get it into review, get some approvals, go through CI jobs and JIRA board requirements, whatever they may be, and then merge the work.

Synchronous but unoptimised, and far too inefficient for the new little old me.

This is an example of a typical day’s workflow as of now. I pick up a ticket. I give the requirements to Cascade (Windsurf) and ideate solutions with it, then it goes off and starts to cook. I get Devin up, I ask him (yes him, he my boi) to pick up PAR-124 from the backlog and work on that until it’s completed, including passing the CI jobs. I’ll come back to check on him later.

Simultaneously, I have Codex doing a daily scan of my codebase for quick wins. It automatically does small refactors, adds missing tests, or correctly types something. Then I also have ChatGPT up, solutionising on an epic and helping me word tickets. Meanwhile, I have Unblocked open, asking it clarifying questions about how my codebases communicate with others within the company and linking that back into my ChatGPT conversation to finalise ticket details.

The thing is, the pessimists or the miserable out there are going to read that and immediately assume I’m exaggerating. I’m not. This is my new level of productivity. This is how I believe you should be using AI in Software Engineering.

Now, I’m not going to beat around the bush here. There’s a very specific thing that needs to be taken into account when using AI in the manner that I do. One specific rule. One god-like guiding star that should never be forgotten or strayed from.

You MUST understand what the AI is outputting. You cannot, and should not, be blindly committing code and pushing MRs into production that you have not treated as if a human being has written them.

Yeah, Cascade is on my ticket, but I’m prompting back and forth, analysing its output, telling it to change the new interface it’s written, or fix what parameters that endpoint should receive, or how that form input handles validation. Yeah, Devin is working on a backlog item in the background, but when I come back to check on him in three hours and he’s given me his MR, I’m not just merging it. I’m reviewing that bad boy like any member of my team put it up.

Cool, I’m asking Unblocked for company information and workflows, but if I’m unsure, or something doesn’t look right, I’m asking my Staff Engineer for sure. And the Epic ChatGPT is helping me create? You bet your ass I’m quickly going over it with Product when we’re done.

Use the tools heavily, but only if you understand, can validate, and can verify the output. Otherwise, and I do bloody love this term, you’ll just be pushing AI slop out. No one wants AI slop. It’s bad for business, bad for reputation, and just generally a pile of crap bad practice.

So why do it? People still do though. That’s why there’s such a bad reputation surrounding AI in the field. A great manager of mine told me last year that “AI itself won’t take people’s Engineering jobs. Engineers using AI correctly will replace engineers that aren’t.” It sounds like a tongue twister reading it back, but you get the point. I’m also pretty sure he stole the quote, but whatever.

Wrapping back round to the point, the world is divided, but it shouldn’t be. If you’re using AI correctly, your brain isn’t “turning to mush” because you’re still using it heavily to mentally debug and test code. You’re not reducing your productivity by getting something else to do the task for you, you’re increasing it tenfold. And more importantly, you’re not losing your job, because you are the new definition of the job.