The human coders hired to mop up AI slop

The advent of AI models that can code has led tech CEOs to make some bold claims about the demise of the human software engineer.

So-called “vibe coding” allows anyone to type in their idea for an app or website and get the code they need to build it, no technical expertise required.

But there’s a hitch: The coding vibes are off.

Freelance software engineers say they’re increasingly being hired to fix and clean up “AI slop,” sometimes costing more time and money than if the client had hired a human to do the work in the first place.

“We've got probably 20 or 30 contracts on deck, large ones, a national hospital chain that's having us clean up all of their AI generated crap that another contractor had generated with AI thinking that they were going to cut corners,” said Phil Anderson, senior vice president of development at Seattle Software Engineers.

Anderson said AI is at least a decade away from being capable of producing a workable, scalable app.

In the meantime, companies are building up what’s called “technical debt,” by trying to build software on top of code that’s riddled with errors, bugs, and security issues.

“A lot of the time when you do AI vibe coding, especially when you have to then fix it, it would've taken you an hour to write the code, had you just written it, but instead it took you five minutes to write the code, but now you have to spend two hours fixing the code because it doesn't work,” said Owen Bick, a Austin-based engineering manager for a healthcare company who freelances on the side.

Freelancers call these “rescue jobs” and say they’re on the rise. It’s a claim backed up by data from Fiverr, the online marketplace that connects clients looking for tech help with coders and other freelancers.

Fiverr crunched the numbers for KUOW and found that jobs on the marketplace looking for freelancers who can fix WordPress errors increased 712% between October of 2024 and March of 2025. Clients looking for help fixing bugs on their Shopify platforms more than tripled, and more general website maintenance requests more than quadrupled.

It’s impossible to know how many of those jobs are cleaning up AI-generated code, but the stats taken with anecdotal stories begin to paint a picture.

“ I've definitely seen an increase in frequency of those kinds of requests,” said Mia Kotalik, a traveling freelance engineer who works in Seattle and a handful of other cities. Her current job is cleaning up an AI-generated prototype for a client.

Kotalik, like the other software engineers interviewed for this story, expects AI tools to improve to the point where they make fewer mistakes. But, she said, there are some tasks AI may never be able to do as well as a human.

“ Most of what the core of the skill is, is actually on the design side of the software, architecture design,” Kotalik said. “That’s like choosing what kind of database, choosing what languages you're going to use, what things you're going deploy your app with. All of those choices are part of the design, and they affect speed and price, and all of these things. That’s usually what a human's going to be better at.”

Kotalik said that AI can be a useful tool to help non-technical people bring their ideas to life, but also issued a word of caution.

"I wouldn't tell them don't use AI tools or try and make a prototype yourself," she said. "It's a great learning experience and I've seen a lot of people get funding with a prototype that they made with AI as a proof of concept … But I wouldn't make any of those decisions without consulting somebody who has that technical background."

Despites their shortcomings, AI tools that can code, like ChatGPT and Claude, appear to be cooling the junior software engineer job market.

Recent graduates with computer science and engineering degrees face some of the highest unemployment rates of any major, according to a report from the Federal Reserve Bank of New York.

By avoiding hiring junior engineers, companies could eventually face a shortage of the experienced decision-makers who can supervise — and clean up after — AI.

Software is seen as a harbinger in the predicted “white collar job bloodbath,” and could be a cautionary tale for other industries eager to embrace automation as quickly as possible.

“ It's not ready yet,” Anderson said. “It's in its infancy stage.”

To hear more about how AI is reshaping the local workforce, listen to KUOW's economy podcast, Booming.