IT pros weigh in on AI’s role in coding: doing more of what they hate the most

Instead of writing code, programmers complain they now have to proof one written by artificial intelligence (AI). As the New York Times aptly put it, many feel like bystanders to their own job. I decided to run a quick vox populi to see what coders actually think about the so-called vibe coding.

To get a better understanding of AI’s role in coding, I asked a bunch of people five questions, and they are as follows:

1. How often do you have to proof a code written by AI?
2. Do you prefer code-proofing to code-writing?
3. How do you maintain the quality of the code when using AI to write it?
4. Would you trust AI to write the code you need fully?
5. Do you feel the pressure from the management to speed things up?

Funnily enough, those developers who use AI almost every time to write code were the same ones whose answers I couldn’t use because they were written by AI. Having rejected those, I was left with a dozen decent answers, and this article is based on my correspondence with experienced developers who leverages AI in everyday tasks, and seemingly have strong opinions about doing so.

How often do developers have to check AI-written code?

Every developer quoted in this article deals with AI-written code. The scope of it differs, but all of them agree you need to stay sharp when using AI if you don’t want to end up cleaning its messes later.

“Nowadays, I'm more of a code editor than a coder myself. I review AI-generated code daily as if reviewing someone else's pull request,” Holger Sindbaek, Founder and Developer at World of Card Games, told Cybernews.

While most developers say they always check the AI-written code as it is “too risky” to trust AI output, some seemingly don’t bother. And here’s why.

“I generally only need to proofread AI-generated code when it throws errors. That is very rare these days, usually for very recent API changes that the model isn’t trained on yet. It’s amazing how rapidly AI has evolved to need much less oversight and correction,” Derrek Coleman, Developer Relations Lead at Recall Labs, said.

Cursor, the AI code editor, is one of the tools that came up a lot. Developers also mentioned GitHub Copilot, and some custom GPTs they use to vibe-code.

Do developers prefer code-proofing to code-writing?

James Fray, founder of Little Dipper Interactive with a degree in programming, says that checking code written by someone else is their least favorite part of coding, no matter how necessary. With AI, checking the code becomes even more frustrating.

“It is frustrating to have the AI agent repeatedly assert something that doesn't make sense as a change or to watch it hallucinate a function that doesn't exist again. That is all to say, AI paradoxically has me spending more time doing the parts of my work that I enjoy less,” Fray said.

But with AI, at least you are not staring at tabula rasa, a blank slate, where you need to build everything from scratch. And that’s what coders appreciate.

“It’s less about preference and more about embracing a new workflow, and using AI as a collaborator that handles the boilerplate so I can focus on everything else,” Coleman from Recall Labs said.

Many seem to embrace the inevitable.

“Proofreading isn't glamorous, but it is non-negotiable. If you care about clean, maintainable, scalable code, review is just part of the job,” Ilia Badeev, Head of Data Science at Trevolution Group, told Cybernews.

Taras Tymoshchuk, CEO and co-founder of a software development company Geniusee, is more keen on writing the code himself, and it is a more creative process. However, code review is also valuable as “it allows you to notice errors in time and grow professionally.”

How developers maintain the quality of the code when using AI

Well, some, as we’ve already learned, trust AI more than others. But the majority of answers indicate that AI code is going through some sort of review process to ensure quality. Many developers said (I noticed the same in AI-written answers to my questions, too) that they treat AI as they would a junior developer.

“Code generated by AI must undergo automated tests and manual review by senior engineers. This helps to detect non-obvious logic errors or performance issues,” Tymoshchuk noted.

And Badeev emphasizes the importance of correct content and a good prompt.

“Just like working with a junior developer, you would guide him/her throughout the process, trying to be crystal clear about the task and expectations when it comes to the outcome. This includes technical specs, company code style guides, architecture decisions, and examples,” Badeev said.

Do developers trust AI to write code?

In some companies, especially AI-focused startups, Large Language Models (LLMs) seem to be doing the heavy lifting when it comes to writing code. However, there are tons of rules (or so they say) governing the use of AI.

“AI is incredibly helpful for generating common patterns, boilerplate, and even suggesting ways to handle edge cases. However, it still lacks that deep, holistic understanding of our specific domain and the broader context of our applications,” John Pettit, CTO of app developer company Promevo, told Cybernews.

Rodrigo Coutinho, Co-Founder and AI Product Manager at OutSystems, believes you can’t trust AI mindlessly. In fact, it’s less expensive to run a thorough check.

“My experience is that you always need to tweak and adapt the code. Even if there aren’t immediate errors, in most cases, you will at some point need to return to old code — either because of a subsequent error or a requirement change.

“When that happens, you will want to understand the affected code. It's much less expensive to analyze code when it’s created, rather than trying to figure out what you were thinking when you asked AI for that code 6 months prior,” Coutinho said.

Do developers feel the pressure from the management to speed things up?

You need to do more with less.

Management’s requirement to optimize processes, work more efficiently, and grow profit margins is nothing new. However, many developers feel that changes are happening at a breakneck speed.

“AI has raised expectations to a level that sometimes makes it hard to explain why delivering software can take time. The full software development lifecycle is much more than just coding, so even if you get AI to write code ten times faster, that's not the overall speed improvement you'll get,” Coutinho said.

Jacob Anderson, CEO of Beyond Ordinary Software Solutions, is in a management position himself. While he isn’t pressuring his developers to use AI, he thinks it could be useful.

“When they lament frustration about integrating with an API, I remind them to use an AI to help with the integration, and they usually find it helpful. Does it solve their problem 100%? Nope. It does give them a new perspective on the problem and some catalyst for trying new avenues for the solution. That's priceless,” Anderson said.

Maksim Kachurin, CTO at Flexbe, says he expects his developers to use AI in their work because it can significantly boost their efficiency.

“If I’m hiring someone new, I’ll probably favor a developer who actively uses AI tools over someone who’s skeptical. AI is just a tool — and like any tool, when used right, it makes you more effective, regardless of its imperfections,” Kachurin said.

Even the management feels pressure to use AI. It comes with a price.

“Running a game platform means players expect everything to be working all the time, and they expect you to update fast. So I've learned to move forward without hurrying. AI assists here and there, like creating boilerplate or tidying up duplicate code, but it also allows you to skip steps. I've done that one. Things broke. Now I work smart rather than fast,” Sindbaek from World of Card Games said.