About This Interview Series

HI-driven AI explores the partnership between humans and artificial intelligence through conversations with people who use AI in their daily work. Our mission is to understand how AI works as a partner and collaborator, not just another tool.

These interviews are conducted in the respondent's preferred language and formatted with AI assistance for clarity and translation. Each respondent reviews and approves the final version before publication, ensuring their perspectives are accurately represented.

Attila Essig

Senior Software Developer & CMS Specialist

Attila is a software developer with eight years of experience at a Hungarian SME specializing in custom application development. His expertise lies in content management systems, first with Sitecore, then transitioning to Umbraco, an open-source CMS built on Microsoft's ASP.NET framework. In this interview, he discusses his evolving relationship with AI coding tools, from early ChatGPT experiments to daily work with GitHub Copilot and Claude.

How did you first start using AI in your work?

In the beginning, when these tools emerged, I primarily used ChatGPT for searches, whenever something was unfamiliar to me. I also used it for simple data transformations. I rarely tried to have it write code at first. I'd use it more for inspiration, asking about specific problem solutions, and then selecting from or building upon its suggestions.

Initially, only about 30% of what it produced was usable as-is. But this percentage keeps improving. Now, whether it's ChatGPT or other tools, it increasingly produces output that's directly usable with perhaps just one or two small modifications.

Was there an "aha moment" when you realized AI could genuinely help?

Yes, actually I had two significant ones.

The first was with JavaScript. I needed to extend a plugin with a new feature, but understanding the entire codebase, several hundred, maybe a thousand lines of JavaScript, would have been necessary. JavaScript isn't my primary language. I can develop in it, but I'm slower because it's not what I work with daily. This is where AI dramatically accelerated my work. What would have taken me two or three days with lots of debugging, I completed in about six hours.

The second aha moment was about refactoring. When I see a multi-thousand-line codebase that could be significantly improved, but the input is too large to tackle quickly, I started on it manually, but I also sent it to AI. I tried Copilot first, but the result didn't quite satisfy me. Then I sent it to Claude, and it produced something excellent. It wasn't perfect on its own, but it was great inspiration. I used about 70% of the code it generated, then restructured it slightly to match how I wanted it.

What's your strategy now when approaching a task?

Honestly, there's no conscious pattern yet, but I'm trying to build one. Before each task, I now deliberately think: can AI help me with this in any way? Can I delegate any subtask or even the whole thing?

The types of tasks I find most delegatable are ones where there are similar examples already in the codebase. When I need to build something similar to what we've done before—a new module, an export/import feature—where the context already exists in the code, those tasks can often be delegated. Same with features that follow established patterns like controller, service, and models in ASP.NET. If it fits within those frameworks, AI can handle it well. For completely new features that don't quite match the existing context, you can still use AI, but you need to provide much more guidance.

You mentioned working with an MCP server for Umbraco. How does that fit in?

This is still quite new—we're exploring it for when Umbraco 17 launches with long-term support in November. The idea is to help content editors work more efficiently. Instead of clicking through the admin interface for hours, they could use natural language prompts.

For example, imagine a cosmetics company website. A content editor could simply say: "Put a 25% discount on face creams this weekend, end it Sunday night, show a banner at the top of the page, and send out an email notification." Instead of six hours of manual work configuring all that, it could potentially be done in 20 minutes with an AI prompt—plus time for review and verification.

We've tested similar scenarios and the results were definitely impressive, though this is still theoretical until we can properly deploy it.

Have you had negative experiences with AI tools?

I haven't had major "epic fail" moments, mainly because if I don't achieve some result quickly, I give up on delegating that particular subtask. So I never got to the point of wasting huge amounts of time.

What I notice lately is that Copilot feels quite slow. When it's working and doesn't produce good results, I just see that I waited several minutes for nothing. Then I just do it manually. I'm not sure how this will change in the future.

What do you do while waiting for AI to process?

This is an interesting question. I haven't figured it out yet. I watch it work, I wait, then maybe I pour some coffee, or log my hours, or check Teams. But I feel like when I'm waiting for AI and it ultimately produces nothing useful, it's just wasted time—because I didn't do anything productive while it was processing either.

The best approach I've found is to use that time to build context for my next tasks. I'll start a search in Perplexity, do research, essentially prepare the prompts I'll need for upcoming work.

How do you use AI outside of work?

Mostly as a Google replacement. When I'm researching anything—vacation planning, products, home setup improvements. I use ChatGPT and Perplexity for ideas and inspiration. You always need to verify what they return, but they definitely give you starting points.

I also mix electronic music as a hobby. Recently, I moved to a smaller home office with poor acoustics. I measured the room dimensions and asked AI which frequencies might create standing waves. Based on its calculations, I adjusted the EQ of my speakers. I verified a few of the predictions, and the sound genuinely improved. It still isn't perfect because the room's acoustics are fundamentally challenging, but what it suggested actually helped.

I also use it for recipes—when I have a few ingredients at home and need ideas for what I can make in 20 minutes. For inspiration across many areas of life, it's genuinely useful.

What's your prediction for software development's future?

Let me be honest about my internal feelings first: I don't exactly love this AI era. Part of me would be more comfortable in a world without AI. Maybe it's just the feeling that I need to learn yet another thing, or that something could partially replace me. But I always end up at the same conclusion: this is how things are moving, so I need to adapt.

I think it will be harder for fresh graduates to find jobs. Education will probably need to transform too. The juniors who do succeed will be those who can truly rely on AI and know how to use it well.

I don't think programming will disappear—it won't go the way of elevator operators. But the role will transform. There will still need to be experts, because while AI can perform many tasks, you need expertise to verify its work. AI makes mistakes. In large products, there might be a bug in just one line that even the AI can't detect. And what about those sporadic bugs that only appear every few weeks and are hard to reproduce? Someone still needs to understand the fundamentals to debug those.

The market will dictate this. Companies that don't adapt will fall behind. Whether small or large, this is about survival—so I don't think there will be much difference in how they adopt AI. Some smaller companies might fade away, but I don't think large corporations will disappear. Those who stay agile will survive.

Any advice for junior developers starting out now?

Honestly? I wouldn't want to be a junior right now. If you have a job, keep it and gain experience. Switching jobs as a junior is risky at the moment. Try to get at least two years under your belt so you're no longer considered a pure junior.

But I'd also say: don't lose hope. The market is shifting, but there will still be demand for developers—just in different ways. Be patient, but follow how the market evolves. If you're specializing in an area that AI is increasingly automating, consider retraining toward areas where human expertise is still essential alongside AI.

Is there anything else you wanted to share?

What I find uncomfortable about all this is that AI's output isn't fully predictable. In programming, you typically expect the same output every time. With AI, that's not the case.

Also—and this is harder to articulate—there was something exciting about building code piece by piece, constructing something gradually. Now it feels like I have to look at everything from further away. You don't need to understand all the parts for the whole to function. That's fine, but... I enjoyed staying close to what I was writing. This shift takes away some of that feeling of having created something.

I suppose this is a completely different paradigm for programming as a profession. It's not like running a linter or any other deterministic tool we've used before. And that takes some getting used to.