DEV Community

Cover image for How to Separate AI Coding Hype from Reality for Real-World QA and Development
idavidov13
idavidov13

Posted on • Originally published at idavidov.eu

How to Separate AI Coding Hype from Reality for Real-World QA and Development

A while ago I posted on LinkedIn that I think we’ve officially passed from the EUPHORIA phase to the ANXIETY phase of AI CODING - link.

Preface

One side hustle of mine is Investing. Initially, I was fascinated by it’s THEORY and PSYCHOLOGY. One of the most famous quotes in the field is: "The trend is your friend."

I absolutely belief that DELAYED GRATIFICATION is the path everyone should embrace. But…we are living in a world, where the latency of 3s, can drive us crazy. We want everything, here, now. We don’t want to sweat, we don’t want to work hard. Instead, we just want vacations, dopamine and a coctail - just for style. Familiar?

In every field there are so few people that actually succeed. No hard statistics, but probably 1 %. Why? Cuz they are just doing that 99% of the people refuse to do.

AI Coding <=> Investing

Let’s talk a bit for the elephant in the room. To be honest, we are all here for that, right?

I just want to add something about the TREND from the INVESTING stand point, because I am trying to make “bad” reference to AI CODING. Here's an explanation of what a TREND is - a direction in which something is developing or changing.

Trend Stages as Emotion

Trend Emotions

For quite some time, everyone has been milking the cow named AI CODING. There is nothing wrong with it, but as always, it is a trend, and it moves in waves.

On the surface, it is fantastic, but let’s not forget about the technical debt we bring to the table by using it.

Back a few weeks ago, when I stated that we are officially passed the EUPHORIA stage and we entered the ANXIETY one. I am using LLMs to learn, brainstorm and develop automation frameworks for more than 2 years, and as a QA I am always skeptical. I try to verify everything that is presented to me. Sometimes, the truth is well covered, and sometimes, it is actually denied. That’s why I do believe that the AI CODING has a dark side attached to it. I already mentioned the technical debt, but the bigger issues are that it ruins problem solving skills, critical thinking and actual collaboration between engineers. The worst part is that we start thinking we DO NOT need juniors. Are we thinking that current Mid and Senior Developers are enough to lead the Earth to greatness? Seriously?

LLMs in QA Field

I try many experiments with Playwright, it’s MCP server, different LLM’s, different IDEs and I can stated that all these tools are helpful. They limit the manual work that I have to do, and they speed my productivity (coding, test case preparation and writing documentation) 3-4 times. But….and there is a BIG BUT, this actually doesn’t speed releasing. It doesn’t speed releasing, because, as before these tools, I spend most of my time discusing requirements, technical implementation, bugs, features, etc.

Currently, I am using Cursor and claude-sonnet-4/chatGPT 5/gemini 2.5 pro as an LLMs, but I limit it in the following:

  1. Tab autocompletion - this workflow gives me the opportunity to keep everything under control, but still improves my speed drastically.

  2. Develop Factory and Helper fuctions - the LLMs are great in short, specific, one function tasks. You can describe your problem well enough, and someone on the world has already solve your exact problem. LLM’s job is just to find the right piece of code and provide it to you. Just like a few years back, but without manually grinding google and Stack Overflow.

  3. Brainstorming Ideas - no doubt that LLMs are great partners in that.

  4. Creating documentation about testing and coding - again no doubt that LLMs are great partners in that.

But what about all the hype about MCP servers giving context to the LLMs, Agentic tools for creating tests alone, and tools, that create test cases from natural language? All these things work only on POC/Demo sites. MARKETING SUGAR COATING.

Nothing of these works in actual Real-World Complex App in development, because the worst thing that can happen to an LLM is to change the context (aka change the requirements of already developed/tested feature). Just do it a few times, and watch the “Michelin Italian Chef” named ${your choosen LLM name} preparing a brilliant piece of spaghetti for you.

Conclusion

My current feeling is telling me that we already passed the ANXIETY phase and we are officialy in DENIAL phase of the wave. This absolute doesn’t mean that LLMs for coding is not helpful. We just start to realize their real value….and real COST.

Before fallin in the trap, ask your self one simple question. Is this an ORGANIC GROWTH or it is a BUBBLE?

Visual Representation of forming and bursting a bubble

Forming and Bursting a Bubble

If you are actually changing the World (doesn’t matter the World or your World) by the use of any LLM, don’t stop! Just be aware of PROS and CONS of it.

Top comments (0)