“Code generated by AI does not enter the project in its pure form.” CTO SharksCode on Vibe Coding, AI assistance, and IT development prospects

— Igor, do you use the Vibe Coding approach in your work, and if so, at what stages of development is it most useful?

In a sense, yes — there are times when this approach is quite appropriate. You can «sketch» and test an idea quite quickly, as well as quickly complete a routine task.

— How exactly does the SharksCode team interpret the concept of Vibe Coding?

From our perspective, it’s about using a regular language to guide the process of writing code.

— What AI tools do you use for programming, and why did you choose them?

Models: ChatGPT / Gemini / Claude — the most famous, cope with their tasks. And IDE plugins: Copilot, Cline, roo code — are easy to use, well integrated into the workflow.

— At what stages of development is AI most useful for you: code generation, refactoring, testing, documentation?

Writing documentation, tests, code generation, refactoring simple sections or explaining code.

— How often do you rely on AI when writing code — as a full-fledged «assistant» or rather as a hint?

There are several cases of coding using AI:

1. When you need to test something quickly without quality requirements and follow-up support.
2. It’s convenient to ask AI to explain the essence of the code to understand it faster — sometimes the details prevent you from seeing the general idea.
3. During refactoring, AI sometimes suggests a good approach.
4. It is also useful in cases of routine work — when the structure is clear, but writing the code takes a lot of time. You can give a description, examples — the AI ​​will do the rest, it remains to check.

— Do you always manually check the generated code, or do you sometimes take it «as is»? Have there been situations where the AI ​​suggested a solution that seemed correct, but led to bugs or problems?

Yes, we always check if the code goes into production. If it’s just a presentation of an idea «for our own people,» then you can take it «as is.»

— How do you assess the quality of code written by AI? Is it different from «human» code?

It depends on the depth of the context and the scope of the task. If the AI ​​has enough context and uses linters (the point is that the AI ​​can either run the linter itself in the process or it can run separately from the process) — often its code is indistinguishable from human code.

But, as mentioned above, this does not mean that verification is not necessary — on the contrary, it should be carried out as thoroughly as possible.

— Does the style of teamwork change when developers actively use AI?

I haven’t seen any significant changes yet. Initially, management tried to estimate deadlines based on AI responses, but quickly abandoned this idea — the models don’t take into account all the nuances of development, and they still sometimes have some problems with time estimation and mathematics.

— How do you do code review if part of the code was written by AI?

Code generated by AI in its «pure form» does not end up in the project. First, the developer decides whether to add the code to the project, and then it undergoes a regular code review, where the code is evaluated as regular code by the developer who proposed it.

— Are there any difficulties in a common code style if everyone uses different AI tools?

Linters and code reviews save us from this.

— Does AI help you learn new technologies, libraries, or patterns? How does AI affect your professional growth — does it motivate or, on the contrary, «relax» you?

Yes, sometimes it suggests tools, algorithms, or libraries that I hadn’t noticed before.

As for motivation, it’s somewhere in between: it’s gotten easier in some aspects, but it’s allowed me to focus on higher-level things.

— Does your company have internal rules regarding the use of AI — for example, not to merge code into open models?

Rather, these are recommendations and exchange of experiences.

The developers are aware that the code can be used to further train models or for other purposes. Therefore, we try not to transfer know-how unless the AI ​​is hosted locally.

— Do you see a threat that AI will reduce the value of «classical» coding skills?

Perhaps in the future it will be so, but for now that time is still far away.

The only real threat I see so far is that it has become more difficult for young people to get into the industry, and therefore more difficult to learn further. I think that over time, business will realize this and start creating opportunities for newbies to further their education.

«AI will get smarter if the speed of light is enough to transmit signals in chips.» CMO, STO and Art Director SharksCode on the prospects of AI and smart tools that make work easier in IT

On the topic

«AI will get smarter if the speed of light is enough to transmit signals in chips.» CMO, STO and Art Director SharksCode on the prospects of AI and smart tools that make work easier in IT

Have you heard about SharksCode? Here’s what IT professionals who have already joined say about it

On the topic

Have you heard about SharksCode? Here’s what IT professionals who have already joined say about it

Read the country’s main IT news in our Telegram

On the topic

Read the country’s main IT news in our Telegram