Can a junior already be replaced by an LLM? And where will senior developers come from? IT professionals weigh in

The number of entry-level positions remains low, with some suggesting that junior roles could be optimized through low-code/no-code tools.

«Previously, Agile transformation was popular; now it’s AI transformation»

Andrey, Product manager with 15+ years of experience

— The fact that IT management has decided, «We’re reducing junior hiring, and their work will now be done by mid-level and senior employees using AI tools.» is a reality. We see this in the numbers.

Yes, juniors are needed for continuity. But companies want to cut costs, so AI adoption is happening rapidly.

I think the next six months to a year will be experimental. Many companies are trying to understand how to use AI effectively—because without proper skills, it won’t deliver the desired results. For example, a recent MIT study showed that 95% of companies don’t see value from AI implementation—only about 5% of pilot AI programs achieved rapid revenue growth, while the rest had virtually no measurable impact on profit and loss (the study is based on 150 interviews with executives, a survey of 350 employees, and analysis of 300 public AI implementations).

For instance, previously, to bring a new idea or feature to market, as a product manager I needed to think, research user opinions, review support tickets, conduct market research, talk to users, visit sales and marketing departments, consult with developers about technical debt. Then form hypotheses and take them to designers and developers.

This was an enormous amount of work that took weeks or months. Now this cycle has been reduced to days—thanks to AI tools.

AI helps create Excel spreadsheets, prepare questions, test them in «synthetic» interviews before talking to real people; prioritize features; analyze user response semantics, detect mood, identify pain points to target, and even suggest new hypotheses. And finally, formulate requirements.

These tools are beginning to act as conversation partners that complement your view of the situation. Yes, you need to verify their output—but it’s still work output that can be reused to some extent.

I’d say it’s already about 60% effective. Previously, I would have asked a junior to perform these tasks—now I handle them myself, just with the help of the right tools.

For business analysts, implementing generative tools also helps reduce work time by 20-30%—if used correctly. It all depends on the people, context, which AI tools are allowed on the project, and what data can be shared.

Conclusion: yes, a junior can already be replaced for certain tasks in product management.

If you want to remain competitive, I would advise studying generative tools and learning new work approaches. Developing human relationship skills is especially important.

«I would advise current QA juniors to explore areas adjacent to AI»

Alexander, Senior QA Engineering Manager with 15+ years of experience

— Recently, our company decided to move toward implementing AI tools across all development and testing processes, purchased licenses for various AI agents, LLMs, IDEs, plugins, and launched an employee training program.

We’ve been in this transformation process for about six months now and have concluded that fully replacing QA as a class is still not possible. But optimizing work or performing certain QA tasks—absolutely.

Ultimately, from our entire AI toolkit, we kept GitHub Copilot, which in the enterprise version can work both as a plugin for LLM chat and as an AI agent for analyzing source code in repositories and generating code or code changes directly in the development environment. Among other things, we use Cursor and Claude Code as additional development environments to analyze results produced by various AI tools and LLM models.

From our experience with GitHub Copilot, it’s already quite applicable for test automation and writing API tests. The GitHub Copilot agent can study the source code in a repository and, based on described instructions, requirements, or test scenario text, create a new API test, making all necessary changes to the project code.

But with other types of test automation, there are still nuances.

Full automation of UI testing or mobile applications proved to be beyond the capabilities of the AI agents and LLM models we tested—both due to the complexity of existing code and the inability of AI agents to access page/screen content and effectively search for elements. We haven’t found a ready solution to this problem yet, so we’re moving toward developing our own MCP server.

Using GitHub Copilot for writing UI tests, the code quality and accuracy of results still leave much to be desired—less than 40-50%, which I consider a poor result. A junior «crafts» such tests maybe not faster, but better.

Mobile applications are even more challenging, with code quality and result accuracy not exceeding 20-25%.

Any business aims to optimize expenses. Why would a company need 50 juniors if instead it can keep 5-10 mid-level specialists who, with AI tools, will do the same volume of work?

Another warning sign for QA juniors may be that companies are increasingly reluctant to take the risk of hiring a QA junior—even one with AI tool experience. Primarily because they lack deep QA discipline experience/knowledge and can’t analyze the quality of results from various AI agents and LLM models (or perform very poor quality analysis).

I’ve personally seen examples where instead of the expected 200-300 lines of code changes, a QA junior using AI tools ended up with 1,200 lines through template code generation. The proposed solution was suboptimal with «broken» patterns. It was quick but inefficient—potentially causing major problems in the long run.

I believe that over time, opportunities for people without experience/QA juniors to enter IT and QA testing and automation in particular will only worsen. The number of vacancies and people needed in this field are already decreasing.

I think very soon, AI tools will satisfy the existing demand for API testing and automation, then UI testing and automation, and later they’ll get to mobile apps and more complex things. Progress will eventually prevail.

AI Models Testing is currently a very fresh and promising area that only a couple of people in several small startups are working on.

«There’s also a pessimistic scenario»

Ilya, Senior Full-Stack Developer (PHP, JS/TS) with 11 years of commercial experience

— Partially, yes. A junior always needs clear instructions for completing tasks, verification, and assistance.

The same work is performed when using LLM—only it generates code faster than a junior does, and almost for free.

The largest companies always hired juniors with future potential to grow into mid-level developers. The reduction in vacancies at such companies is a consequence of budget cuts. Why hire juniors when there are plenty of mid-level and senior developers on the market?

I’ve been using LLMs for a year and a half, more actively lately. First ChatGPT, then Copilot, and now Supermaven for autocompletion and Cursor.

At work, I primarily use them for creating new pages: LLMs generate them excellently in accordance with old ones. Each generation saves me at least an hour, sometimes twice as fast as if I did it myself.

Most of the time is spent checking the result, minor refactoring, and fixing bugs. Fortunately, LLMs make relatively few bugs—sometimes there are generations after which everything works right away.

For simple algorithms, I always try to use LLMs. Yes, I could write it myself in 15-30 minutes, but with LLM it takes seconds plus a minute to visually check.

AI isn’t very good with frontend, as it can’t test the generated layout (or I haven’t found such a service yet). Therefore, perfecting the layout and making subsequent adjustments are still tasks for humans. At this stage, this can definitely be assigned to a junior.

LLMs are also poor at fixing bugs. You can explain everything, yet the model still fixes the wrong thing or looks for bugs where there aren’t any… I’ve tried a couple of times, have little experience, and don’t even attempt to assign bug fixing to the machine since my previous experience was negative.

PHP can’t debug either—and maybe JS can’t as well, at least Cursor didn’t offer me that. So debugging is still a human task for now.

But to my surprise, it turned out that LLMs make minor fixes at a random point in a project faster than humans.

I have a medium-sized project. Since I’ve worked on it for only two years, and I have 4–5 projects in total, I don’t know all the details. I asked an LLM to make a fix—and it did it in a minute. I would have spent at least 10 minutes just figuring out where the code was located. I decided that LLMs should be tasked with code searching; it’s easier for them.

There was also a bug related to syntax. The code was simply syntactically incorrect. On the frontend, the layout was broken because of some quotation mark. The programmer overlooked it, I didn’t immediately notice it either, but the LLM instantly fixed the problem.

Regarding junior training and prospects—it’s a complex issue and not particularly relevant to me. On one hand, I think juniors or students should be prohibited from using LLMs so they learn to code themselves. On the other hand, in 20 years, possibly no one will write code manually.

It makes sense to me that in the future there should be a shortage of juniors and mid-level developers, as the flow of people wanting to enter IT should simply decrease. And yes, there will be a shortage of seniors when the amount of software and budgets increase.

Programming doesn’t stand still: new languages and frameworks will appear, and programmers in one form or another will remain in demand.

But that’s the optimistic scenario, based on the past. There’s also a pessimistic one, in which AGI appears next year. Suppose there were super-intelligent agents and possibly a new wave of programmer layoffs. But for now, that’s science fiction.

«An operator with domain knowledge is needed anyway»

Victor, Senior PHP Developer with 8 years of IT experience

— I think someone needs to initiate and explain the task to AI and then verify the correctness and accuracy of the work. So an operator is needed anyway.

A junior will eventually grow into a mid-level developer (if they don’t become lazy). In general, one way or another, an operator with domain knowledge is still needed.

I use GitHub Copilot daily to improve productivity. For example, for writing technical documentation for API applications. However, I still have to scan every line with my eyes and check the quality, which requires accumulated experience. Therefore, an AI operator still needs to check the quality and functionality of the code.

In general, AI saves me every day with simple routine tasks, but today it still needs constant verification.

I tried writing complex tasks (a large and complex SQL query) using Grok—it couldn’t handle it. It reached cyclic errors—fixing one error, it created another, and so on in circles. I had to assemble it piece by piece myself.

It’s like higher education for two students: one just attended classes for the diploma, while the other absorbed every lecture—who will remain in the specialty in a senior position? I think it will be the same with LLMs.

If a student aims for IT, LLM is an awesome textbook for digesting material. Use it in work only where you’ve already done things manually. In that case, experience will accumulate, and LLM will be a good assistant.

How can beginners increase their value in the job market now? Good question, probably one that should be asked to recruiters. I spent half a year breaking into IT, but that was 6-8 years ago.

I think today the best indicator of a junior’s quality is their portfolio. In interviews, you can advantageously describe your projects in terms of the experience gained, showing the interviewer the quality of your thinking (it’s easier to talk about your own project than one made up on the spot).

Yes, I’m talking about pet projects primarily. Even as a mid-level developer, pet projects make you delve into the depths of technologies. It’s like owning a car—once it breaks down, you’ll have to look under the hood. And that’s valuable experience.

For a student or junior, this will be uncharted territory, but going through even a bit of such territory and showing it to a potential employer immediately adds 100 points.

«Could there be a shortage of seniors in the future? I can only hope so. It would generally be better for me personally»

Vlad, Senior Software Engineer, Fullstack (Python, Go, JS, etc.) with 15 years of experience

— A junior will grow into a mid-level developer and leave for another company, as changing jobs is almost always more beneficial than a new position in the same company. But an LLM won’t go anywhere.

I use LLMs to the maximum—for writing code, tests; translating to another language; UI adjustments. I discuss new feature architecture with Claude and ChatGPT. I use them to create interactive demos of my ideas. I increasingly use ChatGPT instead of Google for information searches.

It’s excellent for quickly sketching an MVP of a new feature. But control is needed, and you must have enough experience to understand when an LLM has produced an acceptable result and when it’s better to discard it and write it yourself.

I can’t blindly trust LLMs, so it seems better to always assign the task to a human, and then they decide which tools to use.

Personally, I have a rather pessimistic view regarding juniors. I’ll answer with the caveat that our company hasn’t hired juniors for about 10 years.

It probably depends on the junior’s skill level. I’ve talked with acquaintances about the level of juniors in their companies. Some could be given a task and solve it themselves. But in some places, juniors are at such a level that LLMs already write better code, and you can give them a shorter prompt than to a beginner.

There was a funny case where, after several calls and explanations, a junior still couldn’t write a test that correctly tests what was needed. Whereas Copilot (not the smartest LLM) managed to do it on the first request.

I have 15 years of experience, and for the first 11-12 years, I think there was a widespread shortage of seniors, and the IT industry was doing just fine.

Of course, in words, we all care about the industry and praise competition as a driver of development. But in reality, everyone tries to create a monopoly—it’s just not customary to say it out loud.