Juniors dead, vibe coding hot: How AI changed software development

By Dmytro Grechko, CEO at Deskree

You all have seen the news: Artificial intelligence writes 30% of Microsoft's code.

And while it might be an exaggeration on Satya Nadella’s side to sell by example, it’s still obvious — AI has enormously changed software development. And as it moves deeper, it’s changing what it means to be an engineer.

1/ Junior developers disappearing

Today this is almost always the rule: What used to be an entry-level task is now a task for AI.

Just a few years ago, if a company needed to build a new internal tool or automate a process, the task would likely go to a junior engineer first, who’d write the boilerplate code, test it, and debug it.

Today, most of the entry tasks are done in minutes by a tandem of a senior engineer and AI.

This has created a shift in hiring: junior engineers are no longer in high demand. According to Financial Times, entry‑level job postings in the U.S. have dropped by roughly 45% since June 2022. Sectors like software development are particularly affected.

Companies can move faster and cheaper with AI. They don’t need to invest time in onboarding or training someone who’s just starting out.

2/ Demand for strong architects goes up

With AI, expectations for human engineers have changed. Now, apart from coding, developers are prompt-engineering, editing the results, and doing all the other stuff around AI to optimize it.

But there’s one area where engineers are expected to be better than in pre-AI times — software architecture. Knowing how systems fit together, how data flows, and how to balance speed, cost, and reliability – for now, only a human can do this well.

Forbes reports that cloud architects are in high demand as companies scale AI systems, calling them “top hiring priority.” 

Skilled architects build solid infrastructure: cloud setups, containers, CI/CD pipelines, security, monitoring, and redundancy. They design for failure so systems don’t break under pressure.

A strong example is Lovable. Its founder came from CERN and used deep architecture skills to shape the product. AI helped speed things up, but the structure and design came from experience. 

Infrastructure — servers, access control, monitoring, scaling — remains human territory. If you’re an architect, this is your moment. 

3/ Vibe coding on the rise

Vibe coding is an emerging software development style, popularized by Andrej Karpathy, co-founder of OpenAI and former director of AI at Tesla.

Unlike traditional AI coding, where developers fine-tune code line by line, vibe coding avoids micromanagement — it guides AI with high-level instructions. Karpathy called it “forgetting the code exists.”

At some companies, this approach is already the standard. At my company Deskree, engineers aren’t allowed to write code manually. Their role is to prompt the AI, make sure the output meets internal standards, and act as reviewers — essentially serving as QA for the machine.

In our case, we created a system where each programmer has their own Model Context Protocol, created by Anthropik. It sends tasks and technical details directly to the programmer’s AI assistant, which knows all the task specifics.

This gives the AI full context of what’s expected before it starts writing any code.

4/ ‘Chat’ becomes the new interface

The next big change is in user experience. In the past, people typed commands in a terminal, then there came apps, browsers, and buttons. For many years, the interface was something we clicked and navigated. Now, it is a chat.

Users say what they want, the AI understands and does the task. It acts as a middle layer between people and software, turning words into actions. This is a big shift for both engineers and product designers.

5/ Companies to build own system design tools

Writing code is only about 20% of software development. The rest is system design, connecting services, managing infrastructure, and fixing problems. AI struggles because it lacks full context.

Engineers use many tools like GitHub Copilot for coding, Grafana for monitoring, and Sourcegraph for navigation. They cover parts of the process, but as Reddit CPO Pali Bhat points out, engineering teams can only manage a limited number of tools and their “integration” and “consolidation” is inevitable. And I agree with him.

My company Deskree, for example, has built Tetrix AI, a tool to cover all layers — coding, communication, infrastructure, monitoring, and operations — in one place. This gives engineers the full picture and saves time. For sure, more companies will follow suit.

Subscribe to our LinkedIn newsletter to get one useful article every week via email and on LinkedIn. We promise each story to be insightful and take less than 5 minutes to read.

If you are an early-stage startup founder building a company that aligns with our investment thesis and seeking to raise funds, you can write to our team on LinkedIn or send your pitch to F1V using our Typeform. Hope we can work together.