ITMO expert Ilya Blagorodov: AI doesn't take responsibility for code off the developer

Ilya Blagorodov has spent 30 years in software development and serves as an expert for ITMO's online master's program "Frontend, Backend Development and AI Solutions," run in partnership with Yandex Practicum. He has published a column on Habr explaining why using AI at work doesn't free a developer from responsibility for their code — and why he himself, despite relying heavily on neural networks, now programs more, not less.

Why AI doesn't replace the developer

The author's central argument: AI assistants — from code autocompletion to full-fledged agents that write entire functions — speed up writing code, but they don't remove the developer's obligation to understand what that code actually does. Responsibility for a bug, a vulnerability, or a suboptimal architecture in production lies with the person who signed off on the commit, not with the model that generated it. That is what separates AI use from simple automation of routine work: the larger the share of generated code in a project, the more important it becomes to read it quickly and accurately.

How an IDE differs from an AI tool

The author draws a line between a familiar development environment and an AI assistant. An IDE is a tool that suggests syntax, catches compilation errors, and speeds up code navigation, but the developer makes the decisions. In that sense, an AI tool can look like a more powerful version of the same thing. In practice, though, it offers ready-made solutions rather than hints — and that makes it easy to lose the habit of thinking through the logic yourself unless you consciously stay in control.

• Author: Ilya Blagorodov, 30 years of experience in software development
• Expert for ITMO's online master's program "Frontend, Backend Development and AI Solutions," run jointly with Yandex Practicum
• Despite actively using AI at work, the author says he now programs more, not less

What to learn to stay in demand

The author describes a paradox many developers will recognize: the faster AI writes code, the more time you have to spend checking that code, understanding the limits of where it applies, and fitting it into the existing architecture. The column comes out in the context of a master's program that trains specialists at the intersection of frontend, backend, and AI solutions — exactly the area where the question "what should a developer be able to do in the era of AI assistants" is especially pressing for educational programs.

The author's position echoes a view heard more and more often in the industry: junior developers who rely on AI tools without understanding what's happening "under the hood" risk getting stuck at the level of copying ready-made solutions, while experienced engineers use the same tools as an accelerator — but check, test, and rewrite the output by hand where needed. That, the author says, is why using neural networks in his own work has increased, not reduced, the amount of code he writes by hand: the more generated code passes through a project, the more effort review and refinement require.

What this means

The column fits into a broader industry debate about how the developer profession is changing as AI assistants are adopted at scale. The tools speed up writing code, but a specialist's value is increasingly shifting toward the ability to understand, verify, and take responsibility for what those tools produce.