ИИ стоит компаниям дорого Программистов вернут?

Overview

An instructor, Alexander Grigorievin, comments on a trending claim that companies are “returning programmers to their jobs.” He interprets this as evidence that AI-assisted development is being scaled back because it is expensive and unreliable for large-scale engineering work.

He also positions his remarks as a continuation of an argument he previously made on his channel: modern AI is often a hype-driven dead end rather than a true solution to software engineering.

Main Points of the Analysis

• AI is costly and not a universal replacement. AI requires expensive infrastructure—such as high-performance hardware, memory, cooling, and power. The economics are contested for both AI providers and companies trying to adopt these tools at scale.

• AI helps with coding tasks, but not engineering/architecture. He argues AI performs best on well-defined problems (e.g., writing specific functions or handling classification tasks). However, it struggles with architectural and engineering-level responsibilities, such as:

  • system design
  • user-facing interfaces
  • overall software/hardware architecture

• Loss of control in larger projects. When AI builds a project largely on its own, teams may lose understanding of the system’s internal structure. As edits accumulate—especially modifications to existing functionality—complexity increases because engineers:

  • don’t know how the system is structured
  • can’t reliably instruct AI for effective changes

• Local changes become “renovations in an occupied apartment.” Using an analogy, he argues that starting fresh is easier than modifying a complex, AI-generated system—similar to renovating a building with many layers of prior work while it is still “occupied.”

• AI may be inefficient for maintenance and support. In technical support and customer service, he claims AI assistants can irritate users, particularly premium/B2B clients, and can fail in ways that create dissatisfaction. He cites cases where users expect a real manager or consultant rather than an AI assistant.

• AI can produce critical programming errors. He describes a classroom example where GPT-assisted code—even for a small and simple problem—crashed due to a null pointer dereference, which he presents as proof that current “advanced” AI still makes unacceptable mistakes.

• The broader economic context matters. He suggests the backlash isn’t explained by AI alone, pointing instead to a global recession. As economic conditions improve, he implies IT budgets and hiring cycles may recover. He also warns that YouTube AI hype is partly monetization-driven.

Conclusion

Programmers are not replaceable by AI. AI can still be useful as an assistant for specific tasks, especially in small, well-formalized projects. However, attempts to use AI at the architecture/engineering level can lead to rising costs, delays, loss of control, and quality issues.

Presenters / Contributors

• Alexander Grigorievin (presenter)