Dmitrii Kovanikov
Software developer arguing that AI cheapens implementation while requirements, design judgment and long-term maintenance remain engineering work; forecasts growth in software on demand.
Dmitrii Kovanikov (@ChShersh) is a software developer and the author of “I'm not afraid of AI taking my job.” In the article he reports 16 years of programming experience, study of computer-science fundamentals, and a month of agent-assisted work on an OCaml project. These are source-reported biographical details, not an independently checked profile.
Engineering is broader than code production
Kovanikov separates fast code generation from clarifying requirements, troubleshooting, weighing trade-offs and owning design decisions. He reports producing in an hour a volume of code that previously took two days; this is a personal anecdote about output, not a controlled measurement of end-to-end productivity. See AI-Assisted Software Development and Speedup Illusion.
He finds agents useful for generation, bug finding and explanation, but describes subtle defects, poor readability and a tendency not to refactor without direction—even with static typing and TDD. In his debugging anecdote Claude Code with Opus 4.5 initially denies a bug, then fixes it after he identifies the relevant function. His practical advice is to build a project, maintain it for years, read books and learn to judge specifications and architecture. This complements Critical Thinking with Agents and Code Reading as Cognitive Labor.
Software on demand and limits of the forecast
He expects personalized applications to expand total demand for software: billions of individual task trackers instead of a small catalog of standardized products. This is a demand-expansion scenario for AI-Generated Personal Software, not evidence that paid engineering employment must grow. Distribution, maintenance cost, willingness to pay and who performs the additional work remain unresolved in AI Labor Market.
His predictions include less manual CRUD work, more complex projects and AI-oriented languages/frameworks. The statement that AI is based on “linear approximations” does not establish an enduring technical limit on model capability; nor does the anecdote establish that humans will always be needed for every class of bug. Treat these as the author's speculative rationale, not verified technical or economic facts.
Mentioned in 3 pages
Engineering with agents
Personal AI and knowledge
Work, teams and companies