AI takes over a lot. From developer to product manager.

Scroll

I haven't coded properly for a year and a half. I don't even look at the code anymore. I discuss plans with AI, let them communicate among themselves — Claude with Codex — and test the result as a user. When something's wrong, I notice it when using the software, not when reading code.

This might sound like a confession. But it's not. It's an observation. And I'm far from the only one who sees it this way.

The voices are getting louder

Andrej Karpathy, founding member of OpenAI and former head of AI at Tesla, coined the term "Vibe Coding" in early 2025 — programming by telling AI what you want and forgetting the code. Just a year later, the industry is already talking about "Agentic Engineering": developers become technical supervisors orchestrating autonomous AI agents. They write, test, and deploy the code.

Jensen Huang, CEO of NVIDIA, nails it: "The new programming language is called human." His message? Domain expertise is becoming more important than the ability to program. Whether biology, agriculture, or education — whoever masters their field and uses AI as a tool has greater leverage than someone who can only write code.

Sam Altman from OpenAI describes how a single developer with AI tools will soon accomplish what takes an entire team today. He emphasizes how rapidly software development is changing — the productivity jumps between 2025 and 2026 are enormous. And Pieter Levels proves it in practice — a solo developer making several million dollars a year with AI tools. He built his flight simulator fly.pieter.com in just 3 hours with AI. Today the game makes over $75,000 to $100,000 a month at its peak.

My own experience

I'll be honest: I'm a hobby coder. Still. But my projects run into tens of thousands of lines of code. I haven't properly programmed myself for a year and a half. I've forgotten a lot. But that's exactly the point — it matters less and less. What counts is: Do I know what should be built? Can I tell when AI is producing garbage? Can I judge the quality?

AI takes over the developing. It already does most of the planning too. With idea generation it's still vague — humans still have the clear advantage there. But the direction is clear: More and more coding work is disappearing into the hands of AI agents.

What remains is product thinking. Asking the right questions. Deciding what gets built and what doesn't. Understanding what the user needs. That's product management — not in the corporate sense with Jira tickets and roadmap meetings, but in the actual sense: building the right product.

Should you still learn to code?

Yes. But not for the reason most people think. You no longer learn to code to write code. You learn it to understand how software works. To recognize when AI makes a mistake. To ask the right questions when planning with AI.

Those who trust blindly will be blindly misled. I don't see AI's mistakes in the code — I see them when I use the software or when I discuss the plan with AI. This judgment doesn't come from nowhere. It comes from having understood, at least once, how things work under the hood.

Will "developers" still exist in five years?

I honestly wonder if the term will even exist anymore. And if so, then with a completely different meaning. Maybe "developer" will become what "webmaster" was — a relic from a time when humans did the work that machines now do.

The new role is: thinker, decider, quality checker. Someone who knows what should be built and recognizes whether it was built well. Less craftsman, more architect. Less coding, more judgment.

AI takes over a lot. And that's okay. As long as we don't forget that "taking over" doesn't mean "understanding."

Sources