Stanford’s Erik Brynjolfsson: AI won’t eliminate programming jobs, it will expand the market

The panic around the "death of the programmer profession" may be premature. Stanford professor and digital economy researcher Erik Brynjolfsson believes that AI will more often not displace developers, but strengthen them — and in the process expand the entire software creation market.

Replacement or Enhancement

Brynjolfsson's main thesis is simple: any technology has two modes of impact on labor. In one case, it replaces a person; in another, it makes them stronger and allows them to produce more products, launch more services, and solve more tasks. In his view, discussion of AI today too often focuses solely on the displacement scenario. He reminds us that similar fears already arose with the emergence of higher-level languages, cloud services, and other automation tools, but demand for programmers didn't disappear — it shifted to new specializations.

"The real value lies in asking the right questions."

This is where, by his logic, the center of human role will remain. A machine can quickly execute part of the work, but it doesn't determine which problem is worth solving, for whom, or how to understand that the result turned out useful. Therefore, those who will win are not just those who can write code by hand, but those who can formulate a task precisely, see the business context, and turn a vague idea into a working technical specification for people and AI-tools.

New Roles Alongside

Brynjolfsson expects not a shrinking, but growth of the entire development ecosystem. If in the past creating useful software required narrow engineering skills, with generative tools, many more people will be able to enter this process. He suggests that over time, software product creation will involve many times more specialists, though not all of them will call themselves developers. Some will assemble applications through conversational interfaces, some will manage agents, and some will be responsible for task formulation and quality control.

• "Chief question officer" — a person who formulates the right questions and directions for AI-systems.
• Agent fleet manager — a role for those who coordinate the work of multiple AI-executors.
• Domain experts without deep engineering background — will be able to assemble useful applications for their tasks.
• Developers with strong subject matter expertise — will gain more leverage and be able to do more in less time.

Against this backdrop, the profession of programmer itself doesn't disappear, but fragments into new specializations and levels of responsibility. Just as databases, clouds, and frameworks once didn't kill development, but created separate career tracks, so AI can expand the field for those who can combine engineering thinking, product perspective, control over automation, and understanding of where machines can be trusted with execution and where humans are needed.

Need for Safeguards and Verification

At the same time, Brynjolfsson doesn't suggest blindly handing software creation over to machines. The more "citizen developers" and agent systems companies have, the more important protective mechanisms become: security, privacy, access control, testing, and verification that AI actually did exactly what was asked of it. The irony is that some of these checks can also be taken on by AI-tools, but only within clearly defined rules.

He describes working with AI as a chain of three stages: first you need to formulate a question, then implement a solution, and finally evaluate the result. Today, generative models are particularly strong in the second part — execution. But the first and third parts remain critically human. This applies not only to IT: the same principle, he says, will work in marketing, art, music, philosophy, and other fields where value is born from choosing direction and judging quality.

What This Means

Stanford's argument here is not that AI won't affect anyone, but that the labor market changes in more complex ways than the "bots took the jobs" scenario. For developers, this is bad news in only one case: if they ignore new tools. For everyone else — it's a signal that the ability to set tasks, check results, and work together with AI is rapidly becoming a separate professional competency.