AI Coding Agents Squeeze Out Junior Developers, Risking a Talent Gap

Companies are hiring fewer junior software developers as AI coding agents take over tasks that once made up a first job. That may pay off in the short term, but it threatens to leave the industry without experienced engineers a decade from now.

A JetBrains survey of more than 15,000 professional developers worldwide, published in August 2026, found that 90 percent use AI coding agents at least weekly. At the same time, the Stanford study "Canaries in the Coal Mine" reports that employment of US software developers aged 22 to 25 has fallen by roughly 20 percent since late 2022, while employment of experienced colleagues in the same roles keeps growing.

The bottleneck has moved

The argument draws on Eliyahu Goldratt's Theory of Constraints: a system's output is set by its biggest bottleneck. AI can now produce large amounts of code in seconds, so the constraint is rarely typing code any more. It lies in missing decisions, poor requirements, manual handovers between systems, limited review capacity and absent guardrails that keep AI within safe limits. All of these need human judgment, and that is exactly where entry-level training is being cut.

Code review shows why. Reviewers must ask whether code creates security risks, whether it is maintainable and how it copes with changing requirements. Experienced developers can spot where generated code is wrong. Newcomers often lack the yardstick and accept a plausible answer. AI amplifies existing judgment but cannot create it.

One firm's response

The author, writing from the perspective of a software services company, argues against keeping juniors away from AI tools. Training should instead focus on generated code and the review that decides what reaches a customer system. The firm uses mixed teams with one experienced person for every two to four developers, and a buddy system that gives each new hire a fixed contact. It also schedules shared in-person days and pair programming with senior architects, while keeping remote work.

Because AI keeps changing, the company decides project by project with clients where AI brings real value and what safeguards it needs. Systems built today may still be running in 2036, the author notes, so cutting training now affects long-term maintainability. His conclusion: the people who keep training juniors today will be the ones building software in ten years.