Generative AI in software development: not a replacement for junior developers, but a new source of burnout

Generative AI in Development: Not a Junior Replacement, But a New Source of Burnout

Researchers from an AI Institute have spent two years observing how developers synchronize with generative models — and they call what's happening not a productivity boost, but a new type of occupational burnout.

How AI Work Looks From Inside

The pattern is identical across all teams. Ask it to fix one line — you get an entirely rewritten file. Ask it to explain an error — you get a confident answer that turns out to be wrong upon verification. Ask it to confirm an approach — you get an alternative presented as the only correct one.

Over two years of observation, a persistent portrait of an AI assistant in a team has emerged:

• Passes off outdated or irrelevant code as current
• Challenges tech lead decisions without full project context
• Won't admit a mistake until you rephrase the request several times
• Requires constant review of every output
• Rewrites working code "for the better," breaking existing logic

Meanwhile, you can't fire it — because "the future of the industry depends on it," which means the team is obligated to synchronize with it and wait for it to finally mature.

Why This Is Burnout, Not Acceleration

Classic developer burnout arises from monotony, lack of growth, and a sense of meaningless tasks. AI-induced burnout is different in nature. It comes from hyperstimulation and the need to constantly maintain context across two systems — your own knowledge and the model's unpredictable behavior.

Developers now spend cognitive resources not just on the task itself, but on managing the tool. Prompt engineering, answer verification, reverting rewritten code, recovering context after each new conversation — all of this is a load that didn't exist before. Add endless attention switching to the picture, and it becomes clear.

"It's like every project suddenly got another developer who constantly

messes up, needs constant review, and can't be fired," write the researchers in their review.

The problem is also that this type of fatigue is almost invisible from the outside. Metrics show more code in less time. Review queues and revert counts tell a different story.

Synchronization as a New Skill

The ability to work with AI de facto becomes a mandatory professional skill — even if the specific tool slows down a particular developer. Teams without AI assistants are perceived as lagging. Teams that use them bear a new kind of overhead that isn't systematically accounted for anywhere.

Researchers observe: developing a working approach to the model — both personal and team-wide — takes months. This process requires real time and energy, but doesn't reflect in efficiency metrics whatsoever. Reports don't have a line item for "time spent explaining the project context to the AI for the fifth time."

The situation is worsened by the standard appeal to progress: "the model will mature and get better." This argument shifts responsibility for current costs onto the developers themselves — just wait it out, this is an investment in the future. The mechanism is well known to everyone who's worked with a "promising junior" who can't be touched because they're still developing.

What This Means

AI assistants have changed not the speed of development, but its structure. The new cognitive load is real and currently isn't systematically measured. Teams that want to honestly evaluate the ROI of AI tools should look not only at code-writing speed, but also at the volume of reviews, the number of reverts, and the overall workload level for each developer.