“Coding Agents” are generating more work.
Opinion piece.
Everyone expects AI to reduce developer workload. What I’m seeing day to day in corporate environments tells a different story.
The promise is compelling: Copilot, Cursor, Claude Code, Gemini CLI. The message is clear: AI will make developers more productive. It will code everything, including the boring parts. Some say it will even replace junior devs. That’s not what I’m seeing.
What Actually Happens
When AI accelerates development, management raises the bar. The backlog explodes. Everything that “would be nice to have” becomes “let’s test it.” Every prototype becomes “let’s plan for production.” Projects rejected for cost or complexity suddenly seem viable.
The bottleneck shifts. Before, it was writing code. Now, it’s reviewing code.
This reminds me of the highway paradox: more lanes make driving easier, so more people drive, and traffic gets worse. This induced demand is exactly what’s happening with generative AI in code.
The Shift in Developer Work
When you write code, you know every detail. You remember why you made certain choices and their trade-offs. When AI writes code, you become a reviewer. You need to figure out what the AI did and why. Validate assumptions. Refactor patterns that almost work, but not quite.
In practice, most AI suggestions get adjusted before going to production. It’s a validation grind. After all, nobody wants to look like “the AI does their job” when things go wrong.
For routine tasks (templates, documentation, CRUD), AI genuinely reduces cognitive load. But for complex requirements, the dynamic changes. It’s like reading “someone else’s” solution who doesn’t explain their reasoning and sometimes makes choices that look right but aren’t.
What About Job Opportunities?
If building software now costs less, more companies build more apps. But someone still needs to review the code, fix bugs, ensure security, understand how everything connects.
That someone is developers. Demand is growing, but the growth is selective. It favors those with architecture skills who know how to validate what AI delivers.
The workload isn’t disappearing. It’s being redistributed.