The Middleman Problem
71% of engineers feel like translators — reviewing code they didn't write, approving solutions they couldn't have generated.
There is a specific kind of work that has become invisible as work.
You read AI-generated code. You evaluate whether it is correct. You ask clarifying questions. You approve or request changes. You review it again. Eventually it ships. You are the person who made it happen.
Except you are not the person who made it happen.
The problem is not that you are doing something wrong. The problem is that the work you are doing — the reviewing, evaluating, approving, questioning — has quietly replaced the work you used to do, which was writing the solution.
What the data says
From our survey of 2,147 engineers:
- 71% feel like "middlemen" between AI output and the actual work
- 63% said they could reproduce less than half of what they shipped with AI assistance
- 44% said almost none of it
These are not junior engineers. These are experienced developers paying close attention to what AI has quietly done to their work.
Why this is not imposter syndrome
Imposter syndrome tells you: you are capable but you don't feel it. The gap is between your reality and your perception of yourself.
The middleman problem tells you something different: the gap is between your reality and your work. The work you do is not the same work you used to do. You have been doing something genuinely different — and it is worth naming.
The Identity Problem
Software engineers define themselves by what they can build. Not what they can review. Not what they can approve.
When the primary work becomes review and approval, something happens to your sense of yourself as a builder. You start to wonder: am I still a software engineer? Am I still good at this?
The honest answer is: you are doing a new kind of work. The skills it requires are real. But they are not the same skills you spent years developing.
One thing you can do this week
Track the middleman ratio. For three days, write down every significant technical decision you made — not what you reviewed, but what you decided on your own.
At the end of the three days, count: how many of your decisions were "this AI output is correct" vs. "I chose this approach because X"?
The ratio is information. Not guilt.