A junior game programmer with a five-year degree landed a freelance gig in their dream field — then two new devs pushed AI hard “for the sake of going fast.” Architecture, systems, and docs are all AI-written, and the breaking point was a submit of 300+ files, some over a thousand lines. The OP avoids AI to actually learn, gets told they are “too slow,” and is weighing a move to better-paying non-game programming.
The thread’s durable points:
- The velocity trap: daily pushes of AI-generated files leave a junior no way to understand the system they are supposed to maintain
- “Trust AI blindly”: the studio uses AI as a replacement programmer, not a tool — the opposite of “think about the overall system, then code”
- Cognitive debt: long-term over-reliance on AI accumulates understanding debt — “all of the people who actually know how to make things” get replaced by people paid less to tell AI what to do
- Delegation kills fulfillment: when you delegate the work, you stop understanding your own changes — regulated industries like pharma offer a few more years of insulation
- Review-process failure: nobody can meaningfully review 100k+ lines in one MR — either the process adapts (automated review, smaller MRs) or you decline responsibility
The OP is not anti-AI — AI as a tool is fine; AI as a programmer with blind trust is not. The practical advice holds up: fix the PR process or formally refuse to review unreviewable merges, and protect your own understanding, because that is the career capital.