Anyone can prompt. The 10x engineer knows when it's wrong.
LLMs write the first draft. What sets you apart is the knowledge underneath: memory, caches, complexity, security. 101x drills those basics until you can steer a model in production and prove its output right or wrong.
Vague prompt
fix the memory leak
The model swaps in a static buffer. The visible tests pass. A hidden test fails. The bug ships.
Precise prompt
in log_greeting, free g after fprintf: the caller owns it
The buffer is freed where it is owned. Tests and memory checks pass, and you can say why.
What you train
Read what the model wrote
A patch can pass every visible test and still be wrong. You learn to see why.
Direct with precision
Vague prompts get plausible answers. Name the owner, the invariant, the limit.
Trust checks, not confidence
Tests and memory checks tell the truth. The model only sounds sure.
Courses
Memory: who owns what
Leaks, dangling pointers, double frees, overflows. Learn to read what the memory checker says and tell the model exactly who owns every byte.
5 exercisesA completion code shows that this account fixed the bug and explained it. It cannot prove nobody helped. We would rather say so.