Do you mean they tell an LLM "push this code", or that the LLM runs the commands as part of a larger agentic workflow. The former is crazy but the latter makes sense.

When Claude says "Shall I push it", it's way easier to just respond "yes" than it is to open a new terminal and run git push, and if you're being graded on how much AI tokens you use, saying yes looks even better for your metrics!