Treating LLMs as Newbie Contributors to Your Development Project

The @twoscomplement podcast's most recent episode

https://www.twoscomplement.org/podcast/speed_of_thought.txt

is not only engaging and thought-provoking, but contains actionable advice from Ben Rady about how to safely and productively engage with code-authoring LLMs.

TLDR:
Set up a container with a copy of the under-development git repo. Ask agent(s) to improve or complete the code, and to push to git remote hosted in the cloud when the work is complete. As a result, the agent has no direct access to the host development system, and requires only intermittent attention to review merge requests and set new prompts. Ingenious!