This project was made possible using agentic LLMs. The authors work on this project in their spare time, devoting it to rigorous research into why the printers are not performing up to spec. That means taking time from families, rest, or other hobbies to make the situation around the printers better.
Given these circumstances, using every possible tool makes the project possible and publicly available. Large language models help with the sheer volume of work involved: documenting problems, analyzing firmware behavior, drafting fixes, and keeping everything consistent across dozens of pages.
The agents in this case are not used to blindly look for problems in code or to generate slop, but to enhance the capabilities of the authors as engineers with backgrounds in EE, control engineering, and software engineering. Everything published here was rigorously reviewed and tested on our own hardware. We stand by those results and welcome any feedback on further improving the situation.
We base our approach on our own experience and the experience of other senior engineers. See RFD 576: Using LLMs at Oxide for an overview of LLM use philosophy that we agree with. We evaluate negative LLM use consequences on a daily basis and try to find the best possible options, taking all externalities into account.