AI Policy

I use AI the same way I think: with nuance. It is a tool in my workflow, not a substitute for judgement. What follows is how I draw that line.

Professionally

1. Every line of code I ship I own. I review it all, line by line.

2. I value quality over speed. LLMs help me move past analysis paralysis and perfectionism, not around them.

3. I assume all LLM output is buggy until proven otherwise. That is not a skill issue or a parameter issue. It is a limitation of the technology, and no amount of guardrails or RAG changes it.

4. I do not vibe code on professional projects.

5. Not all AI code is slop. The skill is in knowing when it is and when it isn't.

6. I prefer locally-run models (Qwen 3.6 27B, 35B) over commercial APIs where practical. On inexpensive, low-consumption hardware, the environmental cost is negligible.

7. Most of what is said about LLMs is hype, extremism, or both. I try to ignore the noise.

Personally

1. I vibe code to test ideas and explore feasibility. I treat all those toy repositories and projects as disposable.

2. When I do share one of these personal projects, I credit the model I used. I have never claimed sole authorship.

Influences

I think the Linux Kernel's policy on coding assistants is the closes to my own views. Debian's proposal 2 aligns well too. And I share some of the concerns the Codeberg team raises about LLM-generated contributions to free software and the environment. My views are very different when it comes to commercial LLMs vs locally-run open-weight LLMs.

These are not contradictions. They are nuance. That is the whole point. Hence the existence of this website.