AI

Painful Vibe coding with Codex

Use of LLMs I’ve been using LLMs for a while, primarily as a buddy that can help me with ideation or noticing gaps in my thinking. I rarely use the LLM output directly because it never feels complete or authentic. For written word, I find it can generate pretty generic wording with little impact or personality. Anything I do take directly, I’ll rework pretty heavily - and it’s just a judgement of whether doing the rework is more efficient than just starting from scratch.

Continue reading

A Policy for AI-enhanced software development

The DORA research group at Google recently came out with an AI capabilities model and capability number 1 is to have a “Clear and communicated AI stance”. Given my experience so far talking to other leaders and on the ground, here’s my proposal for what an effective AI stance could look like: Core Philosophy LLMs should amplify engineering cognition, not automate it This means… Human Ownership: Engineers must remain the authors and owners of production logic.

Continue reading