My current thoughts on generative AI.
8 January 2026
LLMs are actually great in domains where the set of possible outcomes is constrained.
The outputs get worse as the set of possible outcomes grow.
- Search and summary on a specific set of documents.
- Very specific instuction set programming (ex: SQL).
- “I do the first one; you do the rest.” pattern.
- Repetition that’s just a bit out of the scope of a script.
- For example: I write the first test for one function; you (LLM) repeat that for other functions or packages.
24 Dec 2025
LLMs are transformational productivity tools. Adapt, use them to automate the boring tasks, while you focus on the more fun things.