Typed languages should be the best fit for useful AI agents. Context is needed for practical LLM applications and type systems provide a ton of context. Compiling code provides a short loop that can help the agent.
Strongly typed languages like rust are even better. Not only do you get great compiler errors for the agent to incorporate, you can easily tell when LLMs are subtly wrong. For example, when an AI agent was writing a function that used a library, it wrote it based on an older version so it didn’t compile.
See also:
- Static types make it easier work on projects sporadically when you don’t have time to page in the whole codebase into your working memory
- Static types might also help prevent mushy systems as AI contributes more to large codebases
- I tried out goose coding AI agent on some end-to-end features which went surprisingly okay
- More agents are coming as the market gets flooded with AI employees
Links to this note
-
I’ve been testing out
goosean AI agent for writing code that runs on your machine instead of as an IDE co-pilot.