Blogmark

Dax of OpenCode recently wrote:

a weird inversion with LLMs is the models improve faster than the tinkerers

when i see people with custom workflows and setups they're all addressing problems that don't exist anymore

the person naively using vanilla codex is more likely to be experiencing state of the art

I find this to be a challenging conundrum as someone who has typically invested a lot in my tooling and development environment. It seems like as quickly as I learn about and start using a new tool or technique for improving some aspect of AI-assisted coding, it goes out of fashion.

My approach for the past year or so has been: stay curious and try new things that catch my attention. And hold those things lightly and be ready to throw away my assumptions about what works well and what these models are capable of.

I'll also add that the churn can be overwhelming, so feel free to use, as Dax suggests, vanilla Codex or vanilla Claude Code with frontier models. You're still gonna get good results.