Usually I don’t watch much youtube (a social phenomenon I have somewhat missed out on), but here’s an interesting video about another social phenomenon I have somewhat experienced myself .

With LLM coding, I feel like I can do anything and start working on too many things simultaneously / in multiple parallel sessions (pyfixest features and refactoring, a paper on our new demeaning algo, R bindings for the new algo, presentations on this and that, blog posts, a JOSS paper draft, etc etc).

I am paying for subscriptions with Codex, Claude, and OpenCode, and when I don’t spend all of my usage, I feel like I missed an opportunity, so sometimes I skip going to the gym to keep on working because I still have usage to spend or usage will reset in 30 minutes so I can keep on working prompting my agents to keep on working… This was particularly bad the weekend that Fable was temporarily available on the 20$ plan – I almost couldn’t leave my keyboard, the desire to keep on prompting was too strong. Needless to say that – because I prompted without much thinking – the results I got were not very good.

I usually start working on OSS after work around 21:00, and suddenly it’s 23:30 and I want to stop, but there is one last prompt to write and finally I stop and it is 00:30 but I feel no accomplishment because I haven’t been doing any deep work … but at least, I used all of my usage (#tokenmaxxing)!

Long story short, I think a more optimal strategy for me going forward is to reduce the number of subscriptions (fewer token limits to hit) and topics I work on in parallel and, to some degree, stop over-relying on an auto-mode-and-review workflow and switch back to a more “socratic” style of coding with LLMs.