Insights from todays yt dive and workflow optimization attempt
https://www.youtube.com/watch?v=8GRmLR__OGQ
- gpt-5.5 is unreasonably economical in the subscription
- not very good at (web) ui (at least by default)
https://www.youtube.com/watch?v=9tGrhrVKCrE
- you can use hardware you already have as persistent hosts for ai agents, not tied to your main machine
https://x.com/SemiAnalysis_/status/2064815044085318040?s=20
- anthropic, openai prob other subscription subsidies are crazy
https://x.com/kimmonismus/status/2074078414705955006?s=20
- gemma 4 is actually as good as sonnet 3.5?
- And the curve isn't slowing down, meaning that llm intelligence isn't constrained by hardware? Can't be fully correct.
Codex popup feature
I've found value in spending the few minutes for setting up the hotkey and using for the first time really the popup feature of codex. It's useful for the little tasks you have throughout the day, and especially on windows where window management is generally more painful. A hotkey opens the popup, and it disappears on loss of focus. (Which is a bit unexpected as you can't find the window using alt+tab, you have to reopen it using the shortcut again, but I guess you can get used to it.) It opens a project-less chat by default, which is a good default in my opinion, and doesn't take up a lot of screen space so it's more handy when you have something to do in concert with other windows (file browsers, image viewers, web pages, etc., etc.).