Bit more work on Turbo Admin tonight. All hand-crafted.
Think I have three small issues before I get a release out.
Some other, half-done things that need more time will follow.
Still thinking I'll set up a mailing list for customer updates. But it won't be through Lemon Squeezy's email system as their platform appears to be basically defunct and unsupported. Which is crazy.
(Really not sure what to do going forward for sales)
Oh. I do now totally see why solar power and battery storage is a nerd’s dream of a rabbit hole to go down.
If you’re gonna get some solar panels then you probably want a battery. And if you get a battery then you may as well use a flexible tariff to “load shift”.
And… 🤓🤓🤓
I’m imagining a rented, non-deterministic dishwasher that only gets some things clean some of the time, and occasionally makes things dirtier.
– You can rent three of them and once the first one is done you can move the dishes to the second one, and so on.
– You can then also buy two more sets of crockery/cutlery and use whichever comes out cleanest.
But as long as you keep renting them, washing up is dead!!
IT issues…
I've been having really odd issues with laptops lately. My work laptop running MacOS Tahoe refuses to recognise any sound-input device connected by USB, relegating me to Bluetooth-connected or built-in mic’s.
And yesterday I plugged my laptop in by USB-C/Thunderbolt cable to my monitor (which has a built in USB hub and basically works as a dock) and nothing connected. After about 40 mins of swapping things around I decided the Thunderbolt cable was dead.
ARGH!!!
Finally for today, and I've said this before: I'm REALLY curious about this technology. I WANT it to be good. I am trying it (in work context only). I see others saying it's good, but it's just not. If it was good, it might justify the many, many bad things about it. But it's just not. And I don't understand what I'm missing. I really hope it just goes away soon so I can stop thinking about it.
3) I find the process of using a coding agent so mentally taxing. Mental processes to do something are:
– Convert what I want to do into English
– Add context that the LLM will need to do it well
– Wait (and get distracted by something else)
– Get results (context-switch back)
– Figure out what the code output does
– Figure out if that was what we wanted it to do
– Iterate bugs and style
Manual coding is:
– Convert what I want to do into code
– Iterate bugs and style
2) So much of what I hear about AI is caveated.
– Yeah, sometimes it gets things wrong
– Sure, token prices are going up
– Reading code to verify it is so boring
– I had to keep telling it to do better
– It will be better soon
And yet… and yet… people default to using it. "Oh we can just chuck a load of stuff at this and it will work it out.”
But… you just told me that a) it often can’t and b) it's really expensive to do so!! It's INFURIATING!!!
It's hard not to have thoughts about LLMs every single day.
1) I tried "vibe coding" a small internal tool yesterday. It worked, but the code was terrible. It make no useful suggestions about architecture, didn't question anything, didn't ask for extra information that might have helped, use lots of short and meaningless variable names, left no comments or explanations of how anything worked.
Sure I didn't ask it for these things. But it was BAD code by default.