My boy is going for World Book Day as Lindon from Cradle. Made him a little Dross to sit on his shoulder 😄 The white hand was a bit of challange to sort out with paint and corn starch.
I was meant to post this writeup on my dev blog originally, but half way through realised it really is more related to the teaching resources I'm making. Still, thought all my nerdy friends might enjoy some TIL mydragonlibrary.com/blog/playwrite-font-indesign-cursive-fix
More adobe nonsense... InDesign has two Paragraph panels that look identical - one has settings the other doesn't, and that menu is the only way to get Google's Playwrite cursive font to actually join up.
After a bit of time you get a feel for which models/tools works best in which context. Also avoid the hype and overly niche tools unless they seem to *really* fit your use case.
Personally I mainly jump around for free token usage 😂
Tried the Claude Chrome extension for grabbing context off X threads. It does that well enough. No chat history though, and switching tabs kills your session (combined with the above: ugh). There's a connector in the main app too but it chokes on multiple browser profiles. Or is it just me?
ALT: Note from AI: "Note: there's a typo in one of the PDFs — valenrines-bd-a4-single-page-answer-key.pdf (missing the t in valentines)."
This is the type of AI help I'm here for... 😆 Spotting the typo in one of 12 nearly identically named files was actually helpful.
Ooh, TIL gitbutler :) that looks interesting. Though for this specific case there wasn't anything to commit while things were broken. So it's more about documenting the process and catching early enough that I need to start gathering artefacts and notes. :)
How do you all document your debugging process? Especially when you notice things are getting interesting after you're knee deep in fix explorations. I clearly need a system.
The whole fix happened in a single Claude Code conversation though. It was mainly a CSS/print issue, so I asked it to recreate every broken variant on one page. Worked surprisingly well. Instant visual documentation without re-breaking anything.
No notes, no intermediate commits (no point committing if it's broken, right?). By the time I realised it wasn't a quick fix I was on variant 4. Turned out it was worth writing about — but good luck reconstructing that from memory a day later.
I almost lost an entire debugging story because I forgot to screenshot along the way. Again. 🧵
Codex also did a lot of thinking and then gave me a bunch of broken designs. 🤦♀️ Moving to parsing images via a script, which is still not perfect, but the results are closer to workable.
I store these as arrays with numbers mapped to colours, so thought that would have been easy to generate. But nooooo. On one attempt Claude choked after ~10 minutes of rambling about what are the best designs and what palette to use and redesigning the concept without producing any actual output.
I've been working on making a mystery image activity tool. Suffice to say, I'm not impressed with AI's attempts at pixel art. Wanted to do a few quick mockups for the valentine's day activity set. Neither good nor on topic.
The isolated context was annoying, as I don't want to continue the convo in the PR, at this point I would have preferred it in the terminal to preview and then continue.
Even worse, couldn't find the session/conversation in any of the interfaces outside of github. The job output had "Now I'll wait for the commands to be approved." near the end of the output, as if it was waiting for an interactive response.
Trying out a variety of workflows and run into a 'fun' issue today. Asked Claude Code to do some refactoring on a Codex created PR via `@claude` and it happily done the job, but never committed anything. 🤔
Yep. I wish it was a more natural integration though. Can be a bit clunky, and the terminals have hung on me a few times after not using them for a while.
Obsidian + terminal plugin + claude code or any other cli agent is the secret sauce I've been missing. Finally can stop abusing Cursor when working on my MD notes 😂
Refactor went pretty smoothly with the new Codex.app Only run into one issue - it kept insisting on adding a type incorrectly to useForm() 🤷♀️
Did some refactoring today and realised I was creeping towards reinventing form state management as some of the activity component were getting more complex... Well, happy to say I did not do that and setup Tanstack Form instead.
Made me pause while I was walking past a book store. Is the book any good? ;D Be More Bird by Candida Mayrick
Wasn't sure what to expect, as I've only listened to one of the songs before, but Hades Town was fantastic - entertaining performance, great voices, and the way they use the light to shape the scenes! 🤯
Not sure if it's just me but for some reason password protecting a pdf exported from InDesign with a password containing a - makes Acrobat claim the password is wrong, even if I just copied and pasted it 3 times for exporting, confirming and then opening. 🤔 Weird and annoying.
The form filling itself was a bit hit and miss. First convo didn't offer filling the fields. Second convo it offered filling, but then tried to paste everything into the first field 🤦♀️
So slightly helpful - it did read the things already on the page correctly - but not revolutionary.
Main thing that makes it an immediate NO GO for daily usage - no extensions! Come on. I'm not using the browser to manage my passwords... there's a reason I keep everything in Bitwarden.
Downloaded Atlas hoping it might help me fill out some forms quicker when a lot of SEO optimised, repetitive text is needed. Here's my gripes 🧵
My son's favourite books had an animatic made (sadly the Kickstarter wasn't enough to make a full feature animation). It was a lot of fun watching the characters come to life :) They also handled the shortening of teo books into one movie quite nicely! youtu.be/JJMuJCLsYeQ?...
TIL: Starting a question with 'Explain...' in Antigravity, seems to make it act like the Ask mode in Cursor.
Now I can finally get answers to my million questions of ' WHY did you do this AI?!'
