ai-yi-yi

So, I think "intense ambivalence" - in the sense of being pulled in two directions - is the most reasonable place to be about this whole AI thing.

That's kind of a disclaimer - I don't think AI is evil enough to be worth not messing around with, and depending on how things shift, adapting to an LLM doing gruntwork may be a survival skill.

My friend JZ (no not Jay-Z) has been messing around with the AI music maker Suno and I decided to give it a shot... I put in one line of prompt, something about New Orleans Street Band / Second Line style, funky, and got Step On Over. The horns are kinda synth-y, but it's a rollicking bassline, and some of the lyrics are sorta brilliant... I especially like the last line of the chorus "We got room for one more in this song" (although the first line of the chorus "let the good feet roll" is a bit iffy.) Also the final verse during the background is a banger: If your week ran long, come walk it off / If your shoes feel tight, loosen up / Take my arm, we'll find the beat / There's a place for every pair of feet.

Also, a personal friend (and fellow bloger) Shaun McGill set Gemini to make a video summary of his e-book "How did we get to the iPhone" and the result was also pretty impressive.

It's (another, unfortunately) interesting moment in AI. With "Character.ai" filing for bankruptcy, it's easy to wonder if this is the start of a bubble pop - these remix machines have been heavily subsidized for years now, and what happens to the supply/demand curve when and if that becomes untenable?

use the platform, luke

 Why Don't More Developer Use the Platform? 

Sometimes I think not having a build system for 90% of my side projects has been great at keeping me grounded. Browsers do so much, they're such a stable platform year after year.

As a UI/UX engineer, I like to think about how people arrange their virtual workspaces.

15 years ago I wrote about how I liked to move the Windows taskbar to the side, vertically. It made better use of landscape mode monitors, and the width could be adjusted to show full-on title bars when needed.

Since then, Windows has followed OSX's lead in having the default be "one taskbar item per app", not "per open window". Also, on every platform, most apps have switched to preferring multiple tabs over multiple windows. 

So I was interested when Chrome started telling me about its new vertical task bars; the visual parallel to how I would set up Windows back in the day was striking.

It's also interesting whether people tend to have fewer browser windows with lots of tabs, or many browser windows with a few tabs each. I fall into the latter camp; I guess I see each window as a fresh context.

(Of course people who depend on keep specific windows available in perpetuity kind of scare me. They're meant to be transient!)

Destroy Any Website - desktop only, sound on. Super clever; looks like they analyze the page into DOM-ish bits then put it all into a big canvas, and have a lot of fun doing it. 

web app to desktop

Webapps are still an enormously powerful and egalitarian paradigm. 

I've been enjoying using my own webapp "Bubbler" for all my Todo needs. I "add to home page"ed it on my iPhone a while back and it feels and looks great. (The one challenge has been lack of a good offline mode, more on that later.) 

Anyway, it's asburdly easy to use Safari on Mac to make an appified-version of this webpage; I just went "File -> Add to Dock":

Claude helped with a few extremely minor details, but basically everything had been set when I had it help me with the mobile friendly version earlier. 

That's one of the interesting thing about using LLM to help code; stumbling over what's actually quite easy even though it looks a little hard. (The experienced coder has some advantages here, but sometimes the lack of assumptions a vibe coder brings to the table wins out. Beginner's Mind all around!)

ya bunch of tools

 Interesting seeing what tools AI Agents recommend.

(For my professional life, tool choice has often been from one particular architect champion on the team, combined with "what every one else was doing". For side projects, for reasons of sustainability and reducing the footprint of what I had to learn, I've tried to minimize the sheer number of things I use, a bit of "NIH syndrome".)

on doing the math

I was thinking about Simon Newcomb, the eminent astronomer/mathematician who mathematically "proved" that human mechanical flight was forever impractical a few months before the Wright brothers went ahead and did it. 

(Look into what he might have missed, I think he was leaning on the square-cube law: as a machine grows, its volume and weight increase cubically, while its wing area increases only quadratically, by the square.)

I had a similar incident in real life, where my friend EB measured an overstuffed chair and pronounced it impossible to get through a small door and into a back room, and then a few days later my friend JZ and I just did it. (Here, EB really didn't account for the "pivot" of it all - must have missed that episode of Friends.)

That's one of the dangers of reductionism; a model captures what it can, and ignores the rest.

I see a parallel in software testing... unit tests are fast, focused, cheap to run and now incredibly tempting to just throw an AI at to produce. But the success or failure of a working system is an emergent property of all those units working together in a context provided by all the other units... and browsers, networks, external systems, and of course users. Integration and end to end testing are expensive but more important than ever.

So test the units, but don't forget to test the pivot.