Xcode 26.4's Release Notes subheadings include anchors, but unfortunately the anchor IDs are the text of the subheading.

There are a lot of #Resolved-Issues anchors in that document. #AppleDevelopment

Simplifying, but Apple offered APIs using two path kinds: HFS uses colon as a path separator, POSIX uses slash. 25 years later, the HFS APIs are mostly dead. We just use POSIX.

I just wrote a file with a colon in the name and sent it to the iOS share sheet… which shows a slash.

I was going to convert both to hyphen/minus (the low ASCII one) out of paranoia anyway, plus 2026:03:25 is weird. I just wanted to see what would happen: I write POSIX, it shows HFS. Avoid both! #AppleDevelopment

AI is the Ultimate Expression of Open Source

Someone left this comment on my blog last week:

“I swear am not making this up… I tasked Claude with making a Web Cam App on OS X because I couldn’t find one that was easy to use and handled what I needed (Zoom in/out, Colour, Saturation). I noticed a lot of its web searches centred around Celluloid and JakeSpurlock.com… and wow, you’re further ahead of me, in fact I don’t need an app any more.

However, you might get a cheeky PR in the next week or two as one feature missing from Celluloid (that I can see anyway) was a Vignette slider. I like a nice subtle vignette.”

Read that again. Here’s what happened:

The loop closed. We’re both further ahead.

The Flywheel

Open source has always been about this: someone builds something, shares it, others use it, some contribute back. The commons grows. Everyone benefits.

But there was friction at every step:

• Building required deep expertise and time
• Discovery meant knowing where to look
• Contributing meant understanding someone else’s codebase

AI removes that friction.

Creation: I shipped six apps working nights and weekends. Not because I suddenly became a better programmer, but because AI let me move faster. Ideas that would have stayed in my head became real software.

Discovery: Alan didn’t find Celluloid by searching GitHub or Hacker News. He found it because Claude — in the process of trying to help him build a webcam app — kept pulling up my code and my blog posts as reference material. The AI discovered the existing solution while trying to create a new one.

Contribution: The vignette feature Alan mentioned? I built it five days later. It’s already merged. The feedback loop between “I wish this had X” and “here it is” has collapsed.

This Isn’t Hypothetical

I’m not theorizing. This is my actual workflow.

Recently I shipped:

• Today — an RSS reader for iOS and Mac
• ScriptStrip — local transcription for macOS
• Jewel Case — Spotify album art display
• Celluloid — virtual camera with filters
• Draft Night — Little League draft software (open source)
• WalkUp DJ — walk-up music for baseball and softball

All built with AI assistance. All open source or indie. All feeding back into the ecosystem that made them possible.

The models that helped me build these apps were trained on decades of open source code. Now my code becomes part of what trains the next generation. The flywheel spins.

What Open Source Promised, AI Delivers

The original dream of open source was democratization. Anyone could read the code. Anyone could learn. Anyone could contribute.

In practice, “anyone” meant “anyone with the time and expertise to understand a complex codebase.”

AI changes the denominator.

When Claude helps Alan understand my code well enough to contribute a feature, that’s not cheating — that’s the promise of open source finally being delivered. The knowledge isn’t locked behind years of experience anymore. It’s accessible.

More builders means more software. More software means more training data. Better training data means better AI. Better AI means more builders.

The commons grows faster than ever.

The Skeptic’s Objection

“But you’re just generating slop with AI!”

Look at the results. Celluloid works. People use it on Zoom calls. Alan found it useful enough that he didn’t need to build his own — and valuable enough that he’s contributing improvements.

The quality bar is the same as it ever was: does the software solve a problem? Does it work? Will you maintain it?

AI doesn’t change those questions. It just lets more people attempt answers.

Where This Goes

I don’t think we’ve fully internalized what’s happening.

Every open source project is now more discoverable — not through SEO tricks, but because AI assistants will surface them when someone tries to solve the same problem.

Every contribution barrier is lower — not because standards dropped, but because understanding code got easier.

Every builder has more leverage — not replacing expertise, but multiplying it.

Open source always had network effects. AI supercharges them.

Alan tried to build a webcam app. He found mine instead. His suggestion made it better.

That’s not a bug in the system. That’s the system working exactly as intended — just faster than we ever imagined.

The vignette slider is live in Celluloid. Thanks for the suggestion, Alan.

Related: I’m Shipping Like I Have a Team. I Don’t.

Does `@testable import` work on macOS?

I have a test target that builds fine for iOS but not on macOS. The failure is on a @testable import of the main target ("No such module"), which is a SwiftUI application that supports both.

Any other thoughts? #appleDevelopment