Not only Swift

On-device AI with Apple’s Foundation Models framework gives us free, private, and low-latency inference in Swift. But mobile device constraints and context window limits (8k tokens as of iOS 27 beta 3) mean that deep reasoning or large documents still require cloud models.

In my latest blog post, I explore how to build a hybrid AI router in Swift using Dynamic Profiles. By measuring token counts up front with tokenCount(for:) and contextSize, you can keep lightweight workloads on-device and seamlessly route demanding prompts to Gemini (via Firebase AI Logic) - keeping your UI and service layers completely decoupled.

When Apple introduced ContentBuilder at WWDC26, it sounded a bit like “nothing to see here, move along, folks”.

Ok, it does reduce the number of "The compiler is unable to type-check this expression in reasonable time" type checking error messages - and who would complain, given you don’t even have to change your code.

But there is a lot more to it than that.

Fatbobman explains how ContentBuilder works, shows some benchmarks, and digs into the question “why didn’t we get this earlier?”.

As mentioned by Fatbobman in his article, you can use ContentBuilder for non-UI DSLs, too.

Artem Novichkov explains how this works by way of implementing a deep link router, and even shows how to backdeploy this to iOS 17.

SwiftUI isn’t open source, and that’s a huge problem: You never know if what seems to be a bug actually is a bug or if you’re holding it wrong.

You also can’t contribute fixes, but that’s a whole other issue - managing the amount of feature requests and PRs you get when running popular open source projects is a tough challenge.

OpenSwiftUI has been around for quite a while, and despite it not being the actual implementation, it’s a great learning resource.

They now have a proper homepage, and hopefully that will help them gain more popularity.

I’ve got the impression that building proper Liquid Glass UIs is pretty difficult for coding agents. Which isn’t a big surprise, as the training data for most frontier models doesn’t include a lot of code that properly uses Liquid Glass.

Florian built a skill to fix this. I can totally relate to the very first core rule:

Don’t build custom Liquid Glass through adding backgrounds, outlines, blurs and shadows.

Heck yeah. That’s one of the biggest issues, right there. Probably solves 80% of the problem - but the other rules are pretty solid, too.

As app developers and builders, we want to focus on shipping great experiences - not on managing (or even building) the infrastructure behind them.

Focus on what’s important for your app: building amazing features, not the infrastructure that powers it.

That’s how I’d describe Firebase in one sentence.

However, many people are worried that using a backend-as-a-service might cost them an arm and a leg.

Unlike simple billing alerts that only send notifications, spend caps actually pause supported services once a cap is reached to prevent surprise runaway bills.

My colleague Jeff explains how they work and how you can set them up for your projects.

And after you’ve read his article, please also turn on App Check for your app.

Agent Plugins allow you to package your skills, MCP servers, tools, and more in a portable format that works with conforming clients.

If your .agents folder is riddled with dozens of subfolders containing the same set of skills (or symlinks) for each of the coding agents you’re using, you will appreciate what this means.

I can’t wait to clean up my project folders.

Joseph Heck makes a very good point - rather than claiming “it’s so over”, we need solid software engineering practices now more than ever.

Even if it might appear that models and agents can reason, they still really just predict - even if you teach them codify best practices in the form of skills.

When your Wi-Fi dies, you’re essentially thrown back to the age of dinosaurs.

Fun fact: my desk at Google’s Belgrave House office was next to Edward Jung’s in 2014, so I could see the Chrome Dino come to life as he was working on the designs.

Here is your updated list of conferences for iOS, Swift, and AI.