Not only Swift

If you’re a Swift developer, you’ve used optionals before - they’re a language construct that formalises the state of either having a value or not. I thought I knew how to use optionals, but I learned a couple of new tricks in John’s article.

Great overview of asynchronous patterns in Swift, going from delegates and closures over NotificationCenter, Combine, and @Observable to AsyncStreams.

This resonated with me because I gave a talk about this at DeepDishSwift and SwiftConf in Berlin. So if - after reading Khoa’s excellent post - you find yourself craving more, find the slides on my SpeakerDeck and the recordings on YouTube.

The way SwiftUI lays out views and their subviews is often surprising, which is due to the way container views determine the size and position of their subviews.

Subviews can specify their own layout priority, which can lead to surprising layouts.

Natalia shows how this works for some common views like HStack and VStack, and then goes on to demonstrate how to handle layout priority in your own container views.

Oskar shares how he set up a software factory which he is able to operate as a single person. It’s fascinating to read write-ups like his and those of others and see how we’re all in the process of figuring out how agentic software engineering works, and how we can manage it without losing our minds.

Oskar’s conclusion “We need a better interface” resonated with me. Agent UIs are better for orchestrating complex workflows than IDEs with a chat sidebar, but they still don’t provide the level of overview and control needed to manage the software delivery process effectively.

Over the past couple of years months, we all have learned that using LLMs just by themselves is not going to give you great code.

First of all, you need an agent harness to give your LLM of choice hands and eyes, so to speak, so it can actually inspect your code and manipulate it.

And then, you need to teach your preferences (using AGENTS.md), and fill in the knowledge gaps (using agent skills).

But - as I’ve covered before, even with all those tools in place, working with LLMs and agents is not deterministic (and that’s probably a good thing - otherwise, all to-do list apps would look the same ;-P ).

But what if you want to make sure your agent is more deterministic - say, running a script every time it finishes a turn (for example, to format your code)? Antoine is here to explain how it all works and how to set it up in your favourite coding agent.

Letting an AI review the code it wrote is a bit like having a student grade their own exam.

Alexander Grebenyuk suggests we should take a different approach: have agents provide proof of correctness with tests and screenshots before opening PRs.

This way, humans can assume their responsibility: bringing taste, judgement, and accountability into the review process.

A team of researchers published a paper that demonstrates models can edit the weights of (other) models. This means models can effectively modify their own behaviour.

I assume that this also means that agent skills will eventually become superfluous, as you could have models “backfill” their knowledge gap.

But that’s not the big story here. John Resig (yes - the creator of jQuery) reflects on what the findings of the paper mean for the future of coding:

Code that can’t be read by humans, run on machines that humans can’t access, via inaccessible APIs, trained on data that humans didn’t create […] We’re just a couple steps away from code being completely irrelevant.

Depending on your perspective, this could either be terrifying or exciting. I’m personally excited about the possibilities - because, after all, when was the last time you wrote Assembly code by hand?

How do you know the skills you’ve installed actually help, or just waste tokens?

Well, you need to run evals.

Setting up evals isn’t necessarily rocket science, but it’s also not super trivial.

Caliper makes this easier. It’s actually so easy that Claude Code now includes a feature that looks suspiciously similar to Caliper.

I was late to the pre-order, so my iPhone Duo won’t arrive until January. But there are some early adopters who already returned theirs?!

When I first tried the iPhone Duo out in the Apple Store, I felt that handling it would take some time.

Shaun Donaghy actually thought that the form factor was lovely, but usability features like Face ID (or rather the lack thereof) are what made him return the device.

If you’re one of the lucky people who managed to get an iPhone Duo already, how do you like it? And do you plan to keep it?

The idea of a universal assistant is not new. In this video from 1987, we can see the vision Apple had more than 39 years ago.

What’s striking is that Apple assumed that you would get just a handful of messages per day, and that you would be able to deal with them in the morning.

This goes to show how hard it is to predict the future.