Issue 743

13^th February 2026

Written by Dave Verwer

Comment

When reading through my RSS feeds for the week, the number of articles about agent-assisted coding (even ones not supported by default) I read was overwhelming. It felt like there was no other topic for the week!

Am I going to write more about it, too? Well, yes, but not on how to set it up, use it, or whether I like the Xcode 26.3 implementation. Instead, I’m going to look a little into a possible future of working with these tools.

You’ve probably watched Apple’s code-along session introducing the feature. It’s good, and the presenters do a good job of explaining it, but you’ll notice something about the prompts they use. They keep mentioning that you should use detailed prompts, but the ones they use are short and don’t look much like the ones I have had success with.

Don’t get me wrong, I’m not criticising. It’s very easy for me to say a video should use essay-length prompts, but that’s hard to do in a 30-45 minute tutorial session and I understand why the session is how it is.

Ambiguity in a prompt is the one thing that the constant improvement in models will never fix. What happens in the edge cases? What happens when there’s an error? A short prompt can’t contain the detail needed for predictable and reliable software. Some of the agents, like Claude Code, will now ask clarifying questions if they detect ambiguity, but in my opinion it’s much better to think through your software and specify more things ahead of time before letting the agent start to code.

Learning how to use an agent effectively is definitely a skill, and the spec for whatever new app or new feature you’re building is really important. One of the most effective techniques I have used came from watching this workshop by Peter Steinberger where he lets two different LLM contexts try to find and fix holes in a spec before coding. As long as you read the feedback and make decisions where they need to be made, this is an incredibly effective way of working.

What I’d like to see is some of these agentic tools start to integrate that way of working, and none of the UIs I have seen so far encourage this. The agent is always in a sidebar or smaller window and they don’t guide you down a path of making sure you have removed ambiguity before coding. I can imagine a future version of Xcode where the agent works with you to refine a spec for whatever change you’re planning, full screen in the editor, including using multiple context windows to push back against you. I’d love to see the spec in a completely separate context window to the coding agent, too. It would need a lot of work and getting the UI right would be tricky, but I suspect it would be powerful. It would also change the focus of the UI from “the agent is only important enough to live in a sidebar” to “the agent and your instructions are as important as the code”.

All of this is possible already, of course, but only by using multiple tools and you need careful model and context management for it to be effective. I’d love to see Apple and the Xcode team really lead from the front and reduce the friction of working this way. Maybe we’ll see something at this year’s WWDC?

– Dave Verwer

Sponsored Link

Shipping white-label apps used to mean repeating the same steps and signing in and out of App Store Connect dozens of times per release. With Runway, ship everything in one place, just once.

News

It’s that time of year again! The Swift Student Challenge is open for submissions until February 28th. It’s truly a unique experience for the winners, but it’s also a great opportunity to put your mind to a focused, constrained project for a couple of weeks to see what you can create. Check your eligibility (it’s broader than you might think) and get your application in!

Tools

Aryaman Sharda:

SimTag adds a small, unobtrusive overlay to each iOS Simulator window showing the branch that build came from.

That’s it.

What a great idea. 👍

Code

What’s this? An article that doesn’t even mention AI? 😂 Gabriel Theodoropoulos writes about how little tweaks to the built-in SwiftUI morphing animations can make all the difference to how polished your app feels. I love that there are videos for each step of the process, too. Great article.

Alex Ozun:

It goes without saying that just reading alone won’t make you a Swift concurrency expert, you’ll need to continuously put everything you learn to practice.

And yet, also:

But those who do are guaranteed to become top 1% subject matter experts.

I love the format of this article, it’s great.

Videos

Yes, there were Swift talks at this year’s FOSDEM, but this one by Dylan Ayrey and Mike Nolan was the one that caught my eye. They ask the question of whether open source is doomed in the age of LLMs, and it’s really a fantastic talk.

Talking of FOSDEM, there was also a pre-event focused entirely on Swift with nine short talks. Perfect for a few bite-sized watches this weekend.

And finally...

First it became a typeset PDF again, and now a physical object!