Subscribe for weekly commentary and coverage of Swift and Apple platform development. Written by Dave Verwer and published every Friday. Free.

Picture of Dave Verwer

Issue 716

27th June 2025

Written by Dave Verwer

Comment

Several people have asked for my opinion on Apple’s coding assistant feature in the Xcode 26 beta. So far I’ve been saying that it’s ahead of my expectations, and yet still behind the leading (or bleeding) edge of AI-assisted development.

I didn’t expect a chat interface in Xcode this year, multi-file context with automatic context selection, or third-party model support. We got all that, and Apple is ahead of my (admittedly conservative) expectations.

Unfortunately, the most recent developments in AI-assisted development, agents, are a step change over a “simple” context aware chat interface like Xcode 26 has. After using Claude Code, I can say that agents are a huge step forward over the previous generation of tools.

The key points of why agents are transformative are well made by Christopher Trott in his recent post, “Rewriting a 12 Year Old Objective-C iOS App with Claude Code“:

Claude Code unlocked a new level of usefulness for me due to:

  • Its ability to view the xcodebuild output and fix its own syntax errors.
  • Its ability to plan, create its own TODO lists and methodically execute on its plan.
  • Its ability to quickly navigate around my codebase with various amounts of guidance by me.
  • Its ability to incorporate documentation and other context.

The pain point with the chat interface that Xcode 26 has is that your workflow is so much slower without being able to run the build or tests, as you need to explain what went wrong when that happens. Having agents able to run regular builds and look at the output, run tests and look at the failures, and access other command line tools is a huge step forward.

Agents for coding are brand new¹, and it would be hard to fault Apple for being just a “few months” behind under normal circumstances. However, things are changing so fast that it works against their normal approach of introducing major new features annually. Will they choose to iterate on the “coding assistant” during the lifetime of Xcode 26, or will we need to wait a year to see what Xcode 27 brings? If they can iterate, they’re in the race; if we need to wait a year, they’re not.

I’ll finish by saying that it’s possible Apple might not want to enter this race in the way I’m suggesting here, and if that’s the case, it’s not the end of the world. All the other agents work well with Swift code and iOS/macOS apps. While the models don’t know Swift as well as they know other languages, especially when it comes to things like Swift 6 language mode and concurrency, they can produce well-written and reliable code that you can use today. No matter what Apple’s plans are, I’d recommend trying out a hybrid approach to see if it works for you. As a command line tool, Claude Code fits this approach especially well as you don’t even need to change your editor or IDE.

– Dave Verwer


¹ Claude Code entered “research preview” in late February and GitHub talked about the Copilot agent just a month ago.

AI Coding Assistant Benchmarks: Who Fixes iOS Crashes Best?

Discover how the leading AI coding assistants perform at identifying the root cause of mobile app crashes and generating code fixes across iOS and Android. Instabug benchmarks GitHub Copilot, Cursor, Claude Code, and SmartResolve to reveal surprising platform-specific strengths. Read the results.

News

AI sceptic in LLM adventure land

If you’re sceptical of tools like Claude Code, read Aleksandar Vacić‘s latest post where he goes from:

Ever since ChatGPT and its gen-AI ilk arrived, I have been very vocal and adamant that these are bullshit generators

To:

I have never been happier to pay $200/month subscription for anything.

In just a few paragraphs.

Tools

Adding Icon Composer icons to Xcode

Keith Harrison talks about the new Icon Composer app. It’s unfortunately not all good news, as you’ll read, but it’s early days and I’m glad we’re at least on the road to a place where we never have to prepare icons in the old way. 🤞

Code

Apple’s On-Device Foundation Model Is Here.. But Is It Any Good?

Ronnie Rocha on using Apple’s new Foundation LLM:

Don’t think of this as “ChatGPT in your pocket.” Instead, think of it as a highly-efficient formatter and extractor; a lightweight engine capable of translating simple natural language requests into the precise data structures and function calls your app already knows how to handle.

I want to write more about what a powerful concept this is at some point, but please take the time to read Ronnie’s article, it’s great.


Automatic Observation Tracking in UIKit and AppKit

The UIObservationTrackingEnabled Info.plist key, which shipped with last year’s releases, brings state-based UI updates from SwiftUI to UIKit and AppKit. Peter Steinberger digs into the feature showing what it can do, but also covers some reasons why Apple may have made the decision to leave it disabled by default.


Embedding Godot games in iOS apps is easy now

You might have missed Christian Selig’s latest article in the run up to WWDC, but it’s worth reading as it’ll quickly get you up and running with a Godot scene inside your iOS or Mac. There has been plenty of work going on with Godot on Apple platforms recently, including the release of Xogot and plenty of Godot-related libraries, mainly from Miguel de Icaza. It’s great to see an open-source game engine continuing to rise.

And finally...

What does the Xcode “coding assistant” prompt look like? 🤖