
💻 Lead iOS @37signals iOS dev for 18 years. Founder @UnderdogDevs. 🇲🇽 Mexican not Mexican't. Professional SwiftUI hater.
Joined March 2008
- Tweets8.9K
- Following233
- Followers2.4K
- Likes11.3K
Pinned Tweet
🏁 SwiftUI vs. UIKit 🏁
Build a native version of the chat screen in Campfire. One prompt, but this time with skills. SwiftUI got Apple's Xcode 27 skills, UIKit got ones I wrote.
Winner is UIKit 🥳 (2-0) vs. SwiftUI.
UIKit/Swift passed every test.
UIKit/Objective-C was the fastest and used half the memory 🏎️
SwiftUI is still slow (5s of stutters!), even with Xcode's swift-ui-specialist skill.
What would you like me to test for round 3? Code is in the repo (next comment).
Both at work and at my hobbies, I've found that agents outwork me and outcommunicate me. From a recent experiment that failed:
"About 2,100 messages. You wrote 387; machines wrote about 4.5 for each of yours. At the peak you were writing one a minute and getting four back."
It's clear that I have more manpower (agentpower?) than I can handle. During conversations at the 37signals meetup, it came up that we've become CEOs of our own domain.
How do you handle so much information?
4
350
I spent a week in Chicago with @37signals and crew. Lots of interesting and amazing conversations about AI, where we're going, and how we're achieving it.
Time to get back to my experiments! I just published github.com/notolivaresf/uiki… which I'm using for my projects!
I'm super interested in benchmarking skills, so I'll focus on that this week!
Fernando retweeted
We're laser-focused on the Apple platforms development experience for really serious builders. Some examples:
- Providing world class simulator use tools for your agent
- A system that allows agents to work with Xcodebuild in a way that gets out of your way (priority queueing builds based on whether you're looking at them / the agent is doing them in the background, etc. More to come here soon as well, I have something really special cooking 🧑🍳)
- Support for sims even apple doesn't provide (CarPlay sim!)
- A great out-of-the-box suite of skills + tools that help with Apple platforms development
- Best-in-class support for letting your agent use App Store Connect so you never need to worry about things like taking screenshots, uploading builds, etc.
There are many great orchestrators out there, but there are a lot of unique things about development for Apple's platforms that a standard Agent + Orchestrator setup doesn't equip your AI well to handle. That's what Bitrig is here for!
My Opus 5.5 agent has been so dumb.
I asked for a UIKit implementation, and it wrapped SwiftUI in UIKit and then proceeded to tell me both implementations had the same laggy behavior.
What?
BUT then I learned some of my rules were preventing it from doing what I asked.
SO I have revised my agents .md file based on Opus 5.5 and GPT-6 Prompting guides.
Make sure your rules are easy to follow.
Honestly, this wasn't a rule I asked for. I tried to slim down my Agents file, and this weird rule was written.
It’s not even close. Why are we still using SwiftUI when we have agents?
UIKit was boring to write. We’re not writing anymore and you get a performance boost.
I compared UIKit and SwiftUI performance with the most ridiculous scroll view possible.
Memory usage wasn’t dramatic, but was interestingly spiky. Perhaps due to the different numbers of gifs and different resolution of the images rendered on each cell.
When it came to profiling peak memory and energy usage, I expected UIKit to yield the same kind of results I saw in SwiftUI: We’re still displaying an enormous number of elements, animating them the same, and these images are not any bigger or smaller, right?
Now don’t I look silly.
Memory usage hovered around 92MB for UIKit, compared to 248MB in SwiftUI.
blog.jacobstechtavern.com/p/…
Replying to @EquityDiamonds @fromjrtosr
SwiftUI has too many random behaviors to make it sensible for AI to even get right
While the code might look elegant, there’s so many little gotchas, there’s so many little paper cuts, and it’s so easy to use incorrectly
You can have way too much state driving the animations and creating unwanted animations
You can have way too many things causing layout glitches
And for whatever reason, agents can’t see most issues until it’s at runtime because of however they reason about it
Fernando retweeted
Replying to @PaulSolt @fromjrtosr
Basically zero reason to use SwiftUI if you’re vibecoding
Fernando retweeted
^^^^ exactly
related: I originally used PHP for my app's backend years ago because it's all I knew, but now that AI writes all my code I decided to have it rewrite my backend in Go which I've never used
rewrite is flawless, 1/5th the LOC and literally blistering speed
Replying to @fromjrtosr @paulostergaard
Verification and seeing the issue is the biggest challenge for agents
Otherwise, they’re just coding blind, especially around SwiftUI, which is so complicated
Fernando retweeted
I think it will be important to look at this from a completely different angle.
What if the tokens usage is not an issue. And what if all the tests are passing and the expected behavior is being satisfied and humans never have to look at the code.
Does the underlying code quality still matter?
As long as the code is working as expected... who cares what is going on underneath.
And if we go with this route does it mean that the understanding of our code is now owned by AI companies.
This is an interesting question.
As I manually type more code (Yes I am a dinosaur) and review my code using AI tools, I am finding how bad AI suggestions are from an architectural perspective. It behaves like a teenager, who when asked to clean the room simply pushes all the garbage under the bed.
When I confront AI with those bad decisions it says "Oh yeah you are right....".
I really don't understand how people are 100% trusting AI and committing all the code without reviewing.
Fernando retweeted
Peak memory is 300% (!) between UIKit/Objective-C and SwiftUI.
103MB vs. 35MB.
UIKit/Objective-C is also twice as responsive. I don't have to tell you that for drawing, every millisecond counts.
34ms vs. 86ms
Are the agents finally bringing back Objective-C? 😆
Fernando retweeted
SwiftUI vs. UIKit 🏆
ROUND 3
Build a native version of the chat screen in Campfire. One prompt: SwiftUI got Apple's Xcode 27 skills, UIKit got ones I wrote.
This time, agents have an additional +3 hours to optimize their build.
Winner is UIKit again!
SwiftUI did win on jumping to message 2,500 by 13ms. It is still slower at startup by 39% and peak memory is still almost double (44.4MB vs. 77.2MB) which is mind blowing... consider this is only _one_ screen!
What would you like to see next? Some suggestions were @twannl skills, ObjC++, and AppKit. I'm personally super interested in AppKit, but we'll see!
Excellent advice. Incredible advice, really.
I've gotten complaints about the experiment I'm running. It's not perfect. It's not scientific. It's unfair to SwiftUI.
Maybe? But I'm trying to figure out the future with agents. It was unreasonable to build the same app twice with SwiftUI and UIKit myself. It's now completely within reason and I'll do it again and again.
If that ends up exposing that SwiftUI sucks, so be it. If it ends up exposing that SwiftUI with skills is way better than UIKit, so be it.
Be unreasonable.
Fernando retweeted
"An early answer to your question: yes. The state machines are finding bugs before a line of the build is written. Just laying out each state and its data against today's code turned up these."
Sold! Will be using them everywhere from now on!
























