Mobile developer. iOS, Android, KMM. Open-sourcerer. AI whisperer.
World
Joined July 2012
- Tweets882
- Following331
- Followers78
- Likes1.6K
Why my Swift debouncer got 50× faster and what the Swift Runtime has to do with it: A measured investigation of task reuse, Swift runtime metadata, generic specialization, lock placement, and callback latency
ivanmagda.dev/posts/swift-de…
I use both Claude Code and Codex, and I got tired of project knowledge staying with whichever agent I had been using. I moved the reviewed parts into the repo so both can work from the same context.
ivanmagda.dev/posts/project-…
Pretty cool, it turns out that Telegram has native support for displaying AI agent progress.
I added it to swift-claw, and now the progress preview looks almost like ChatGPT/Claude.
Wrote a short post about how it works:
ivanmagda.dev/posts/telegram…
Sad to see Swift Talk come to an end. Thank you, @chriseidhof and @floriankugler, for everything you’ve shared over the years. I learned so much from you both. Wishing you all the best!
talk.objc.io/episodes/S01E50…
I wrote a personal AI assistant from scratch in pure Swift: one always-on binary on my own Mac, controlled from a private Telegram chat. No Node, no Python, no Docker.
Open source: github.com/ivan-magda/swift-…
That is what the clip shows: the card prints the fully resolved target path, I tap Deny, and clawd writes nothing. Typing "yes" in the chat does nothing at all. Any OpenAI-compatible endpoint plugs in as the model, cloud or local.
Anyone using your Swift app's AI feature probably already pays for ChatGPT. Asking them to buy API tokens on top is a second bill for the same models. I open-sourced ChatGPTAuth: sign in the way Codex does and call the API on the plan they're paying for.
Device-flow login plus a credential that refreshes itself, built as a Swift actor. No third-party runtime deps. iOS, macOS, tvOS, watchOS, visionOS, Linux. Swift 6, MIT. My own AI agent runs on it daily.
github.com/ivan-magda/swift-…
SpeechLanes: on-device speech-to-text for Apple platforms that handles more than one language. Configure your locales, hand it an audio file, get the best transcript back. Voice messages that mix Russian and English come through too. iOS 26+, MIT. #iOSDev
github.com/ivan-magda/swift-…
Ivan Magda retweeted
Book: About Accessibility in iOS!
I’m writing the most important book about accessibility for mobile developers and designers.
8 years of experience in one place.
Last June, Foundation Models was a Swift API to one small on-device model. After WWDC26 the same session API runs Apple's server model, Claude, Gemini, or a local Qwen. I spent the week in the betas. Three posts on what changed:
My favorite part: Dynamic Profiles. One session swaps instructions, tools, and the model mid-conversation via a SwiftUI-style body. I built these mechanisms by hand for my coding-agent series, so I mapped Apple's versions onto ours, footguns included: ivanmagda.dev/posts/foundati…
The overview maps everything else, including the bit nobody mentions: macOS 27 ships an fm CLI, and fm serve gives you a local OpenAI-compatible endpoint backed by Apple's models. I'll update the post as I publish the rest: ivanmagda.dev/posts/wwdc26-f…
Most developers blame the model when their AI coding agent starts producing bad output.
5 things I learned about making AI coding agents actually work: compact instruction files, project structure, fresh sessions, self-verification, and systematic debugging.
ivanmagda.dev/posts/five-thi…
#AI #CodingAgents #SoftwareEngineering
Anthropic: "Our Max plan gives you 5x the usage of Pro, surely that's enough!"
Me: hits 100% usage and burns through $127 of extra compute in a few days 😭
Listen @AnthropicAI, I think it's time for a Max 20x plan. My code isn't going to write itself 💀💸
