So I recently downloaded Apple Intelligence on my IPhone 16. I charge it only to 85% to save battery, but after I downloaded the AI, battery can’t last a day without a charge. Also, the system data file is enormous. I have downloaded it 2-3 days ago. Is this storage and battery thing going to calm down? Every answer is highly appreciated 😁
Spent the last few weeks building Pattern — Personal Pattern AI.
It reads your:
All on device.
Then it writes one paragraph a day:
The pattern you would have missed in a chart.
The one that hooked me:
My deep-tagged sessions almost all started above 55ms HRV.
The shallow ones were mostly below 42ms.
No matter how hard I pushed.
Focus tracked recovery, not willpower.
A few things I cared about from day one:
Solo build.
No funding.
Live on the App Store now:
https://apps.apple.com/us/app/pattern-personal-pattern-ai/id6766061942
Happy to answer anything about:
I need to vent because the baseline user experience on iOS has become completely unusable.
First, Apple’s voice-to-text / dictation is easily the weakest link in my entire tech stack. I use it daily, and it constantly misunderstands basic words, forces me to go back, erase, and repeat the same process over and over again.
Second, the predictive text is just as bad. Instead of helping, it actively wastes my time by throwing out completely wrong suggestions and breaking my typing flow.
To top it all off, I tried to actually use Apple’s official support channel to complain. Siri completely failed to direct me to the right place. When I finally navigated to the support page myself, the site forced me to create a username, made me check a dozen boxes, and then—in classic broken fashion—looped me right back to the login page to start over.
Is anyone else dealing with this absolute regression in basic software quality? How is a trillion-dollar company letting core features like dictation and customer support loops remain this broken?
TestFlight is now open for external testing. iPhone 15pro and iOS26+. Feel free to try it and let me know what works, what feels confusing, or what could be improved.
Thanks so much for helping test it!
Just shipped an audiobook app and ended up leaning on the on-device Foundation Models framework for four things: "previously on…" recaps for series, cleaning messy filenames into typed structs, tightening publisher back-cover blurbs into 2-sentence summaries, and one-line witty captions for a year-in-review screen. All running on the 3B model, all offline, all free.
The real killer feature is @Generable + @Guide. You get a typed Swift struct back instead of parsing model output, and the per-field descriptions feel like a more reliable prompting surface than system prompts.
Rough edges: the safety classifier is aggressive on both input and output (book blurbs about war history or politics will trip it, permissiveContentTransformations plus stripping pull-quotes/URLs fixed most of it), and the model will confidently hallucinate authors for obscure titles, so treat output as a suggestion not truth.
Overall though: structured output, free inference, zero data leaving the device is a wild deal for a 3B model.
So I recently bought a MacBook M5 air and my iPhone was a 13 Pro which do not have access to Apple Intelligence, so I did not get to see any features. now that MacBook can use Apple Intelligence like most of the features are useless. image playground is just a gimmick. If I’m being honest, I really enjoyed summary because you could just select multiple paragraphs of text and just take the summary, especially if you’re reading the news and you don’t want to just read everything. I really enjoyed the Apple Intelligence rewrite function in writing tools. I was writing an essay and I just did not like the way a sentence sounded so I just decided to try out this feature and the sentence look instantly more better when I edited the new sentence.
I needed something to help me prepare for therapy sessions. I kept showing up and blanking, then remembering everything I wanted to say on the drive home.
So I built Prelude.
It was my first time working with Swift. There were walls. Voice echo during agent playback, responses getting truncated, the AI starting to sound repetitive after a few exchanges. Figured them out one by one.
The app runs on Apple Intelligence and uses the premium on-device Apple voices for TTS. The whole thing works offline, nothing leaves your phone.
It has a voice agent that talks with you before your session, generates a structured brief from the conversation, and tracks your emotional trends week over week.
Free forever. No ads, no in-app purchases. Built it as a charity project for the mental health community.
This app helps you turn voice into searchable notes with AI generated tags and summary.
Problem: AI summary for voice notes is helpful but most of the apps are paid because they have to make API calls.
My solution: We generate summary and tags using Apple Intelligence. It is not as powerful as using OpenAI or Claude API but it's free.
We do not collect any user data.
Plus the app is really small (2.9MB) so you don't have to worry about the app using up too much space.
Apple just confirmed something I honestly didn't expect to happen this soon.
iOS 27 will let you swap out Apple's AI models for third-party ones Claude, Gemini, ChatGPT across Siri, Writing Tools, and Image Playground. The feature is called "Extensions" internally, and Mark Gurman reported it two days ago.
This is a big deal. Apple's entire brand is built around keeping things inside the walled garden. Letting users choose their own AI model is about as un-Apple as it gets.
Why Is Apple Doing This?
Because they have no choice.
Apple Intelligence launched with a lot of promise and very little delivery. The company has been promising a smarter Siri since 2024. It's now mid-2026 and people are still waiting. The Google Gemini partnership was supposed to fix this, but even that integration has been delayed and spread across multiple iOS updates.
Meanwhile, Android users have had Gemini Live for a while. Samsung Galaxy S26 can place a Dunkin' Donuts order through DoorDash via voice. Siri still struggles with basic multi-step tasks.
Opening up to third-party models is Apple admitting they need more time and they'd rather give users options than lose them entirely.
What "Extensions" Actually Means
According to the Bloomberg report, Extensions will let installed apps surface their AI capabilities directly through Apple Intelligence features. So if you have Claude installed, you could use it inside Writing Tools or Image Playground on demand.
This applies to iOS 27, iPadOS 27, and macOS 27. Google and Anthropic models are apparently already being tested in internal builds.
The base layer will still be Apple's custom Gemini-licensed model. Extensions sit on top of that, giving you an override when you want something different.
WWDC 2026 Is June 8
This is all coming into focus at WWDC, which is five weeks away. iOS 27 is shaping up to be the biggest software release Apple has put out in years Gemini-powered Siri, Extensions, new photo editing tools, and potentially hints at the rumored touchscreen MacBook Pro.
For Mac users especially, the Extensions feature in macOS 27 could be genuinely useful. Imagine triggering Claude directly from any writing surface on your Mac without switching apps.
My Take
Apple has been slow. Painfully slow. But this move is smart. Instead of shipping a half-baked internal model, they're building a platform. Let the best models compete inside iOS. Users win, and Apple keeps the hardware and ecosystem lock-in.
Whether it actually works as advertised at launch is a different question. Apple Intelligence has a history of promises not matching reality.
WWDC on June 8 will tell us everything. I'll be watching.
What do you think will you switch to Claude or Gemini inside Siri, or are you waiting to see how it actually performs?
So I made something with the shortcut app so Siri can give me the weather and everything like that (things on my calendar today) and at the end it assumed I was Christian. It gave me some random Bible emoji while saying “Enjoy your day, Tristan”. Luckily for her, I am Christian, but if it was a Muslim, it would have been disrespectful and could have made a scandal. Siri needs some updates guys.
So I made something with the shortcut app so Siri can give me the weather and everything like that (things on my calendar today) and at the end it assumed I was Christian. It gave me some random Bible emoji while saying “Enjoy your day, Tristan”. Luckily for her, I am Christian, but if it was a Muslim, it would have been disrespectful and could have made a scandal. Siri needs some updates guys.