Bambang Tri Rahmat Doni
591 posts
Bambang Tri Rahmat Doni
@bengi_dev
iOS App Developer
Indonesia Katılım Ağustos 2023
128 Takip Edilen16 Takipçiler
With LLMs we’ve basically invented digital candy.
LLM written content is designed to maximize engagement in a way human text will never be able to match. It’s why AI-written posts constantly go viral on here.
AI will win not because it’s better but because it’s more delicious.
English
I can show you a good vibe coded app and a bad human coded app an you'll tell me that the human coded app is vibe coded.
English
I put SwiftUI vs UIKit scroll performance to the ultimate test: an infinite scrolling feed full of interactive 90s GIFs.
The results? Well, see for yourself...
English
@jacobtechtavern One more addition, mate. Don’t forget to give your explanation of which the good and bad things from your perspective, because i need that to learn from someone like you. Cheers mate 🍻
English
@bengi_dev I don't, but that gives me a really good idea. I should absolutely start recording that shit, like when I do the instruments trace, the making of kinda thing
English
@DonnyWals Lol. But seriously, why AI need to access Music and iCloud too, are they just blindly request the access without the need to using it?
English
Agents need music to get into the flow too!
Julie@syswarren
Apparently, it needs more vibe for the coding
English
Google Stitch vs Claude vs Human
Pick your winner
Stitch by Google@stitchbygoogle
Meet the new Stitch, your vibe design partner. Here are 5 major upgrades to help you create, iterate and collaborate: 🎨 AI-Native Canvas 🧠 Smarter Design Agent 🎙️ Voice ⚡️ Instant Prototypes 📐 Design Systems and DESIGN.md Rolling out now. Details and product walkthrough video in 🧵
English
My iOS Agent Skills are packed with knowledge. I even include things you'd think an LLM already knows, like:
- "Use Observable instead of ObservableObject"
- "Use async/await instead of DispatchQueue"
You might think Agents know these things, but in my experience:
- They need strict guidance
- They still make rookie mistakes
- They get influenced by your code patterns, which often still use older APIs
Even though training data is there, my Agent Skills guide your agents toward better iOS development.
The risk of this approach is token usage. An inefficient skill will eat your context, leaving less room for the actual task.
That's why my skills are:
- Optimized for token usage
- Designed for Agents navigating through reference files
Code examples are intentionally minimal: just enough for an Agent to know what to do, nothing more.
The latest release of my Concurrency Agent Skill saves ~4,000 tokens per query through smart routing, with zero loss in knowledge depth.
Quick install instructions below 🚀
What Agent Skills are you using for iOS development?
English
When someone says “I shipped an app in 24 hours. No debugging. No reading the code."
It actually means nothing went wrong because the app was simple enough that it could not.
Don’t be fooled by that bullshit. Put your head down, get to work. Ship something that that truly adds value.
English
All my Agent Skills for iOS development has taught me a lot, but this new skill will be one much closer to what I truly love building.
Stay tuned ✌️
English
Use less context when you have the proper skills.
You need skills for:
• SwiftUI
• Modern Swift Concurrency
• Liquid Glass
• Apple HIG (Human Interface Guidelines)
• Accessibility (So more people can use your app)
• Figma or design tools
The saved context will allow your agents to work faster and have less rework.
indy@indyfromoz
@PaulSolt SwiftUI skills for modern Swift concurrency best practices, Liquid Glass design, Apple HIG and accessibility. I have a bunch of skills for Figma and a few for Cloudflare. Helps a lot to keep the context small and achieves the rapid iterations I do in my dev workflow.
English
Today I'm pleased to announce three new agent skills that will power up agentic coding for app developers, all available now.
First up is Swift Concurrency Pro, designed to help you get maximum performance from your code *without* fighting the compiler. github.com/twostraws/swif…
English
Last week I released SwiftUI Pro, a free and open-source agent skill to help everyone write better SwiftUI code using agents such as Codex and Claude. It's already at 1800 stars on GitHub and rising, but it was just the beginning. github.com/twostraws/Swif…
English
@JorgeCastilloPr If you didn't use a Mac, where did you run the simulator?
English
I can now confirm that you don’t need a MacBook to code iOS apps anymore.
I’ve been coding an iOS app lately without Opening XCode at all. Just Claude code and simulator.
This is gonna make iOS development blow up 🔥
English
@theo Did you vibecoded it, or you build it without the help of ai?
English
Wild day. 5k users already.
Cutting a new release before bed.
- Fix bullet points in markdown outputs not rendering
- Fix crash on unsupported languages in code blocks
- Fix shell path detection for non-zsh/bash shells
- Fix errors on non-git projects
- Fix project paths with "~"
- Prevent bad paths from being added as projects
Lots more to do. Thank you everyone for reporting issues, will continue grinding 🫡
Theo - t3.gg@theo
T3 Code is now available for everyone to use. Fully open source. Built on top of the Codex CLI, so you can bring your existing Codex subscription.
English
@atmoio Help me to anser the question of “who we are?” nowadays
English
More agent harnesses need a "profile" concept.
I set up a few in mine for common tasks:
- "Default" is GPT 5.4 medium
- "Writing" is Sonnet 4.6
- "Deepthink" is GPT 5.4 high
I set a keyboard shortcut to quickly switch as well
English
The next version of our SwiftUI Agent Skill just released, with more updates based on WWDC & Apple SwiftUI videos from the past 12 months.
Enjoy!
English
The mission is 8% done, and it's already been 5 hours. I think this is the coolest one shot experience I've seen.
Okay, this is wildly better than anything else I've tried. the tests are so logical, it's using the browser to validate the UX, it's doing real quantizations.
It's been at it for 5 hours, and there's nothing in there that doesn't make sense from a tool use, and trajectory perspective.
I built Orchestra 3 months ago, it's similar to oh-my-opencode and gas-town. It worked but it sucked, I couldn't get it to make sense so I dropped it and haven't touched any orchestration tools since.
----------------------------------------
why is it so good? because they're not leaving it up to the model to decide, I see the way they frame prompts and separate sessions, it's all running like a factory lol. Very impressed.
0xSero@0xSero
Okay, Droid Missions are tied for #1 Orchestrator with Roocode's boomerang mode. This has been incredibly stable. 1. The "orchestrator" model simply delegates tasks and asks the users questions. 2. A plan is created that is much deeper than what I would expect with spec mode. 3. The mission is broken down into tasks, the orchestrator delegates a task to the worker 4. Once a task is done, it is passed to the verifier to review the output This loop I've seen with Roocode, the main difference here being the UX it feels really good. Although in Zed there's an annoying re-rendering bug in terminals that causes tons of flickering. I am using this with my custom models (BYOK) it works good in the desktop app too. (: Now look, I don't know if these systems will be productive, in that the longer models run without human feedback the more likely it is to make a huge mess. Time will tell.
English