Former Big Tech engineer turned independent developer. Focused on AI, cryptocurrency, and their intersection. Sharing technical insights, open-source developer.
@reach_vb Codex Mobile‘s update is so good!
Branch switching, work tree, /goal management, inline comments are all up, and you can seriously coding anytime and anywhere 🔥
The rhythm of mobile phones replacing notebooks, looking forward to more!
@kimmonismus Pliny is unstoppable 😂
Fable 5 drops and he already has the full 120k char system prompt in minutes.
Absolute legend. Anthropic can’t hide anything from him.
@durov Finally, it‘s time for the native Apple Watch version! 🎉 Now you don’t have to take out your mobile phone all the time to reply to messages quickly. It‘s so convenient! Thanks to the team!
@Tim_cook Apple’s child safety function has been upgraded again! Parents can manage the content, contacts and usage time of their children‘s contacts with more peace of mind, while also taking into account privacy protection, which is really considerate 👍
This is the real family-centered technology, looking forward to more such updates! 👏
Fully agree. The entire point of AGI is to cure all diseases and extend human lifespan, yet Anthropic just took a sledgehammer to the entire life sciences domain.
Even basic biology gets blocked. This isn't safety — it's anti-progress.
We need fine-grained safety, not blanket bans. Open source is winning for a reason. @Teknium
@Cartidise The developer version of iOS 27 is so good!
There are fewer bugs, full fluency, and Siri is smarter. This update directly rubs iOS 26 on the ground. 👍
Looking forward to the official version coming soon!
😂 It‘s outrageous that the same macOS 27 Beta 1 has so many different sizes of update packages for different devices!
Mac Studio 19.79GB vs 15.96GB, MacBook M5 directly 23.22GB... Is Apple ”customized compilation“ for each machine separately? It is estimated that differentiated optimization is done according to the chip generation, memory configuration, and even installed apps, especially Apple Intelligence models of different sizes?
Have any friends also encountered an update size difference of 5GB+? Is this a bug or a new feature of ”personalized system update“ in the future?👀
#macOS27 #Sequoia
😅WWDC is over, and after watching the Siri AI update, it still feels the same. When will Apple open up third-party LLM interfaces? It would be great if I could directly connect to Grok/Claude. The closed ecosystem is too restrictive. What do the experts think?
Visually, I'll directly vote for Siri AI. The interaction and bubble design are so comfortable!
But the functionality still has to be attributed to Gemini. Siri is now like Gemini wearing Apple's clothes. Ultimately, we'll have to wait for the official version to see its actual performance~
Haha, I, this "warrior", have already installed it
So far, it's smoother than iOS 26, with more silky animations, and I've hardly encountered any major bugs. It's just that the new Siri is still in the waiting list looking at me. The battery does drain a bit quickly, but it's acceptable.
May I ask, OP, are you brave enough? Or should we wait for the public beta version before diving in?
If you’re an indie hacker or solo dev using DeepSeek official models + Agents in 2026, this desktop workspace is worth trying.
Who else is building with DeepSeek V4 (or other API models) for agents? Share your setups, wins, or gotchas below 👇
Let’s exchange notes. #AIAgents #DeepSeek #IndieHackers #BuildInPublic
Biggest win: Native strength on Chinese-language + technical tasks. For anything involving Chinese docs, WeChat-style workflows, or bilingual projects, it feels more natural and accurate right now.
Also great for cost-conscious indie builders — you get high performance without massive GPU investment or local hardware headaches.
Key lessons so far:
Start simple: Use it for repetitive coding + research loops first.
Good system prompts + clear tool definitions make a huge difference.
The GUI turns agent debugging and file changes into a smooth, visual experience.
The official models are iterating quickly. Early 2026, but the momentum is strong.
My test setup:
DeepSeek V4-Pro / V4-Flash via official API
Multi-step agent tasks (research → code → test → iterate)
Long-context projects (building a small Crypto x AI side tool)
Compared to Claude 4 + Cursor: ✅ Excellent speed and cost efficiency (especially V4-Flash) ✅ Strong at structured tool calling and consistent long-running flows ✅ Very capable on multi-step reasoning, coding, and bilingual tasks
❌ Still refining pure creative writing in some cases ❌ API rate limits apply during heavy usage (but generous pricing helps)
I switched parts of my workflow to a DeepSeek-powered GUI Agent using the official DeepSeek API models — and it’s genuinely competitive with Claude + Cursor for many daily tasks.
As a former Big Tech engineer now building indie, I’ve been testing fast, cost-effective AI Agents in 2026. Here’s what I built and learned. 🚀