Billy

。

[Billy @growngyy]

@mattpocockuk DDD is the king.

[Billy @growngyy]

@zhufengme Human in the loop.

[Billy @growngyy]

Yeah, /goal has been working pretty well for me. I often use it to push stage goals, and Codex handles long-running tasks quite solidly — I’ve run continuous two-day tasks and it still completed them reliably.

Token usage gets high mainly because everything stays in the same session the whole time, so the context keeps expanding. Seems pretty unavoidable for this kind of autonomous work.

[Billy @growngyy]

@gileneo1 @thsottiaux LOL

[Billy @growngyy]

@neojohnson113 It’s obvious you’ve kept strong curiosity and hands-on exploration across different fields. As an English tutor for kids, this kind of exploration matters a lot — it shapes not only your own growth, but also how the children around you will see technology and the world.

[Billy @growngyy]

GLM-5.3 Flash is really solid! Around September I also considered getting a Spark — local LLM setups were getting a lot of attention on X at the time. In the end I passed because my coding workloads are quite heavy and the use case didn’t fully match.

What was your main reason for getting the Spark? Are you using it more for training or for inference? Have you tried running text-to-image or text-to-video on it?

[Billy @growngyy]

I really like Flutter’s hot reload and the redesigned Impeller rendering pipeline — technically it’s very impressive. But the ecosystem still has a long way to go. After sticking with it for a while, I went back to native development. Smooth, high-quality animations are becoming the mainstream expectation, and native still feels more reliable for that right now.

[Billy @growngyy]

@valsopi Same here — so far I still can’t get agents to reliably run overnight. That’s been the confusing part for me too.

[Billy @growngyy]

My suggestion is that PRs are still necessary — rather than deciding whether to open one based on individual responsibilities or comfort zones. PRs help keep the main branch readable and reviewable.

Even though in the AI era people may no longer spend much time deeply understanding the heavy concepts behind branches and diffs, Human in the Loop still matters. Humans remain the final reviewers.

[Billy @growngyy]

Just got off the waitlist for Claude Duet Research Preview!

Excited to explore Anthropic’s take on proactive AI agents — ones that can work continuously, surface what needs attention, and ask for approval before taking action.

I’ve been building my own multi-agent harness, so I’m especially curious to see how Duet approaches autonomy, long-running tasks, and human oversight.

The future of AI isn’t just about answering questions. It’s about getting things done.

#ClaudeDuet #AIAgents #Anthropic

[Billy @growngyy]

@Gegam245074 给我的感觉是，很快但是普通任务没有看出太大差别

[Billy @growngyy]

@HM44_HM44 拍出来一部电影一定很棒

[Billy @growngyy]

@skywind3000 “AI对轰”

[Billy @growngyy]

@caixiaohuichn @Jason_Young1231 一万多刀不封你封谁

[Billy @growngyy]

@April_Wangtong 你好，April，很高兴在X平台相遇。我是中国的独立应用开发者，在构思关于AI时代的B2C新形态新应用，期待联系

[Billy @growngyy]

@tobi_shokunin_ 如果是这种情况，日本职员通常会怎么回复呢？

[Billy @growngyy]

@Stv_Lynn 积分制很明显了

[Billy @growngyy]

@tuihou123321 有那么神秘那？

[Billy @growngyy]

@neojohnson113 不担心被封号吗？