Pinned Tweet
Duet is different from other agents like Muse, Dot, or Grok.
The waitlist is open for small businesses. You must enter your business name to join. It’s a small detail, but it shows they’re mainly targeting business use cases.
Looking further at the demo, all the use cases are around business activities: managing, operating, and collaborating with your team. There’s no mass-market consumer use case like booking flights or arranging hotels. They even say it directly: personal errands like booking tickets aren’t quite as strong yet.
Because it touches business data, a big part of the announcement focuses on safety and security. They even admit its limits openly: “This is an early version of Claude Duet, and it still might misread a thread or miscalibrate what’s urgent.” In a business context, that kind of transparency matters.
It’s aligned with Anthropic’s current direction. They’ve been focusing on enterprise, and now they want to reach a larger market. A product for smaller businesses is a natural move.
We might argue that all the other bots can do the same.
Yes, that’s true if we only look at the features: always-on agents that have access to many tools, do the job for you, and proactively communicate with you.
The point is how they package those features and position the product. That decides which audience they focus on acquiring. And in the long run, it shapes how people think about Duet.
So on the product surface, they’re the 4th to join the race. But in positioning and market segment, they’re the first to go after it this clearly.
Time will tell how far they go. For now, it’s an interesting game to watch.
So here is my prediction:
Fable 5.5 WON'T release today!
It's based on what people predicted in the last 7 days.
Spread this for a world without skeletons
People rightfully pointed out that mobile apps CAN show you something instantly because you already have all the frontend pre-downloaded on your phone (unlike websites).
In this case, showing a loading indicator is fine if it takes a while, but PLEASE don't show fake content wireframes unless you 100% know final dimensions. If your content is dynamic, the effect on the user is the same as a layout shift. Just have the fancy white flash over the container or something like that.
But of course, optimize your app that you don't have to show them for most users at all!
I've been slowing down in the last 3 weeks.
Every time when I have >2 things running I can't manage and review them well. Some tasks I only need to prompt and forget, some need my attention. The latter is what I set the limit.
Even when I go slower, it's still way faster than when I write all the codes ourselves. So I need to find the rhythm that I can manage the quality.
Vu retweeted
Working with AI when you're a literal person can be a chore. AI is trained to presume people mean something different than what they say, but I mean exactly what I mean, and it makes me bang my head when it says, "I thought you meant..."
With 30K PRs per day sometimes they accidentally make some builders look for new project ideas 😅
ANTHROPIC 🔥: Personal Finance feature on Claude is being prepared for the upcoming release (again)!
> Personal Finance will work through Plaid integration.
> Claude will be able to analyse user’s bank data and build various dashboards on top of it.
> During development, this feature was named as “Claude Money” but has been renamed in the latest update.
> Given the new onboarding screen and an addition of Plaid integration logic, there is a high chance we would hear about it relatively soon.
“Claude Monet” would indeed be an epic name for this.
Soon? 👀
I think about this post from you @loveAIwang when write my post :D Would love to learn how you feel about apps in Chinese. I hope I don't bring some weird points from outsider 😄
xitter.cf/loveAIwang/status/2110…
I spent 7 years in the core team building the biggest app in Vietnam (70M+ users). I researched the Chinese app ecosystem heavily, especially WeChat. From what I've seen, most of the mess comes from business decisions, not from a lack of design skills or a different cultural taste.
There are 2 different things here: the artistic side of the design, and the complexity of the information. The latter overwhelms the former.
In China, and partly in Vietnam, companies want to build super apps. A super app is one application people use for every single purpose in their everyday life: messaging, payments, ticket booking, short videos, and hundreds more. When you have that ambition and try to fit in that many features, the classic UX rules get traded off, and many of them simply can't be applied. As a result, if you're used to the simplicity of Western design and how much it cares about information architecture, Chinese apps look like chaos.
And they are indeed chaotic. No cultural explanation can change that. People do find it hard to find anything outside of what they usually use. But in my humble opinion, these are deliberate decisions, not poor design skills. When building a super app, you must optimize for cross-selling. When users use one service, you must get them to use more related (or even unrelated) services. What's the most basic design approach to achieve that? Yes: put everything on the screen, next to each other, near the one users just picked. Do it intensively and you create a mess. But that mess perfectly fulfills the original objective.
Another thing that makes it harder: mini apps. Super apps open their platform to thousands of third parties, from big brands to small shops. Their interfaces are built by those partners, not by the super app's designers. There are design guidelines, but enforcing them is very challenging. Strict review slows down partners, while the platform wants the ecosystem to grow fast. And users don't see the boundary. To them, it's one app. So every low-quality mini app makes the whole super app look rough. That's why at many points, the design looks odd and low quality.
So the nature of the ecosystem creates a different level of information architecture, which leads to a chaotic design on purpose. When you see that mess, it's very hard for you, who are familiar with and inspired by Western design, to call it good design. Even Chinese people see it as a mess. A mess is a mess, regardless of culture.
At this point, I want to emphasize again that this is not a skill issue of Chinese designers. They have good designers, a lot of them. I've known many great designers from Tencent, the maker of WeChat. When we spot basic design issues like bad padding or misaligned elements, they are just... design bugs. Just like design bugs in Western products.
No cultural difference turns bad padding into good design in China.
Then let's talk about the artistic side of design. It's difficult, almost impossible, to make that mess beautiful in the common sense. Interestingly, they approach it in a different way: by bringing more cultural elements into the design. For example, most Western apps look the same throughout the year. But in Chinese, and some Vietnamese, apps, you'll see something special for every traditional event. The colors, the animations, the decorations... everything changes to celebrate the occasion. Think Google Doodles, but at a much larger scale. This is how they solve the artistic side of design from an emotional perspective. So culture does show up in these apps, just not as an excuse for the mess. There are many other examples of how they adapt their design.
This is an interesting topic. I first touched on it in 2013. 13 years later, people are still questioning it and trying to explain it. Seeing it come up again today made my weekend.
I love these discussions on X where I can learn every day.
It’s astonishing how many people believe RAG is dead just because models got smarter and agents can now “call search tools”.
Smarter LLMs don’t eliminate the need for retrieval. Context windows are still finite, and when the answer is buried somewhere among millions of private enterprise documents, the real challenge is finding the right information efficiently.
Agentic search doesn’t magically eliminate that challenge either.
When people say “just let the agent search,” the question is: what actually powers that search tool?
An agent can decide what to search and when, but it still needs an efficient retrieval mechanism underneath. At enterprise scale, that often means proper document processing, indexing, filtering, and ranking. Giving an agent a search tool doesn’t make that infrastructure disappear.
And accuracy is another story. In domains like finance, finding a relevant document isn’t enough. Extracting exact figures from complex tables or connecting information across multiple pages requires specialized retrieval, document understanding, and verification. Simply giving an agent a “search tool” doesn’t solve these problems.
Another common misconception is that RAG is synonymous with vector search. It isn’t. Vector search is just one of many retrieval techniques.
RAG isn’t dead. The naive “just dump everything into the context window” or “just let the agent search” approaches are.
20 year in HR and doesn’t know how to behave. If any HR or people does this in my team they are in trouble.
Replying to @fw_andyyy
20 year HR professional here. Many times I’d make candidates wait a few extra minutes to see how they handled it when they got frustrated. Weeded out a lot of hot heads. Me, I’m looking for patience and understanding.




















