Founder @ ShengFM 声动活泼(China’s leading podcast studio)
Host @ Voice of Context | 科技早知道 (What’s Next in Tech)
Exploring how AI reshapes what comes next.
OpenClaw agents just went mainstream in China.
But some of the first users who rushed to install them are already uninstalling.
Hype moved faster than the use case.
Meanwhile, cloud vendors, AI firms, and even people charging a few hundred RMB for installation are already making money.
The early agent economy is getting messy.
Full breakdown ↓https://t.co/2HofmX64ea
I actually think the idea of people having agent “twins” that interact socially could become meaningful.
But Moltbook itself doesn’t seem like a particularly strong product or data asset.
This feels less like a traditional acquisition and more like PR — or simply buying attention around the “AI agent social network” narrative.
Sadly, after our latest podcast episode on AI and careers, I saw a lot of anxious comments from listeners in China.
Some said:
“This will be the golden age of knowledge elites.”
Others said:
“Ordinary people will feel hopeless.”
What struck me is that both reactions assume the same thing:
a fixed-pie world.
But reflecting on it, I realized something about my own perspective.
I’ve spent much of my career inside the tech and startup circles — in Silicon Valley and in China — surrounded by founders, investors, and builders.
Sometimes that bubble makes it easy to forget the broader reality of society.
If you look at the world honestly, it’s more like a bell curve.
Most people have a job, a boss, and a relatively defined role in an organization.
Only a small minority consistently operate with strong curiosity, agency, and the willingness to execute.
Some of them become entrepreneurs.
Others become the people inside companies who can independently drive projects and build things.
What AI may change is not human nature — but the distribution of tools.
For a long time I’ve been influenced by Peter Diamandis and his book Abundance.
It shaped how I think about technology: as something that expands possibility rather than simply redistributes scarcity.
AI feels like one of those moments.
It doesn’t guarantee success for everyone.
But it dramatically lowers the cost of learning, experimenting, and building.
In that sense, it democratizes capability.
If someone can learn quickly and act on their curiosity — like many builders I see today — the future still looks much more like abundance than scarcity
And the people who benefit the most will likely be those who choose to explore the tools rather than fear them.
During GTC Week, we’re hosting a Multimodal AI event in Silicon Valley — and the Seedance team will be joining us to present, alongside leaders from TikTok, Shopify, Adobe, NVIDIA and more.
The conversation will focus on applied AI video, commerce, and the realities of scaling production and revenue.
If you’re around for GTC, would love to see you there.https://t.co/0QF2aD6Lnm via @LumaHQ
Every founder I know is experimenting with OpenClaw right now.
Some observations from running a media company and trying to integrate AI into our workflow:
1. AI does save a lot of time.
We haven’t rolled it out across our entire company yet. My co-founder and I are testing it separately inside a few controlled Lark channels.
Because most of our team is based in China, the setup itself is a bit different from what many people in the U.S. use. We currently run our internal AI workflows through Tencent’s one-click deployment, combined with Lark as the collaboration layer.
We’ve also tried several standalone AI cloud services from different model providers. Some worked well, but teammates occasionally ran into stability issues, so this setup ended up being the most practical for now.
For research, topic discovery, and crawling news sources, the time savings are very real.
But that doesn’t mean we need fewer people.
We are still hiring producers — people with taste, curiosity, and strong editorial instincts. AI doesn’t replace that.
2. Workflows have to become agent-readable.
I realized something surprising: many workflows that work perfectly for humans simply don’t work for agents.
Training an AI agent is very different from training an intern.
You have to break everything down very explicitly:
(a) what rules allow it to gather the right information
(b) what format and style the output should follow
Things that feel like “one step” for humans are often two or even three steps for machines.
Large models do have a certain level of consistency, but the output often still diverges from what I actually want.
Interestingly, this process of working with agents has forced me to rethink and clarify my own workflow. In trying to make it understandable for machines, I ended up making it more structured for humans as well.
3. Speed of change
One thing that really struck me this year is how fast the tooling itself is evolving.
Last year we already had some internal automation in place. We were building small tools and workflows, often reusing them with the help of external contractors.
At the end of last year, we were still writing vibe code ourselves.
By the beginning of this year, we were already asking agents to generate most of that code for us.
The speed of change has been honestly a bit shocking.
4. A thought about organizations.
I named our internal system “The Mind”, inspired by Iain M. Banks’ Culture series.
There is a line I particularly love:
“One day the Minds would start thinking
how wasteful and inefficient
the humans themselves were.”
It’s a beautiful irony.
Humans are both incredibly wasteful and incredibly efficient at the same time.
Which makes me think that the next phase of work may not actually be about efficiency.
Maybe the real frontier will be taste, creativity, and perspective.
It's very calming when I realize that.
we just wrote the ultimate beginner's guide to OpenClaw
almost everyone @every has one now, and they have completely changed the way we work and live. we're using our claws to:
- build product
- answer customer service queries
- book hard-to-get restaurant reservations
- track our reading notes
and much more
this is the guide we wish we'd had at the start:
https://t.co/66n3Wz6MT0
Kimi’s 20-day revenue surpassed its entire 2025 total.
This wasn’t about better models.
It was about execution velocity.
In the agent era, speed is becoming a strategic moat.
Full analysis ↓https://t.co/wHBBExtDfm
Zhipu’s GLM-5 topped OpenRouter this week.
Less than 24 hours after launch, I spoke with the head of https://t.co/U3MG5VoNj0 . @ZixuanLi_
The ranking wasn’t the whole story.
Here’s what actually matters from https://t.co/yEFVq3OWVb 👇
The bigger issue is compute saturation.
When one generation succeeds, it consumes the capacity needed for the next.
Upgrade too fast → cannibalize revenue.
Upgrade too slow → lose users.
This supply shock dynamic defines Chinese model economics.
@Handy@HandyCX The cleaning “pro” is a no show and the contact info is wrong. This “service” is a nightmare. We have people coming and no time to clean. Thanks a lot, fraudsters. Stay away, everyone!
Portfolio hiring: SAL is building vertically integrated software + hardware platform to power our real-time retail distribution network. https://t.co/41Df6bDTbI https://t.co/RYNVWtiZQx