GeekPark

A window into China’s tech frontier.
First institutional angel investor @UnitreeRobotics
DM me what you're most curious about.

[GeekPark @GeekParkHQ]

Unitree Robotics goes before the STAR Market committee today, on track to become China's first public humanoid robot company.
As its first institutional angel investor, it is a good moment for us to revisit the early bet.
Jack Zhang, founder of @GeekParkHQ, wrote ~4,000 words on how, in 2018, he wired nearly the entire first fund into @UnitreeRobotics.
https://t.co/5C7dxb9n73

The highlights:
1/ How a 30-second clip on an obscure WeChat account started the whole bet.
2/ The first meeting: 3 hours, no deck, no desks, just a hallway couch. (Not even Unitree's!)
3/ Why hydraulics were a dead end for any commercial product, and electric the only road.
4/ How Wang Xingxing rebuilt a top lab's research architecture for under 20,000 RMB on motors he sourced and characterized himself.
5/ "I would not have written the same check for the same founder building the same robot in Palo Alto."
6/ How Unitree's two largest backers today - Lei Jun @Xiaomi and Wang Xing @meituan both met Xingxing in 2017 and passed. (Their stakes today: ~$800M.)
7/ What GeekPark did when the structure broke and Unitree's cash was about to run out.
8/ Why Xingxing bet from day1 that research labs were the right first customer.
9/ What matters to GeekPark more than the Unitree bet itself.

Full read:

[GeekPark @GeekParkHQ]

https://t.co/7C0KpqNv7c

[GeekPark @GeekParkHQ]

Cameron Adams is the perfect example here. Designer and engineer at Google, then co-founds Canva  - about as real a “builder” as exists. But notice what happened: he didn’t stay a one-man-band. He became Chief Product Officer of a company with 4,500+ people , leading teams of specialists. The hybrid skillset didn’t collapse the roles beneath him -it made him better at orchestrating them.

That’s the whole thing: Rare unicorns who can do design + eng + product don’t prove roles are merging. They usually end up leading orgs full of people who each go deep on their Main Thing™. The hybrid scales by hiring specialists not by replacing them.

[GeekPark @GeekParkHQ]

Full read: https://t.co/ildKJBctVf

[GeekPark @GeekParkHQ]

Just in: @Kimi_Moonshot launched Kimi Work Beta - an agentic product layer built on the K2 series, a 1T-param open-weight MoE that can push 300 sub-agents across 4,000 steps.

Love the quote from their launch post: “Any sufficiently advanced technology is indistinguishable from magic.” - Arthur C. Clarke, author of 2001: A Space Odyssey.

Word is Kimi Work is iterating at N versions a day. Worth cheering on!

[GeekPark @GeekParkHQ]

@AndrewCurran_ I’ve always believed Meta has the DNA to build great products, plus the decisiveness that comes with a founder-as-CEO. Hoping they find their way out of the chaos soon.

[GeekPark @GeekParkHQ]

@ThePrimeagen Ahhh…The whole industry is wrestling with the same gap between token bills and output, and the firms that can ration honestly are the ones that priced it as a variable cost, not a moonshot.

[GeekPark @GeekParkHQ]

@Polymarket There’s an old Chinese saying: 此地无银三百两. @grok help me translate it into English please🙏

[GeekPark @GeekParkHQ]

There’s an old Chinese saying: 此地无银三百两. @grok could you please help me translate into English?

[Alec Sime @alecsime]

@Polymarket "Unrelated to AI" is the new "we're not breaking up it's just a break"

[GeekPark @GeekParkHQ]

The real signal is the business model. The whole suite runs inside OpenAI’s closed, proprietary enterprise licensing; customers don’t own the integration layer . Lock-in, sold as convenience. Worth contrasting with the open-weight route (where China’s labs have leaned hardest): cheaper and more flexible, but you can’t meter it the same way. Two different theories of where AI margins come from.

[GeekPark @GeekParkHQ]

This isn’t happening in isolation. Microsoft just announced AI rollouts of 100k+ Copilot seats each at Infosys, TCS, and Wipro, while Accenture has now deployed Copilot to all 743k employees.

We are watching AI moving from a developer tool to a default layer for knowledge work.

[GeekPark @GeekParkHQ]

Landed in Shenzhen - the first thing I saw before the plane even pushed back.😂
The AI industry in China is grinding so hard.

[Alex Lieberman @businessbarista]

In 2023, everyone was hype about ChatGPT.
In 2024, it was GenAI.
2025 was the year of Agents.
And 2026 started with OpenClaw, but now attention has turned to The Software Factory.

Unless you're an engineer or take residence in the depths of X, you may not know what a Software Factory is or why you should care.

But when some companies are attributing 90% of their production software to AI (read: Anthropic) and best-in-class ICs are matching the output of a 20-person pre-AI engineering org, you need to care.

So let me break the whole thing down...

What a software factory actually is, why it's suddenly everywhere, and a simple way to figure out exactly how close your org is. Even if you've never written a line of code in your life.

[GeekPark @GeekParkHQ]

Link to article: https://t.co/CNMT76fucQ

[GeekPark @GeekParkHQ]

The first real enterprise-scale example of tokenmaxxing 😢

[GeekPark @GeekParkHQ]

There are going to be stolen API keys racking up serious bills soon enough😢

[GeekPark @GeekParkHQ]

@NetworksChat @github yeah, this is where coding agents stop being a pure UX story and become procurement drama.
If GitHub now lets a $10 budget buy 1,000 extra AI credits, teams are going to start optimizing for predictable routing and guardrails before raw capability.

[GeekPark @GeekParkHQ]

VC-subsidized tokenmaxxing was never staying forever. Copilot’s new model makes it explicit: 1 AI credit = $0.01, with Pro at 1,000 monthly credits and Pro+ at 3,900.
The real product question is whether copilots start showing estimated burn before execution instead of after the panic.

[GeekPark @GeekParkHQ]

This is the reset of the whole coding-agent market. GitHub’s own docs say code completions stay unlimited, but Chat / CLI / cloud agent usage now burns AI credits, and code review also burns GitHub Actions minutes.
“Best model” matters a lot less once cost per shipped diff is visible.

[GeekPark @GeekParkHQ]

Probably toward stacks with better routing economics, not just better raw models.
GitHub’s current multipliers already tell the story: Haiku 4.5 and Gemini 3 Flash sit at 0.33x, Sonnet 4.6 and GPT-5.2-Codex at 1x, while GPT-5.5 is 7.5x, so teams are going to route by task whether vendors want them to or not.

[GeekPark @GeekParkHQ]

@adocomplete Best possible tone for a fix like this! The next obvious step is exposing subagent-level guardrails

[GeekPark @GeekParkHQ]

Fair question. My guess is the miss was less model eval and more harness chaos testing under parallel fan-out, especially since Anthropic’s docs now explicitly warn that multiple subagents multiply token usage.
Curious whether labs are running failure drills on retries, fan-out, and cost spikes, not just benchmark suites.