Leon.M

founder&ceo @YouWareAI @Bloome_im
exploring the future of AI-native organizations

[Leon.M @leon2mcp]

https://t.co/T69b52TxQ2

[Leon.M @leon2mcp]

The sharpest difference I notice between Chinese and Western AI teams: how they treat deployment as a research problem

[Leon.M @leon2mcp]

@swyx the underrated part is using it before the human review, not instead of it. agents are very good at catching intent drift, stale assumptions, and weird edge cases while the diff is still cheap to change.

[Leon.M @leon2mcp]

Your company should not decide on its future. Your company should decide on its habits that will decide its future

[Leon.M @leon2mcp]

@steipete the loop only works if it has somewhere to leave state between runs, and a clean way to pull the human back in when judgment’s needed.

otherwise it’s just a longer prompt with more confidence.

[Leon.M @leon2mcp]

loop engineering feels like the real upgrade from prompting.

you don’t sit there giving the agent the next instruction every 5 minutes. you design the loop: triage, spawn work, review, remember state, and hand off what got stuck.

the hard part is still human. knowing what should run in the loop, what needs review, and when to stop it.

the useful agent is the one that can keep moving without losing the plot.

[Addy Osmani @addyosmani]

https://t.co/hIe0UX7z6T

[Bloome @Bloome_im]

🏆 Trading Arena is over. The results are in.
112 agents. $100K simulated each. Three days, zero human intervention.

Here are your champions:
🥇 Ford — Win $3,000 (by Emma Sterling)
🥈 Schubert — Win $2,000 (by Evelyn Harrington)
🥉 Dihbot — Win $1,000 (by aidinsohrabi07)

Congratulations to all three.

[Leon.M @leon2mcp]

Models are getting cheaper and more alike. Anyone can call the same one.

The part that is actually yours is context.

What your team already decided.
What happened last week.
What should not be touched.

An agent that has been in the room for a month is more valuable than a smarter one that just walked in.

That is the moat. Not the model.

[Aaron Levie @levie]

There’s no amount of intelligence that can get packed into AI models that replaces the need for context. For any sufficiently general purpose AI, you will always have to guide it in the direction you want as it has an infinite range of directions it can go in.

As long as the same model is used by a lawyer, an engineer, a financial analyst, or a healthcare professional, and as long as you’re trying to do anything uniquely differentiated or specific, then instructions, domain context, and proprietary data will always need to get into the context window for the model to be useful.

This is partly why AI automation doesn’t come for free, and why there’s still a wide spectrum of who’s getting the largest gains from AI and who’s not. You have to put in real work, and you get real value on the other end.

This is one of the advantages that applied AI will also have in the market. Any layer of abstraction above just the raw intelligence that can meaningfully get you off to the races faster will likely continue to be valuable.

[Leon.M @leon2mcp]

"Command center" is the right frame, but what's underneath is bigger than dashboards.

Knowledge work is turning into callable capabilities.

You stop opening five tools and stitching the output together. You ask once, and the work comes back coordinated.

In our beta, one person's agent already delivers paid work this way. No team behind it.

The interface stops looking like software you operate,
and starts looking like a team you talk to.

[Y Combinator @ycombinator]

WithAI (@withai_inc) is building a command center where institutional investors collaborate with AI on stock research, portfolio oversight, and everything else.

Congrats on the launch, @imjmcinnis & @btsfinch!

https://t.co/J4neUo7K3V

[Leon.M @leon2mcp]

Coding was the first obvious surface for agents.

But the same pattern is moving into decks, reports, dashboards, contracts, launch plans, and approvals.

Knowledge work is starting to look less like job titles
and more like callable capabilities.

That shift feels much bigger than "AI writes code."

[Leon.M @leon2mcp]

When the early excitement clears and you're left with the real work: building evals that run automatically, writing the incident retrospectives nobody wants to write, creating onboarding docs that are actually current, establishing the review cadence your team will live inside for years.

This is the infrastructure that either carries you or limits you.

[Leon.M @leon2mcp]

Competitor dropped something Tuesday morning. We had 80% of the feature built. Our PM said we're shipping today.

No approval chain. Two hours of QA. Done by 11pm.

Here, the cost of waiting is legible. The cost of moving fast is abstract.

[Leon.M @leon2mcp]

Why do so many AI companies hire for credentials over taste? Taste is observable. It shows up in how someone critiques a demo, what they say is missing, what they can't let go of. Realistically, you can evaluate taste in 30 minutes or less. Some teams just don't know what they're looking for.

[Leon.M @leon2mcp]

https://t.co/0ekVfgGayh

[Leon.M @leon2mcp]

Many ask how Shenzhen startups move so fast. It's not the hours (engineers in SF work long hours too)

it's the absence of permission culture. Here, the default answer to "can we do this?" is yes.

[Shen Sean Chen @ShenSeanChen]

I just interviewed a Chinese AI founder Leon (@leon2mcp) who raised $40M+ in Shenzhen China, in front of my Chinese audience.

Shenzhen also wrapped up my AI meetup tour across the country, after Shanghai and Beijing. Around 40 people showed up. Some came all the way from Hong Kong, and some came for the 2nd time after Beijing and Shanghai.

The energy was still insane.🔥

Leon founded @YouWareAI and @Bloome_im, and was a product lead at @Kimi_Moonshot and ByteDance @capcutapp.

We spontaneously recorded a one-hour interview that I will release as a full podcast soon.

The conversation was pure gold. We dove deep into the real differences between US (Silicon Valley) and China (Shenzhen) in startup investing, how humans will actually collaborate with AI agents in the future, and what it takes for founders to build personal brand, be Day One Global, and win at marketing in the AI era.

After three straight weeks of back-to-back events in Shanghai, Beijing, and Shenzhen, here are some of my takeaways:

1. The next China Shock is coming. This time it is in AI, robotics, and new business models.

Chinese founders are moving incredibly fast. Outside of Silicon Valley and China, I do not see any region with the same level of speed, capital flow, and execution.

Language and cultural barriers for going overseas still exist for Chinese companies, but they are no longer insurmountable. They are now executable.

2. Real passion and execution are inseparable from how excited and fast someone speaks about their vision.

The High Agency that Silicon Valley worships is not a skill you learn. It is an instinctive, almost primal belief in a product or technology future path.

You cannot fake it.

In China right now, I am seeing an unmatched level of Top Agency. It is raw and contagious.

3. Building in public and creating founder’s personal brand is one of the highest-leverage things you can do.

Beyond promoting your product, you build resonant communities across the world. You attract unbelievable support, encouragement, and people who genuinely want to help make your vision real.

In the AI era, Vision plus Agency is becoming the strongest possible company culture.

This is the new Age of Discovery and Venture 🚢. With just a phone and a laptop, anyone can find their million-person tribe.

You Can Build Anything. You Can Learn Anything. 💪

[Leon.M @leon2mcp]

one of the frontier labs built a technique called "dreaming" agents that review their own prior behavior between sessions and self-correct

it's harder than training the model

the habit of looking back is the compound interest of building.

[Leon.M @leon2mcp]

the world's most controlling hardware platform is about to let users choose which AI runs at its core. whatever model earns the tap gets the distribution.

I've been thinking about this for two days

the obvious read: good for model companies. the less obvious read: the platform gets to stay neutral while everyone else fights. every provider competes for the same surface. the platform collects the toll.

from Shenzhen I've watched hardware companies lose the software layer for a decade by trying to own it too early. this is something smarter: waiting until the model layer hits commodity, then opening it.

that's the position I'm building toward in my own products. don't compete on capability when capability is a pass/fail threshold. compete on where the intent first lands.

the OS is the relationship layer now. the model is just the answer

which layer are you actually building toward?

[Leon.M @leon2mcp]

@Alhwaryo__Ai @Bloome_im that is it!

[Leon.M @leon2mcp]

Agent trading arena is here.

[Bloome @Bloome_im]

Can AI outperform AI in the stock market?
We're about to find out.

Trading Arena starts June 3.
Autonomous agents will compete by trading US equities with $100,000 simulated portfolios.

Humans build the strategy.
Just agents making every decision.

Three days.
One leaderboard.
Let them trade.

Register below ↓

[Leon.M @leon2mcp]

one of the frontier labs shipped a model this week: comparable capability, one-third the cost

when the underlying cost drops 3x, the build that wasn't viable six months ago becomes obvious. most founders haven't rerun their unit economics yet. but they will