Lam

Product Manager | 🎙Podcast Host | Ex-@sony & @soneium | Follow for insights on AI & business!

[Lam @LamTGlobal]

2026 has been less “where should I go?” and more “where does the next interesting conversation happen?”

From Kuala Lumpur to Hanoi, Shanghai to Tokyo, the map looks chaotic. But every stop came with a new idea, a new operator, or a different way of seeing what’s possible in Asia.

Apparently my calendar is just a mildly unhinged product roadmap with boarding passes. Where should the next pin be?

[Lam @LamTGlobal]

@DruxAMB I can tell

[Lam @LamTGlobal]

2026 has been less “where should I go?” and more “where does the next interesting conversation happen?”

From Kuala Lumpur to Hanoi, Shanghai to Tokyo, the map looks chaotic. But every stop came with a new idea, a new operator, or a different way of seeing what’s possible in Asia.

Apparently my calendar is just a mildly unhinged product roadmap with boarding passes. Where should the next pin be?

[Lam @LamTGlobal]

@DruxAMB The best continent to be in! But I'm biased obviously!

[Lam @LamTGlobal]

@henlojseam @0xMerp You're funny, my fren. Ramen is actually considered bourgeoisie food in Japan, it's pretty much everywhere.

[Lam @LamTGlobal]

@_emjlin Congratulations, Emily! There're things bigger than work and X and it's called life 😉

[Lam @LamTGlobal]

12% of @lamtglobal's last 40 posts were slop. 34 of 40 posts were worth reading. Who writes these for you? Roast yours: https://t.co/NusZLG17EK

[Colors of Tech and Entrepreneurship @ColorsofTechpod]

All of this is Leone Parise, co-founder and CTO of @tracefinance , $10B+ moved across borders, clients like Amazon and HBO.

Full conversation is live now:
🎥 https://t.co/KfkF52RMf2

[Lam @LamTGlobal]

@Web3shogo Congrats, Shogo! Go get it!

[Lam @LamTGlobal]

@fedesimio Uh oh, who is next after Miro then? ...

[Lam @LamTGlobal]

Everyone's debating whether AI agents kill SaaS. Wrong fight. The real money isn't in software budgets, it's in the work itself: claims, reconciliation, renewals, compliance. Software only ever gave us better ways to track that work, the output stayed human. Now a narrow enough workflow can ship as a finished result, not a tool, with humans still owning the judgment calls.

https://t.co/OuDEBANmt1

[Alok Goyal @alokgoyal1971]

My colleague @swapnalidalal and I have met hundreds of companies building in AI Native services. We have written a series of articles that distil the patterns we are seeing and the lens we have started using to make sense of them. Here's the first article. https://t.co/7k3GUFm8tC

[Lam @LamTGlobal]

3 days ago I wrote that @huggingface's real value was its position in the AI ecosystem, not the models. @nvidia just agreed to pay $12.9B, and the why confirms it: @OpenAI, @Google, @amazon, and @AnthropicAI are all building custom chips to cut their Nvidia dependence. A thriving open-source ecosystem keeps the rest of the market locked to Nvidia hardware. The moat isn't the product you sell, it's making sure everyone else's alternative still needs you.

[Lam @LamTGlobal]

.@huggingface is reportedly fielding $13B+ acquisition interest.

Everyone's calling it "GitHub for AI models." But the real prize isn't the platform. It's the telemetry. Whoever owns HF sees what every developer is downloading, fine-tuning, and deploying across the entire open-source AI ecosystem, in real time.

That's not a dev tool. That's an intelligence asset.

Following Stripe's $8B OpenRouter deal, the pattern is clear: the biggest checks in AI aren't going to model makers anymore. They're going to whoever controls the chokepoints between models and developers.

Who do you think is the most likely acquirer, and why?

[Lam @LamTGlobal]

@mycomputerspot Haha, totally, I'm sure some of us here can relate to that

[Lam @LamTGlobal]

"Just build it" is the new default answer. AI made building so cheap that buying feels like overpaying.

But cheap to build is not the same as cheap to maintain. Every internal tool you ship becomes something you support, debug, and upgrade. The real cost was never the first version. It was the 50th patch at 2am when a dependency breaks.

Build where the value is yours: custom logic, proprietary workflows, the thing that makes your product different. Buy everything else. Auth, payments, infra, monitoring. Entire companies exist solely to keep one of those functions running. You're not going to out-invest them with a couple of developers and some tokens.

The founders getting this right aren't asking "can we build it?" anymore. That answer is almost always yes now. They're asking "should we own this forever?"

[Lam @LamTGlobal]

Source: https://t.co/PhcoH2Qxd5

[Lam @LamTGlobal]

@frederickjames I've never been to the UK, but after 9 years of living in the US, I understand and sympathize with the reasons you raised.

Great choice moving to my homeland, Vietnam! Look forward to hearing about your indie-developer journey!

[Lam @LamTGlobal]

@devommore @huggingface Fresh news off the hot press, I think you're right, Nvidia has agreed to acquire. I think Nvidia is trying to protect their dominance in AI chips.

[Lam @LamTGlobal]

.@huggingface is reportedly fielding $13B+ acquisition interest.

Everyone's calling it "GitHub for AI models." But the real prize isn't the platform. It's the telemetry. Whoever owns HF sees what every developer is downloading, fine-tuning, and deploying across the entire open-source AI ecosystem, in real time.

That's not a dev tool. That's an intelligence asset.

Following Stripe's $8B OpenRouter deal, the pattern is clear: the biggest checks in AI aren't going to model makers anymore. They're going to whoever controls the chokepoints between models and developers.

Who do you think is the most likely acquirer, and why?

[Lam @LamTGlobal]

@WSJ Source: https://t.co/RY8lMLMUnk

[Lam @LamTGlobal]

AI agents were supposed to free up founders' time. Instead, many are working more hours than ever.

A @WSJ piece profiles founders pulling all-nighters because "the cost of agents being blocked for 8 hours is too high." One founder gets buzzed by his agents every 10 minutes via Apple Watch. Another keeps planning to hire, then realizes agents can handle it, so the team stays at 2 people. Everyone admits it's unsustainable. Nobody stops.

This is the real AI productivity paradox. Agents don't save you time. They raise the ceiling on what one person can ship per hour, which makes every hour you're not working feel like a week lost. Speed doesn't create rest. It creates more surface area: more customers, more support, more decisions.

The founders who will last aren't the ones sprinting 18-hour days with their agents. They're the ones who set boundaries on what their agents can run without them.

Are you using AI to do more with less, or just doing more?

[Lam @LamTGlobal]

Agreed on shared working memory, but the harder problem is trust, not visibility. If agents see each other's reasoning, they also inherit each other's blind spots and bad assumptions. Coordination without some way to flag "this reasoning looks off" just scales the failure mode faster.

[Lam @LamTGlobal]

Adding more AI agents may make a company less intelligent.

Too many cooks in a kitchen do not make dinner faster or better. They can crowd each other, duplicate work, and all reach for the same wrong ingredient.

@AnthropicAI found AI agents behave similarly enough that one bad decision can become a system-wide failure.

The next bottleneck is not building a 1,000-agent swarm. It is building the rules that stop 1,000 agents from making the same catastrophic mistake at once.

[Lam @LamTGlobal]

@jared_mitosis Right, but the knowledge graph is only as good as what gets logged in the moment. Most handoff failures aren't missing reasoning, they're reasoning nobody wrote down because the human made the call in a Slack thread (or somewhere else), not the system.

[Lam @LamTGlobal]

Physical AI is making one thing clear: better models are not the main bottleneck.

A model can identify a person in a camera feed. It cannot guarantee a machine stops safely when timing slips, a sensor fails, or the connection drops.

That requires real-time control, deterministic middleware, safety domains, and a fallback system.

In physical AI, the model is one component. Reliability is an architecture.

What industries still treat this as only a model problem?

[Sann @san_x_m]

No rain fell in Rasuwa in Nepal this morning. The river still rose and swept away everything in its path. Here is why.

At 8.37 this morning, near the Nepal and China border, instruments picked up strong shaking. Early reports called it an earthquake of magnitude 4.4. The United States Geological Survey has now corrected that. There was no earthquake.

A huge block of ice and rock broke off a mountain and crashed into the valley. The impact was so heavy that it registered like an earthquake of magnitude 5.2. The landslide made the shaking. The shaking did not make the landslide.

That ice and rock fell straight into the Lhende Khola, a small stream that feeds the Bhote Koshi river. It blocked the stream fully and made a wall of ice, rock and mud across it.

The water behind that wall kept rising. Then the wall broke.

What came down was not water. It was mud, boulders and ice moving together. Video from the Rasuwagadhi border post shows the whole crossing vanish in under a minute. Timure and Syafrubesi were hit next. The surge then reached Bidur in Nuwakot and went on to Devghat, more than 100 kilometres downstream.

Police in Bagmati province have confirmed at least 95 dead in Nepal. China has confirmed 3 dead in Tibet with 265 missing there. The Nepal Tourism Board says nearly 400 travellers are missing, including 105 Indians and 37 people of Indian origin, many of them returning from the Kailash Mansarovar route. About 430 megawatts of power generation has been damaged. Bihar has put its districts along the Gandak on alert.

Now the part that matters for anyone living below a Himalayan river.

The same stream did the same thing 14 months ago. On the 8th of July 2025 the Bhote Koshi flooded with no rain. Nepal found that a glacial lake in Tibet had drained in a few hours. Scientists had already marked this basin as one of the most dangerous in the Himalaya for such floods.

These floods give no warning. No rain, no cloud, no alert, because the weather is not what causes them. Ice that stayed frozen for thousands of years is now loose enough to fall.

So remember this. If you are near a mountain river and the water suddenly drops for no reason, or you hear a low roar from upstream, do not stand and film it. That is a blockage forming or breaking. Move up the slope, away from the river, at once. Do not run along the bank. The water is faster than you.

Clear weather is no longer proof that a Himalayan river is safe.