🧁Baker’s debrief #6: what do you actually need to be there for?
AI, Silicon Valley, and the human layer of building globally
Silicon Valley is very much my home base. Most of my work, relationships, and professional life are still deeply rooted in the Bay Area.
At the same time, I’ve been fortunate to work with founders and startup communities across Asia, Europe, and other parts of the world.
That combination has made me increasingly curious about what “building globally” actually means now.
If AI lets smaller teams do more, talent can be more distributed, and products can reach users before a company has a local presence, how much does where we build still matter?
That was the question behind the panel I brought together at S-Tron Shanghai:
Beyond Silicon Valley: how is AI changing where and how we build?
I was joined by Patricia, a repeat founder and scientist now building again in biotech and healthcare; Angus, a longtime Silicon Valley technical founder and operator; and Feiyu, founder of MemOS, an open-source AI infrastructure company built from Asia for a global developer community.
What emerged was not an argument that geography no longer matters.
It was a much more practical conversation about what has become easier to decentralize — and what has not.
Execution is getting cheaper. Judgment isn’t.
One idea came up repeatedly: judgment matters more.
AI can already take on a meaningful amount of execution. Research gets faster. Engineering teams can cover more ground. Agents can handle work that would previously have required additional people.
That changes the underlying logic of how founders think about building teams.
Patricia is starting again from zero after taking her previous company through exit. Angus has spent more than two decades building and scaling software companies and is now back in a much more 0-to-1 mode with his own product. Feiyu sees the same shift from inside an AI-native infrastructure team.
The question is no longer simply:
Who do I need to hire?
It starts one step earlier:
What can AI already do — and given that, what do I actually need another person to do?
Twenty or twenty-five years ago, building serious enterprise software often meant assembling a substantial engineering team and physically going to where the talent was.
Today, someone who knows how to work effectively with AI — how to delegate to agents, evaluate their output, manage uncertainty, and know when not to trust them — can carry far more than one person could before.
AI does not remove the need for humans.
It changes what we need humans to be good at.
A similar idea has been showing up in Silicon Valley around experienced founders gaining new leverage in the AI era: not because age itself is the advantage, but because domain knowledge, taste, and judgment become more valuable when execution becomes easier.
That also resonates with what I’ve been building with Ryme: the goal is not to replace founder judgment, but to help founders prepare better for the conversations and decisions that still require it.
When more people can build, knowing what is worth building becomes a bigger differentiator.
Smaller teams can be more borderless — but capability still comes first.
AI also changes where the people need to be.
Remote work was already loosening that constraint. AI pushes it further.
If models and agents can take on more execution, teams can stay smaller for longer, and founders can increasingly choose people for capability rather than proximity.
But borderless does not mean frictionless.
Feiyu’s experience with MemOS makes the prerequisite clear: the product still has to be good enough that people want to use it.
The core team is primarily in Asia, while developers, contributors, and users already participate from different parts of the world.
Open source lets the product travel before the organization does.
Developers discover it, use it, contribute back, and bring it into their own communities before the company has built a formal presence there.
The simplest version may be:
Build something useful enough that people want to take it with them.
The product can arrive before the company does.
For developer infrastructure in particular, the traditional sequence of market entry is starting to reverse.
Instead of:
office → local team → marketing → users
the product can go first.
Usage creates the first foothold. Developers become contributors. Contributors become advocates. The community begins to exist before the company is physically there.
MemOS is already seeing that pattern across regions.
And yet Feiyu still sees Silicon Valley as an important place to invest more time and presence.
That is not a contradiction.
Product can create initial adoption remotely. But once the work becomes about partnerships, adjacent infrastructure companies, repeated conversations, and longer-term relationships, being in a dense ecosystem makes the next step faster.
Remote access removes distance. In-person presence can still compress time.
Silicon Valley still matters — for specific reasons.
Patricia’s perspective made that distinction especially concrete.
After more than a decade in Silicon Valley, following earlier years in Europe, her answer to whether founders still need to be there was essentially a strong maybe.
For financing, the Valley remains unusually important.
Capital is concentrated there. Leading investors are close to each other. Signals travel quickly.
And biotech adds another constraint entirely.
Labs matter. Scientific partners matter. Specialized infrastructure matters.
That is why the recent conversation around the “borderless founder” is useful only if we do not confuse borderless with placeless.
The barriers to participation are lower.
But some places remain unusually efficient nodes for particular things.
Silicon Valley is still one of them.
So the more useful question is not:
Do I need to be in Silicon Valley?
It is:
What do I need to be there for?
AI infrastructure is already moving into ordinary industries.
Another signal from the conversation was how quickly capabilities such as memory, context, and orchestration are moving beyond frontier AI companies.
Patricia pointed to biotech and healthcare, where data already needs to move across different databases and complex workflows. Feiyu shared a security-related smart-device example where memory becomes part of how the product works.
The point is less about any one use case.
It is that capabilities that recently sounded like “agent infrastructure” are already becoming practical building blocks inside established industries.
Eventually, people stop talking about the infrastructure.
They talk about what the product can now do.
A more pragmatic AI market.
Jev felt like another small but timely signal.
It is designed for narrower, structured decisions inside software rather than trying to be the most capable model for every task.
What interests me more than the model itself is what the attention around it suggests.
The AI market may be becoming more pragmatic.
Not every task needs a frontier model.
Cost matters. Latency matters. Reliability matters. Fit-for-purpose matters.
Sometimes the more sophisticated choice is not to use the most sophisticated model.
Don’t build on the hype.
One line from Patricia stayed with me:
“Don’t build on the hype.”
That may be one of the most useful founder reminders in the current AI cycle.
It is unusually easy to chase whatever is getting attention: a new model, a new interface, a new category.
But durable companies still tend to start somewhere much less fashionable:
What do you actually know?
What problem do you understand deeply?
What are you unusually good at?
What insight remains useful when the current trend moves on?
AI makes execution easier.
It does not give everyone the same domain knowledge, taste, or judgment.
Be there for a reason.
I went into the conversation asking how AI is changing where and how we build.
I left with a more practical conclusion.
A lot of company-building really is becoming more decentralized.
Early products can be built from more places.
Teams can stay smaller for longer.
AI can take on a meaningful amount of execution.
Products can find users before companies establish a local presence.
A lot can now happen earlier, with fewer people, and without depending on one geography from day one.
But there is still a layer that is much harder to decentralize:
the human layer.
Trust.
Relationships.
Repeated conversations.
Local context.
Being around long enough for the second and third conversation to happen.
And in the startup and venture world, Silicon Valley still holds an unusually dense concentration of that human layer.
That density continues to act like a magnet.
Companies can build elsewhere, find users elsewhere, and develop real momentum elsewhere — and still choose to spend more time in the Valley once capital, partnerships, talent, and relationships become more important.
That may be one of Silicon Valley’s most durable advantages.
Not that everything has to start there.
But that strong founders and companies still have specific reasons to come there.
That is increasingly how I think about it in my own work too:
Use AI for leverage.
Let the product travel.
Build where it makes sense.
But when the work depends on trust, relationships, and people, be there.
Not by default.
For a specific reason.
Hopefully, a year from now, we can come back to the same conversation and ask:
Which of the assumptions we made today are already no longer true?
Given how quickly the way we build is changing, I suspect there will be more than a few.