Two weeks in SF: Yes, AI labs are coming for everything, and here's what founders and investors are doing about it.
May 30, 2026
I just got back from two weeks in the SF bay area.
The first week I met with startups — portfolio companies like Gamma and Zenode, and a string of companies I've been talking with that are building agentic AI infrastructure on the frontiers. The second week, I was at Camp Hustle, my favorite investor event in the world, with 200-plus VCs, angels, and family offices.
Two different crowds, with the same shadow hanging over both: The big AI labs are coming for everything.
It came up at every turn. The evidence isn't subtle anymore. Anthropic and OpenAI started poaching senior consumer fintech and consumer healthcare operators about eighteen months ago. So yeah, they’re coming after those apps too, just quietly.
What’s not quiet: The labs have stood up joint ventures with mega funds — a billion dollars each — aimed straight at embedding AI products, APIs, and infrastructure into large enterprises. That's system-integrator work at lab scale. Your Deloittes and Accentures.
They aren't selling tokens anymore. They're going after the consumer surface and the enterprise surface, vertical by vertical.
For investors, why bother investing in anything other than Anthropic SPVs?
For startups, how do you even compete?
Founders and funds I met are all responding, each in their own way. Some responses are sharp. Some, frankly, are not.
Here's my read on the most notable ones.
How the startups are responding
Across the strongest companies I met, including portfolio companies and ones I'm interested in backing, the responses sorted into three distinct patterns. Each is sharp in its own way, but also hard in its own way.
Response 1 — Race to the edge
Get physically close to the customer, build a human relationship, and absorb the complexity of the AI stack on their behalf. Everything is moving too fast for the customer to keep up, so you become the layer between the frontier and their workflow. That embeds you in their machinery.
Grant Lee, co-founder and CEO of Gamma told me that even though revenue was still growing fast, they have to move faster. New offices will open in major cities. And a new role he's inventing: the Forward Deployed Designer. His framing:
"You have to be there, because customers don't always know what to do with the amazing features you've built. You have to be alongside them to figure out how to embed it into their workflows."
Another startup (whose round has gotten too hot for us to join and so I won't name them) ran an identical playbook in the financial and insurance world — engineers embedded inside their largest enterprise customer, building fifteen to twenty workflows side by side, shipping new features the moment the customer needs them. There is a product, but that's just a conversation starter, a sales kit almost.
Likely 70% of the features will be built alongside customers in the next 24 months.
I heard Deb Liu speak at Camp Hustle. She's a superstar in the tech world, now cofounder of Ember AI which is embedding AI into businesses.
"No one knows what will happen, or what the paradigm will be", she said, "so you get close to customers and really understand what they're trying to do, and build it for them, really quickly, sometimes in an afternoon, and the wow is what sells."
This isn't services-on-top-of-software. It's a structural recognition that at this historical moment, both euphoria and confusion are at a maximum. When everything is changing, and changing fast, trusted relationships are what endure.
Five years ago, you could swap consultants but Microsoft and Salesforce were etched in stone. Now? Any software might be swapped out, but the trusted human who settles you down, picks the model mix, and architects the rest of your stack — your AI guy, call him Steve — is who you keep.
When even Anthropic and OpenAI are betting on this, you see where the future is headed. And startups are betting - correctly, I believe - that the big AI labs won't cover even a majority of the customer surface.
Response 2 — Reinvent the medium
A completely different move, hard in a totally different way. Instead of getting close to the customer's current workflow, you reset the workflow itself — by reinventing the underlying medium people work in.
Gamma did it once already, by totally reimagining what a presentation looked like: HTML generation bridged into an editable Tiptap surface, so AI output stays editable by humans yet still maintained a sense of alignment and spacing. Then Google and NotebookLM went the opposite way: fixed, image-based slides that were beautiful yet totally non-editable. Both are new takes on an old medium, and both gathered massive usage.
Someone will do it again, and Gamma is betting that it will, a second time.
There are other great examples. HeyGen open-sourced Hyperframes, which renders HTML compositions into video — video as a derivative of HTML, not its own primitive, so any agent that can already write HTML can now produce video. Blew my mind!
While I was in conversation with investors in Los Gatos, talking about markdown files, the Claude Code team tweeted that HTML was the new .md file, and we were like "okay what's happening now!?"... they said you should stop emitting walls of markdown and output interactive HTML instead. Even file primitives are being re-imagined on the fly, on a monthly basis.
The pattern is the same:
don't compete inside the existing medium. Reinvent the medium and force everyone else to play your game.
This is hard because you have to be right and early.
If you're wrong about the substrate, you die there. If you're right, your moat persists for a while, because it takes everyone a minute to even understand what they're looking at, and most incumbents think you're an idiot even when you're not.
Response 3 — Buy the unhyped, where AI is invisible from the back
Don't fight in the AI arena at all. Go after businesses that aren't part of the hype but are better than ever.
Take one of my angel portfolios, for instance. A tech-enabled primary-care play I invested in two years ago, with low single digit millions in revenue, and barely profitable. Now? Tens of millions in revenue, and higher pure profit than the revenue level at my time of investment.
I had lunch with the CEO, and he said
"absolutely no AI, not even the mention of it", at least at the clinic level. Nurses and admin staff hear "AI efficiency" as "you're about to be replaced."
so the company's explicit stance is "We're not going to replace you, and we're not going to push AI on you at all."
The AI lives at headquarters, quietly — new digital product lines layered on top of the physical service network, but all handled by HQ, never imposed as a sales quota on clinic staff.
Investors look that way too: Our friend Eric Bahn at Hustle Fund wrote something similar to his LPs: So much capital is concentrating at the top, into generative-AI-native companies, that plenty of genuinely good businesses — now better than ever because AI lifts their back office — are raising at lower valuations, because the attention is elsewhere. The unhyped businesses are the asymmetric trade.
How investors are responding
The second week put the mirror up. 200+ VCs, angels, family offices, and operators in the same room. The labs are forcing every investor to rethink their investment approaches, and frankly, the responses are even more varied than the startups.
Trade 1 — Price discipline outside the gravity center
Cap the universe of deals around $5–10M post. Hunt for early traction where multiples haven't been pulled up — Southeast US, Canada, parts of Europe, Japan. Lean on personal networks to source.
But cheap alone isn't a strategy; it has to come with a clear articulation of which vertical, geography, or asset class is undervalued and why you are the manager who can reach it. To be honest, most "bargain-hunters" I met couldn't clearly articulate a thesis for finding and getting bargains.
Ranchos Ventures was one exception. It doesn't do AI startups. It does franchises, and especially consumer brands that are scalable - think Crumbl cookies, which surpassed US$1 billion in sales a year 🤯. It's led by a GP who comes from the franchise world, and who has a whole playbook to pick brands and help them become franchises.
If you wanna stay inside the AI world, though, you gotta play — one manager told me about a hellishly expensive pre-seed she broke discipline for, now growing like crazy: "Sometimes you see an amazing builder and you just have to go for it."
You can't be price-disciplined inside the AI gravity center anymore. You have to physically move to where the gravity is weaker. But few folks I met knew where to move.
Trade 2 — Concentrated secondaries on the bellwethers
The polar opposite. Pay full freight to be on the cap table of Anthropic, OpenAI, and the foundation-model adjacents. Concede the valuation argument entirely and bet that being on the list is the only alpha that matters this cycle.
One fund I met has its top five positions at ~70% of the fund, with the math riding on those names doing 5x at gross. All of them hot names that you see in secondary SPVs.
This is the labs-are-coming-for-everything trade run at the LP layer: if you believe the labs compound to multiples of where they are, you concentrate there and accept that the alpha is access, not stock-picking.
Trade 3 — The generational shape change
The one I think most interesting, and also the most puzzling to the veteran investors, because it's tied to generational difference, and to fuzzy things like vibes and culture.
The most familiar term I'd use for it is GTM, but it's actually more than that.
Yes, the thing that investors like to press founders on, GTM, is now indispensable for investors as well.
The funds doing this best are relentlessly focused on distribution, leveraging a combination of content and community, plus splashes of flair that sometimes border on the bizarre.
The Pitch, for instance, is this "Shark Tank but real" program that I'd been listening to since its very first episode, back in 2017 I think? And I met creator Josh Muccio at Camp Hustle and totally fanboyed like hell. They poured their heart and sweat into the show for years before they raised a fund, and no surprise, fundraising from LPs wasn't much of a problem for them, because they fanboyed and fangirled like me 🤣.
Another manager who built a >100-member operator-angel community now sources ~60% of her deals through it. Operators in big tech, spotting brilliant colleagues who are leaving to found companies, and getting their first checks in. They already have a 50x markup, and yep, that one came from the community as well.
(The secret's out though, my cousin who works in big tech says her VC friends are constantly pressing her on who's leaving to found companies.)
Another fund I met, the Honors Fund, is a single LP (large family office) fund that tapped a popular entrepreneurship professor to connect with students — as a charismatic teacher and trusted mentor, this offers a way into student communities.
What AI cannot collapse
These strategies also pointed at a few things that AI cannot collapse. Yes, AI has completely collapsed the cost of building software, but some very valuable things remain out of reach (so far).
One is hardware. Yes, most investors I met wanted to do hardware, and deep tech. But truthfully I met very few who knew how to.
Second is attention. The Pitch show spent years building attention. So did Hustle Fund. Certain startup founders, for better or worse, have mastered the art of getting attention. Until AI proves that it can conjure viral videos and durable influencers, all artificial and autonomous, humans remain the masters of attention.
Third is community. And friends from SF to NYC to Germany are telling me that IRL is back in a massive way. Hiking groups. Alcohol-free raves. Poker nights. Private dinners. Hell, a week after Camp Hustle, someone in the Whatsapp group posted a Luma invite for a hard-tech hangout in a Costco parking lot. The last one drew about a hundred people. Ridiculous, and absolutely marvelous.
Showing up IRL and being a human being still requires... a human.
It's not that the older generation can't run this shape. It's that the shape looks unserious to a lot of them. A podcast looks like a side project. A community looks like marketing. A parking lot looks like nothing at all. That's the mismatch — and it's exactly why the trade is underweighted.
Focus on what's hard, and what's human
I came home from these two weeks more convinced of something I'd only half-articulated before.
Every response that's working — racing to the edge, reinventing the medium, buying the unhyped, building the funnel before the fund — wins for the same reason: it's hard. And hard is the one thing the models haven't learned to collapse.
The work that survives this cycle — for companies, for funds, and for people — is the work a model can't write. Being at the customer's edge. Reading the shape of their work. Cultivating relationships and communities. Knowing which unhyped thing is quietly compounding while everyone stares at the top of the market.
And all of it is hard, and very much human. Taking customers through workshops, sorting through workflows to discern what they really want, having your solutions rejected — I've been there, and it's hard as hell. Rethinking what a medium means, be it presentations, videos, or email, takes years of struggle and dead ends. I've tried and failed. Building audiences and communities takes years too. Hard, and human.
But that's the way it is, because models are collapsing the cost of everything that's easy, and increasingly things that used to be hard, too.
So you must go where the going is still really, really tough. Where there's struggle, pain, and dead ends. Like a founder who goes and works in an elderly care home for a year in order to learn what using IT goods really feels like in that setting. Or a CEO who spends months going door to door, meeting customers one by one.
For anyone who's willing to go there, the opportunities are everywhere.