Sarah Guo

AI is the new dropshipping: distribution + a model in a niche

Sarah Guo frames the easiest AI money-maker as combining internet distribution with an increasingly powerful foundation model to do one useful thing for a niche, echoing the dropshipping wave where internet kids made their first $100k.

“if you can figure out internet distribution and then get, you know, super powerful models getting increasingly powerful to just do something useful in a niche, those two things together, that's like the new dropshipping. You know how like for maybe 7 or 8 years, I'm like too old to know what the exact timeline was, but there was a period of time where people were like, oh, you know, I'm an internet kid. I'm gonna figure out some dropshipping thing and like make my first $100,000. I think this is it.”

Steal thisPick one useful task, wrap a foundation model in a niche-specific template + SEO site + Stripe, and treat distribution as the real product.

One-person AI companion apps printing $1M cash flow

Guo says solo builders ship niche AI companionship apps (digital girlfriend/boyfriend) and generate around a million dollars of cash flow for themselves, not because they're deep AI people but because the underlying need is basic and universal.

“I have several friends who have shipped like AI companionship apps. Just look at, you know, paid apps in the App Store by Charting. And, you know, some of those people are generating a million dollars of cash flow for themselves. It's not because they're deep AI people.”

Steal thisShip an AI companion app for an underserved niche; the engagement and willingness-to-pay are already proven.

Watch for anomalous consumer behavior, not just metrics

Guo's consumer-investing tell: pay attention when a product's behavior pattern stands out sharply from its category. For AI companions the standout signal is that people spend hours a day with them, a level of engagement few products achieve.

“when something just really, really stands out from all other products in their category or previous categories, that's when you pay attention. It's like the, you know, dumbest metric, but it is really clear when something has special consumer behavior around. And the thing that is really interesting to me about Character or the companion apps that work really well is like people spend hours with them”

Steal thisHunt for products whose engagement is a clear outlier versus the category; that anomaly is the real signal of a winner.

HeyGen hit tens of millions in revenue with zero paid marketing

Guo, an investor in HeyGen, says its photorealistic video avatars are so novel that users create unbelievable content and share it organically, driving the company to tens of millions in revenue without ever spending a dollar on paid marketing.

“I am an investor in a company called HeyGen. You can make a video avatar of yourself. You cannot tell the difference. And, you know, reaching that bar of quality is new as of this past year. And like people create content that is unbelievable and they share it. And so like, now HeyGen is in tens of millions of revenue. Great. They've never spent a dollar on like paid marketing.”

Steal thisBuild a product whose output is so novel users can't help but share it; the wow factor becomes your free growth channel.

A Sims-meets-companion AI game world will get very big

Guo predicts that a product blending an AI companion with a game world, where characters have memory, goals, and realistic video and voice, will become a very large business.

“instead of it just being like, I'm talking to a person, it could be, you know, that person has some combination of memory of me, other interactions, goals, and like the media experience of them is richer. And we haven't gone there yet, but I think like there's a version of that company that's somewhere between like a companion and a game world that will be very big.”

Software 3.0: manipulate foundation models, don't train them

Guo defines Software 3.0 (after Karpathy's 1.0 hand-written code and 2.0 dataset labeling): the next generation of software is built by guiding pre-trained foundation models with reinforcement and business-specific info, so the value is in last-mile customization for large real-world niches.

“Software 3.0 is the idea that the next generation of software, a lot of it is about manipulating foundation models, and they're called foundation models because they have a lot of capability out of the box. You don't need to train them from scratch. You just need to give them like guidance, reinforcement, the information specific to your business.”

Guo flipped on healthcare and backed an admin-automation company

After years of dismissing healthcare as too slow and misaligned for startup speed, Guo changed her mind and invested in a healthcare administration automation company, reasoning that billing, authorization, coding, claims, and patient support are expensive, manual, and fertile for AI.

“And we just did a healthcare administration automation company. So I'm like, oops, like changed my mind. Real hypocrite here. And like one of the reasons being, I'm like, well, if you think about the mind-numbing work that happens in healthcare administration, like billing authorization, coding, claims processing”

Already-outsourced work is the best target for AI automation

Guo's idea-generation framework: look at what parts of a job have already been separated out and outsourced as a service (e.g. medical scribes sent offshore). If a task was already split off to outsource to cheaper humans, you can likely outsource it to a machine next.

“one framework for like your listeners, like thinking about different ideas is like what parts of work have been outsourced services already, right? Because like it used to be the doctor taking the notes and they were like, wow, we pay this person a lot.”

Steal thisFind tasks already carved out and outsourced to offshore labor; those are the ripest to automate with AI.

The US has spent ~$2 trillion on broadband to date

Responding to Sequoia's '$600 billion hole' AI CapEx warning, Guo contextualizes it against the broadband buildout: roughly $2 trillion spent on broadband to date, which she argues was worth it, suggesting AI CapEx will likewise pay off.

“like the broadband buildout, like, well, we wanted the internet, you know, like we spent about $2 trillion on broadband to date. Like, we're not there yet, right? That was worth it.”