Howie Liu

Live shopping is a $100B+ GMV market in Asia

Howie Liu pitches AI-avatar live shopping by citing the scale of the existing market: live shopping (influencers peddling products on livestreams) transacts tens of billions, possibly $100B+ in GMV in Asia.

“I mean, you think about live shopping, which is a huge thing in Asia. I mean, there, there's like probably tens of billions, if not $100 billion plus, uh, GMV in commerce, uh, dollars that are getting transacted through live shopping.”

AI-avatar live shopping with one-on-one celebrity interactions

Howie pitches applying AI avatars (like HeyGen's) to live shopping so each shopper gets a personalized one-on-one interaction with a celebrity avatar, e.g. a Kim Kardashian avatar walking you through Skims products.

“In this case, like I think applying the avatar concept to live shopping where you can now have this one-on-one interaction with— and it could be a celebrity, like maybe you get some, um, uh, some like real influencers to sign on to this. And you know, it's like, uh, Kim Kardashian type who's actually like represented in avatar form, and each person shopping gets to have a one-on-one interaction with her about Skims and, you know, the Skims products that they want to try out.”

Steal thisLicense a celebrity likeness, build an avatar, and let every shopper get a personalized one-on-one product walkthrough.

Auto-generate personalized avatar thank-you videos from an Airtable row

Airtable wires HeyGen into their stack so they can automatically generate a personalized avatar video from each row of a table, e.g. a thank-you note that looks like the CEO for every event attendee.

“We actually have a setup internally where we can now generate an avatar video powered by HeyGen automatically from content in a row of Airtable. So it could be a list of different event attendees and that use case I just described, like I may actually do it for, um, thank you notes for different events that I host in the future.”

Steal thisPipe a contact list into HeyGen via API to auto-create a personalized avatar video for every person in the row.

An AI news aggregator you can talk to, not just a summarizer

Howie wants an AI news product that knows the topics he cares about, goes out and finds the relevant articles, and then becomes an interactive agent he can interrogate, like having a super-informed friend on call.

“I just want, like, an AI news aggregator that goes out and knows what I care about. And more importantly, it doesn't just summarize the news for me into like, here's the summarized articles. It actually becomes this interactive agent that I can talk to.”

Steal thisBuild news as a chat agent that fetches and explains, not a static summary feed.

White-label AI newsletter platform with injectable personalities

Howie extends the AI news idea into a Substack-like platform where creators spin up their own branded AI newsletters, and where the AI can be injected with personality (e.g. a TMZ-style gossip reporter) rather than dry AP-style summaries.

“Totally. And maybe like you could take it to the next level, which is like it actually becomes a platform where like Substack, you allow creators to actually create their own branded newsletters and they push the distribution promo of each one. But because you're powering this like kind of white label AI, you know, powered platform, like they can very easily spin up their own custom newsletter.”

Steal thisLet creators spin up branded AI newsletters on your white-label platform and have them push the distribution.

AI avatar wealth advisor giving everyone the high-net-worth treatment

Howie pitches an AI financial advisor (only possible post-o3-level reasoning) that builds a CFO-grade personal financial plan and delivers it through a human-like avatar you can call up anytime you feel stressed about money.

“So I think the idea here is using the really smart reasoning capabilities of the newest models. I don't think you could have done this pre-O3, where you actually can come up with a very thought-out, personal financial plan and, you know, recommend like, here's how much you should invest into equities versus bonds, and like, here's how much you should spend. But do it like with a, you know, an avatar on top of it so that it's a very human-like experience and you get to talk to them and just like call them up anytime you're feeling stressed about your financial situation.”

Steal thisWrap an o3-level planning model in a callable avatar so everyone gets a CFO-grade personal financial plan.

The AI roll-up playbook: buy traditional firms, automate, expand margins

Howie describes a roll-up playbook where you buy traditional low-margin businesses (like property management) and use your own AI to automate a large part of the cost structure, taking the firm from low-single-digit to tens-of-percent margins. He cites a German operator doing exactly this to become one of Europe's largest property managers.

“They're, they're buying up different property management firms and basically using their own AI tech to automate a large part of the job. And actually they don't fire people. They keep everybody employed, but they take on more business. As a result. And so they've quickly become one of the largest property management companies in Germany, if not in Europe, by just going and aggregating these traditional businesses and creating that AI leverage, taking what probably was like a low-percents margin business to become a, you know, tens-of-percents margin business.”

Steal thisBuy a labor-heavy traditional business outright, automate its cost structure with AI, and keep staff to take on more volume.

Sell to the top 1% power users who'll pay 10x for an hour saved

Howie's pricing thesis: there's a 1% power-user audience (VCs, lawyers, realtors) for whom email is their whole job. Make them even slightly more efficient and they'll pay $200 or even $2,000 a month, not $20, because the tool saves hours and prevents missed high-value opportunities.

“Yeah. And so actually one of the premises here is that there is a top 1% audience and not necessarily meaning like 1% richest people, but there's a 1% power user audience of email for whom email is their life, right? Probably for you, definitely for me, every VC, maybe even realtors, lawyers, et cetera. Email is everything to them. And if you can make them 10% more efficient at email, they'll pay not $20, but $200 a month, right? Maybe even like $2,000 a month.”

Steal thisTarget the 1% for whom the workflow IS the job and price for the hours saved, not the seat.

Win a 5% niche of a giant TAM to build a multi-billion business

Howie's bull case for Cursor: ~30M developers (Stripe's number) expanding to 50-100M, each rationally paying $1,000-$10,000/yr for a 2x productivity coding agent, implies a $30-300B TAM. Because coding is fragmented by language and stack, capturing even a 5% niche yields a multi-billion-revenue business.

“Okay, now you're talking about like a $30 to $300 billion TAM, like carving off even a small few percent niche of that, especially when like coding itself is so fragmented. You've got like different types of coding in different languages, front end, back end, etc. Like all you have to do is win like a sizable 5% niche and you've built like a multi-billion revenue business.”

Steal thisPick a TAM so big that winning a 5% niche of a fragmented market still makes you a multi-billion business.

Two startup paths: broad 'Cursor for X' or deep vertical PE roll-up

Howie's framework for choosing what to build: go broad ('Cursor for X') by picking a huge market with tens of millions of payers where capturing a niche is still huge; or go deep and narrow into a vertical with maybe 5,000 customers, extracting maximum value by buying the firms outright and running the PE playbook.

“I mean, I think there's like, you can either go really broad, like in this, you know, Cursor example or Cursor for X. So pick a market that's just so big, like there's so many potential target users, like tens of millions at least, and for whom the value is significant enough that they will pay for it.”

Steal thisDecide upfront: go broad on a tens-of-millions market, or go deep on a vertical and own the firms outright.

The AI-native consumer app will be built by artists, not technologists

Howie argues the first post-ChatGPT wave was driven by technologists close to the models, but the breakout AI-native consumer apps will come from people who are more artists than technologists, who understand consumer psychology with AI as a new paintbrush, just as nobody predicted Twitch.

“Like, an example would be like, what's the next consumer social app that's AI native, right? And I don't think that just means like, you know, it looks like Instagram, it looks like TikTok, and it's got some like AI features to help you like write posts and so on. Like, I think it might mean like, I mean, there are a few startups that are trying to do this where, you know, it's like you actually have like social networks or chat groups where half the participants are actually AI avatars, right?”

Shaan canceled the dinner that would have gotten him into Airtable early

Shaan recounts DMing Airtable's Howie Liu years ago when the company was valued under ~$100M, inviting him to a mastermind dinner. Sam and another guest canceled, so Shaan canceled the dinner and never met Howie — a missed shot at an early friendship and investment.

“And I was like, Howie, yo, love Airtable, think it's fucking great. I think it's gonna be big. You know, you're a fellow Duke guy. That's awesome. Hey, I host these mastermind dinners, you should come to one. And he's like, oh, I'd love to. And we scheduled it. And you were scheduled to go to it. And then I think you and one other person canceled and you were like, oh, I can't come. And so I canceled the dinner. And so I never met him. And like, you know, fuck, who knows what could have been. We could have been homies by now. We could have been invested a long time ago, you know, under $100 million.”