Do You Really Have to Start This Damn Company? | A Conversation with Jason, Founder of Articuler.ai

Summary

• Articular AI has moved into offline event matching and is scaling: invitation emails see an ~80% open rate and 15%–20% conversion, the platform has just over 70,000 registered users (more than 40% in the Bay Area), and it co-hosted July’s biggest Silicon Valley event, AGI Summit, as well as a Codex developer event with OpenAI. The business solves two core problems for event organizers: attracting attendees and getting them to show up. A 20%–30% turnout is considered good for a free Silicon Valley event; after matching, one organizer found the venue they had booked based on expected attendance was too small. Repeat business followed.
• Jason’s moat is not just searching for people, but computing them. He turns people’s backgrounds and intent into high-dimensional vectors, then layers in post-training, dynamic multi-agent matching strategies, and expertise tailored to specific settings. The core asset is real interaction in which users “vote with their feet”: friend requests and responses are directional signals with varying strength, a data dimension missing from previous-generation databases. “The world doesn’t need another LinkedIn.”
• Jason sees Luma and Eventbrite as partners, not competitors: they are essentially ticketing or registration systems, earning a 5% ticketing take rate and $70–$80 a month in API membership fees, with no event-invitation service. He says Articular AI could open up its invitation capabilities to help Luma make money, while keeping people-matching on its own platform: “That’s our most critical asset.”
• The market foundation is 1.6 billion annual attendances at professional events worldwide, and “people’s intent carries through before and after an event.” The more radical idea is to use a distribution network to replace the meeting itself: once registration closes, match people directly based on their backgrounds and intent. “I don’t even want to do group chats.” Rather than burn cash building a broad, two-sided network, the plan is to use public data to help users find people on day one, then build an “atomic social network” within events.
• At the company’s darkest point, a U.S. investor’s term sheet was pulled over an identity issue (one of its LPs had ties to the U.S. government), team members left, and cash nearly ran out. Jason sold his own stock to fund the company, took no salary last year, and slept on friends’ couches. A co-founder told him, “You don’t have money; I do.” In hindsight, Jason thinks the company might have died if it had raised a few million dollars more to hire and spend on marketing.
• Jason’s blunt advice to founders in transition: if you can’t bring in 1,000 new users a week, growth is a problem. Find a user every hour and send them a message; AI automation tools can help keep up the outreach. “If you don’t even know where your customers are, I’d say don’t start this business—quit while you’re ahead.” Failure to find PMF may mean you’re dreaming up demand in isolation. “There’s nothing new under the sun”: understand what already exists, then make incremental improvements.
• On the industry, Jason jokes that 90% of AI application founders are building tools and many are pivoting (AI-powered PowerPoint is his example); he sees social matching as a rising tide that lifts all boats—the stronger the models, the better the product. He thinks many U.S. B2B applications are customized development under a new name: FDE. He also dislikes the term “going overseas”: “If you’re doing business in the U.S., why aren’t you an American entrepreneur?”

Deep dive

1. No money for ads, so handing out flyers on BART led to events

• The starting point was blunt: “Last year, there was genuinely no money in the account. If you wanted me to run ads, I simply couldn’t afford it.” Jason rode BART alone to university entrances and events in San Francisco and Palo Alto to hand out flyers. In the process, he saw that events are settings with clear intent: people are looking for peers, founders, investors, or customers. In San Francisco alone, there are thousands of events of all sizes every day.
• The core insight: “It’s no different from cavemen dancing”—people are in the same place at the same time but have no idea who they should meet. “Even if the most important person for you is sitting right next to you, you don’t know who they are.” People still rely on introductions or luck. Qukai said he sometimes wonders who the people around him at Blue Bottle are; Jason added that understanding someone’s background, intent, needs, and what they can offer makes it possible to connect people more efficiently.

2. Helping organizers bring in attendees: how the 80% open rate happens

• The process is simple: organizers provide an event topic and audience profile, and Articular AI searches a pool of 1 billion people worldwide for likely attendees and emails them invitations. The open rate is around 80%, with a 15%–20% conversion rate. Jason stresses that this is matching, not just finding people.
• The best-performing subject line came from an event hosted by an RSI scientist in Silicon Valley. The message was roughly: “Other researchers in your area are already coming—would you like to join? We’re discussing your recent research, and there’ll be plenty of intellectual cross-pollination.” Qukai questioned whether it sounded like marketing and might land in spam. Jason said the emails are AI-generated from start to finish and highly personalized; they are invitations, not sales pitches. The team has also put work into technical details like domains and DNS.
• The lesson on email length: “The shorter, the better.” People scan for keywords first and open the message only if something catches their interest; no one wants to read a wall of text.

3. Why organizers pay: acquisition, attendance, and access to the right circles

• The first two sources of value are attendee acquisition and turnout. Luma and Eventbrite are essentially ticketing or registration systems: they can take a 5% cut of ticket sales, while API customers pay membership fees of around $70–$80 a month. They do not offer event-invitation services. For a free Silicon Valley event, a 20%–30% turnout is considered good; matching gives people more reason to show up. Organizers often tell Jason they booked venues based on an expected 20% or 30% turnout, only to find that the room was too small. After one event, they often come back for a second and third.
• The third value is access to the right circles. Jason says not to romanticize the U.S.—everywhere has its own networks. When he was an investor, he spent half his time traveling in the U.S. and built trust at key nodes in those circles. Starting with people nearby and serving them well turns them into word-of-mouth channels.
• The biggest thing an event organizer can do is put a group of relevant people in the same place at the same time. Large events cannot cater to every attendee: some are looking for customers, others for investors or researchers. Unless someone knows everyone in the room and can introduce them one by one, doing that manually is unrealistic. AI can do it.

4. Luma can’t copy it—and doesn’t need to: the moat is “computing people”

• Jason’s stance on Luma: “We’re not competitors; we should work together.” Articular AI could open up its invitation capabilities to help Luma make money. Jason wants people-matching to happen on his own platform because it is the company’s most critical asset. When Qukai suggested Luma could build its own ticketing system, Jason said the event logistics SaaS stack—pricing, ticketing, check-in—is a vast system they cannot, and are not equipped to, replicate.
• Jason says matching is not simple to copy. Articular AI does not turn intent into database-friendly keyword queries and scrape results; it represents people as high-dimensional vectors containing their background and intent. “While others are still searching for people, we’re already computing them”（当别人还在搜索人的时候，我们已经在计算人了）. The system also combines model post-training, dynamic multi-agent matching strategies, and expertise and know-how added for different settings—a layered engineering effort.

5. The Chinese team’s unfair advantage: consumer-product detail and iteration

• Asked whether any YC team from Stanford could copy the product in a few weeks, Jason said founders from different backgrounds have different mandates. The main focus for U.S. founders today is still B2B cost reduction and efficiency. Over the past 20 years, Chinese teams have been better at consumer products, where the game is won on details, operations, and product iteration.
• Jason is candid about where his own strengths lie. He could go sell B2B in the U.S., but against an all-white team with 10 years of sales experience, he would not feel confident—and he does not think any Chinese team would. His edge is engineering, iteration speed, and years of operational experience.

6. From pilot to scale: already inside the top-tier event ecosystem

• The timeline: the need became clear late last year; the product was built out in the first half of this year; the team piloted at small scale in May, ramped up in June, worked on infrastructure in July and August, and started scaling this month. Wins include co-hosting Silicon Valley’s biggest event in July, AGI Summit; co-hosting an OpenAI Codex developer event; and participating in several YC Startup School after-parties. The platform now has dozens of events a week, with more than 10 events listed for October’s SF Tech Week. The goal for next year is more than 200 events a week.
• The platform has just over 70,000 registered users, mostly in major U.S. cities. More than 40% are in the Bay Area, with users also in Seattle, Boston, and New York. Expanding to another city does not require an in-person pitch for every event: Jason says online organizers can find them through their own product.
• A new feature, Signal, uses users’ backgrounds and current intent to surface recent, relevant signals from across the internet. Someone looking to connect with event organizers or community owners can see events coming up in the Bay Area next month, along with key contacts and their details. Salespeople can also track new developments from customers and assess whether they may have new budget. Signal began as an internal GTM dashboard and was later opened to all users. The current bottleneck is team bandwidth; automating the product has been a daily discussion lately.

7. No second LinkedIn: the asset is a vector of real interactions

• Jason says a business can essentially be understood as a database, but “the world doesn’t need another LinkedIn”（世界不需要第二个 LinkedIn）. Copying LinkedIn’s existing data would be foolish. The opportunity comes from different data dimensions and measures of value. For example, if the product recommends 10 researchers to Qukai and he sends friend requests to 8 of them, that reveals a preference. If someone responds and they connect, that is a stronger signal. These are choices users make in real settings—votes with their feet—and a kind of signal that previous-generation databases do not have.
• Jason once tried to get data on conversations between two people, but it was noisy. To use it, he would need the full dataset; scattered fragments can lead to bad conclusions. If you see only two messages and then a deal, for instance, you might wrongly conclude that the person is especially easy to close.
• Professional events reach 1.6 billion people annually worldwide, and “people’s intent carries through before and after an event.” A small private dinner around a table may not need the product. Once a group reaches around 20 people, Jason thinks matching becomes necessary.

8. The radical idea: stop holding conferences—and a different view from Body

• First principles: a conference brings a group of relevant people together around a topic, and what it delivers is person-to-person relationships. Jason wonders whether a more efficient distribution network could replace the conference. Bay Area founders could register, with sign-ups closing at 8 p.m. on Friday, and then be matched based on their backgrounds and intent. “I don’t even want to do group chats. We’ll deliver person to person, one-on-one.” Qukai suggested going further and choosing the topic based on the backgrounds and intent of the 20 people who sign up.
• Jason likes Body’s product and thinks its marketing and product design are creative. But he is cautious about Body’s plan to build a two-sided or multi-sided network among users on its own platform first: early on, user numbers and density are too low, and it takes a lot of money to get the network off the ground.
• Articular AI sees two paths: first, use public data so users can find and contact people who are not on the platform from day one; then build two-sided or even multi-sided networks in specific times and places, rather than trying to create a network for everything. The team calls the latter an “atomic social network”: start with one event and let people in the same place at the same time interact naturally.

9. Last year’s low point: a pulled term sheet, departures, and selling stock to stay alive

• Jason’s day-to-day emotions are a roller coaster. He wakes up at 7 a.m. thinking the company is about to go under; by 1 or 2 a.m., he is wired, convinced the company will soon IPO and is the best in the world.
• The financing blow: a reputable U.S. institution gave him a term sheet, then pulled it over an identity issue. One of its LPs had ties to the U.S. government, so it could not invest. Jason thought the company was about to make it, then suddenly found himself back at sea. In hindsight, he thinks a few million dollars more might have led him to hire people and burn cash on marketing—and the company might already be dead.
• Some team members left because they had not raised money and the business was still searching for direction. For a stretch in 2025, cash was close to running out. Jason sold his own stock to put money into the company, took no salary last year, and slept on the couches of a college roommate and friends. A co-founder told him, “We’re willing to work for you for free for two months. You don’t have money; I do.” To survive, the team took B2B orders helping Chinese companies expanding overseas find sales leads, including sellers of building materials, machinery, forklifts, and cranes.
• Re-reading the book about Elon Musk changed how Jason saw it. When he read it in high school while preparing for the SAT, he just thought Musk was impressive. Reading it again in Silicon Valley this year, Jason said he “really admired him and felt what he went through.” He paced downstairs at a friend’s home, repeatedly asking himself whether to give up, then decided this was the right thing to do and something he should pursue.

10. Signals that it works: one check and an annual membership

• One example is an Indian founder who graduated from Berkeley and wanted to find investors but did not know what firms existed in the U.S. He used the product to match with investors in March or April this year, got his first check, and has now just joined YC. He told Jason, “If I hadn’t gotten that first check, I would have given up on this product.” Jason was deeply moved.
• Another moment came at AGI Summit. A user looking for people interested in charities and willing to donate saw the first match and signed up for an annual membership. Jason says users are what turned things around this year: the user base is growing, and both payment and retention look good. When he is exhausted, he does user interviews and listens to what people found and what results they got.
• The long-term vision is that no one should have to spend time searching for the right people or opportunities. If people have something to offer, the platform should make their backgrounds and abilities visible so others can find and recognize them.
• Jason says the company ranks somewhere between third and fifth in its category on G2, where community discussion is one ranking factor. He is encouraged that users are not only using the product but also discussing it on Reddit and social media. Organizers who have worked with them recommend the platform to friends as well.

11. 90% of AI tool builders are pivoting; social matching rises with the tide

• Jason says 90% of people building AI applications are making tools. His example is AI-powered PowerPoint: he recently made his own fundraising deck with Codex and thought the result was very good, but there are surely far fewer people building AI presentation tools now. When models improve or a model provider enters a category, tools can be displaced quickly. He jokes that 90% of AI application builders are pivoting.
• Building a social product put Jason outside the mainstream narrative; things were rough, and the company did not raise much. But he believes this is a rising-tide business: the more capable the models, the better the product experience. The team is also studying a new model transcribed as “Jav” and plans to move its whole dynamic matching strategy over to it, expecting it to be more efficient.
• Jason says not to rely only on news translated from Chinese media to understand the U.S. market. Many U.S. B2B application companies are really doing customized development, now called FDE. Custom work often does not transfer well between projects, but some of it can now be productized or distilled into know-how. Some friends from his undergraduate years do this kind of work and have ARR in the tens of millions of dollars, with profits as well. Jason also says Chinese teams will fly to clients and sleep in their offices to deliver; with high U.S. labor costs, customers will pay if a product can replace one person.

12. Where the persistence comes from: an underdog’s urge to prove people wrong

• Jason’s answer: “I think I’m the one who should do this. If I don’t, the world will lose a great product, something beautiful.” He admits he is competitive: “If I can be first, I won’t settle for second. Are kings and generals born to their stations?” At last year’s lowest point, he felt a need to prove people wrong. Negative feedback from investors was subjective; they already had user feedback and revenue while others were still on a waiting list.
• The underdog story: Jason did not start preparing to study abroad until his second year of high school. At USC, many classmates had come from elite Chinese schools like the High School Affiliated to Renmin University and Beijing No. 4 High School, or from top U.S. high schools. Only after starting college did friends tell him they had published papers in high school; he did not even know what a paper was. He thinks underdogs should change things, and that persistence did not start with this company.
• Jason’s first round came from Qiju Ventures. The firm warned him not to treat entrepreneurship as a fashion trend or a cool label. Jason also dislikes treating entrepreneurship as a fashionable identity and then dismissing failure as simply having spent investors’ money. People give you money to do something you believe is right; every dollar, every day, and every opportunity should be valued.

13. Blunt advice for founders in trouble: keep a fallback, find 1,000 new users, and stop inventing demand

• On mindset, Jason says, “Everyone’s suffering is different.” He knows people who arrived in the U.S. with no money and went to events for the food, and others who slept in tents near Berkeley; he thinks the latter were braver. He does not think “startups are hard” is something worth celebrating. Finding it hard may mean you have not found a path, but you can also choose to meet the challenge. Keep some fallback: Elon Musk once said he had enough money for one more rocket launch, and still had a little money left after it exploded. “Staying alive is no disgrace. The disgrace is doing something empty and ending up destroying yourself.” He also cites Airbnb’s founders selling cereal when things were difficult.
• Jason disagrees with “we don’t have money for growth.” He tells his team that if they cannot bring in 1,000 new users a week, something is wrong. Find a user every hour of the working day and send them a message; use AI and automation to extend your reach. Social media and paid ads are also options. Early on, you do not need to start with advertising—the founder should be talking to customers. “If you don’t even know where your customers are, I’d say don’t start this business—quit while you’re ahead”（你要是连客户在哪儿都不知道，那我建议你这个业就别创了，趁早。）Founders need to be able to sell.
• On “not finding PMF,” Jason thinks many people build in isolation and imagine what a product should be. He once imagined a new need himself, but believes there are no new needs: “There’s nothing new under the sun.” First understand what has already happened, then make incremental improvements.
• Another lesson from the front line: “Money is a curse.” If the company had $20M on day one, it might have spent its time hiring or crafting a story for the next round instead of going into the field to observe and talk to users. “Being poor has its advantages and disadvantages.”

14. He dislikes “going overseas”: be an American entrepreneur from day one; the biggest fear is moving too slowly

• Jason strongly dislikes the term “going overseas.” If you want to do business in the U.S., he says, you should be thinking from day one about putting down roots and becoming an American company. When figuring out the direction, talk to users; do not just ask the Chinese community for advice. The U.S. is a country of immigrants. Being able to explain yourself and understand English is enough. His biggest concern is that people spend some time there and assume they will suddenly have an epiphany.
• His biggest fear is “moving too slowly.” Fast product validation early on does not mean the product is finished; from here, growth determines whether the company lives or dies. Jason thinks AI applications often fail not because the product is terrible, but because they cannot sell it. He does not really believe “ghost stories” about raising a $30M Series A without spending a penny on paid acquisition; acquiring customers always costs something. The real concern is the narrow window: how to put limited “bullets” to work as efficiently as possible, and how to find more ammunition.