18VC Podcast - Next Gen Builders

Everyone talks to LLMs like they're people. Professor Robin Jia, who has been studying them since before ChatGPT existed, says that's the wrong picture: inside, an LLM works more like a democratic society, where parts of the model vote, disagree and sometimes fight over every word.On Episode 25 of 18VC, Philip and Corbin sit down with Robin Jia, computer science professor at USC Viterbi and head of the ALLEGRO Lab. They cover what's actually happening inside the black box, why copyright law may not save creators, the risks Anthropic flagged in its S-1, and what students should do about AI eating entry-level work.Highlights & Key Insights:- The "democratic society" model of an LLM: attention heads writing to a shared bulletin board and voting on the next word, and why the experts don't always win- How a hidden code on every New York Times page could prove a model trained on it, and why that still doesn't prove copyright infringement- Why a model will double down on an answer it can tell is wrong, and what that means for auditing AI- "They're really good actors": the case against AI consciousness, and the real-world harm of believing otherwise- Why the OpenAI–Anthropic race is "a recipe for disaster," and the case for an FAA-style watchdog for AI- Why he brought back oral exams in his machine learning class, and what he looks for in undergrad researchersTimestamps:(00:00) Highlight reel(00:42) Welcome Professor Robin Jia(03:53) From computational biology to NLP(08:38) Industry labs vs. academia(13:51) How the ALLEGRO Lab evolved(16:44) The thread tying it all together(18:57) What is an LLM, really?(22:46) Scaling laws: have we hit a wall?(27:38) Opening the black box(30:02) The LLM as a democratic society(36:02) AI in medicine & high-stakes fields(41:37) Copyright, memorization & watermarks(46:27) Can a model unlearn?(51:35) Anthropic's S-1 & catastrophic risk(55:25) Who keeps the AI labs in check?(59:41) Why models won't admit they're wrong(01:01:36) Can AI have feelings?(01:06:46) Overhyped vs. underrated research(01:09:34) ALLEGRO's next big problems(01:12:55) AGI, world models & embodiment(01:16:17) Advice for students in the AI era(01:21:44) Life outside the lab(01:23:42) Wrap-upSubscribe for the earliest signal on the founders worth watching.— Connect with Professor Robin Jia —Website: https://robinjia.github.io/LinkedIn: https://www.linkedin.com/in/robin-jia-a50b9623a/X: https://x.com/robinomial— ALLEGRO Lab, USC Viterbi —Website: https://allegro-lab.github.io/— Connect with 18VC —18VC Website: https://www.18-vc.com/18VC Podcast on Spotify: https://open.spotify.com/show/6b3YoAMLjZLziwqHWy3EXb18VC Podcast on YouTube: https://youtube.com/@18vcpodcast?si=l69vI6hHwLgIvw4N18VC Instagram: https://www.instagram.com/official18vc/18VC LinkedIn: https://www.linkedin.com/company/eighteenvc/18VC X: https://x.com/Official18VC

Katie VanArnam was diagnosed with lymphoma at eight. On the night of her first chemo session, her mom Casey Glazer cut open an old T-shirt and sewed Velcro over the port. Thirteen years later, that shirt is Courage Tees, and it just took first prize at USC's New Venture Seed Competition.On Episode 24 of 18VC, we sit down with Katie VanArnam and Casey Glazer, the daughter-and-mother team behind Courage Tees, about adaptive port-access shirts that let patients keep their clothes on, and keep feeling like themselves, during treatment.Highlights & Key Insights:Why the shirt switched from Velcro to plastic snaps, and the diagnostic-test rule that made the change necessary Fox Chase's gift shop already sold a port shirt that nobody bought. One patient's line explains why The "companion shirt," which turned classmates, families and even oncologists into customers Why they stayed for-profit: hospitals and organizations buying shirts for patients turned out to be the best distribution Katie didn't make the finals the first time she pitched. The mindset she borrowed from the previous winner won it the second time Building a company with your mom: who does what, the arguments, and why it has been "weirdly healing"Timestamps: (00:00) Highlights(00:57) Pre-show warm-up(02:47) Welcome: Katie & Casey of Courage Tees(03:56) Diagnosed at eight(06:43) Feeling exposed during treatment(08:17) The first Velcro shirt(09:15) From Snoopy shirts to a business(10:23) Companion shirts & the nurses(12:35) What "Courage Tees" means(13:58) Why start the company in 2025?(15:41) What a chest port actually is(17:13) Competitors, snaps & the design edge(19:05) From homemade to manufactured(20:54) Expanding to adults, babies & toddlers(22:58) Who's buying today(24:49) Nonprofit vs. for-profit(26:58) Michael's Way & the give-back model(28:17) Amazon knockoffs & the community moat(32:33) The five-year vision(33:25) Losing, then winning New Venture Seed(36:33) Journalism meets entrepreneurship(37:26) Co-founding with your mom(39:52) The next 12–24 months(41:12) Advice for young foundersSubscribe for the earliest signal on the founders worth watching.Subscribe for the earliest signal on the founders worth watching.— Connect with Katie VanArnam —LinkedIn: https://www.linkedin.com/in/katie-vanarnam-6012692b5/Instagram: https://www.instagram.com/katiee_van/— Connect with Casey Glazer —Instagram: https://www.instagram.com/caseyglazer1/— Courage Tees — Website: https://couragetees.com/Instagram: https://www.instagram.com/courage_tees/— Connect with 18VC 18VC Website: https://www.18-vc.com/18VC Podcast on Spotify: https://open.spotify.com/show/6b3YoAMLjZLziwqHWy3EXb18VC Podcast on YouTube: https://www.youtube.com/@18vcpodcast18VC Instagram: https://www.instagram.com/official18vc/18VC LinkedIn: https://www.linkedin.com/company/eighteenvc/18VC X: https://x.com/Official18VC

A kid learns hide-and-seek in one round. AI needed hundreds of millions of tries. Prof. Saty Raghavachary spent 16 years building the effects tools behind DreamWorks films, and he thinks that gap tells you exactly where AI is headed and where it isn't.On Episode 23 of 18VC, Corbin and Philip sit down with Prof. Saty Raghavachary, USC Viterbi computer science professor and DreamWorks R&D veteran, on what intelligence actually is, why agentic AI is really "LLMs in a loop," and why he'd bet on living neurons over bigger data centers.Highlights & Key Insights:- The two-word definition of intelligence he boiled down from decades of competing definitions- "If it's reversible, it's not real": the one line he says separates AI training from how a brain learns, and the moment he says it's time to get scared- Forget the Turing test: the kitchen test no robot can pass, and his four-letter framework for what machines are missing- Why world models look stunning right up until you try to crush a flower petal or snap a candy bar- Living neurons in a Petri dish that learned to play Pong, and why organoid intelligence is his pick for the real path toward AGI- His blunt advice for CS students: one-trick ponies are already obsolete, so go broad and build on things made of atoms, not bitsTimestamps:(00:00) Highlights(00:42) Welcome to Episode 23(02:36) Metallurgy → DreamWorks → USC(05:49) What academia doesn't teach(12:25) Intelligence = "considered response"(15:45) Agentic AI: LLMs in a loop(18:39) What AI still can't do(20:50) Organoid intelligence & the road to AGI(25:38) "If it's reversible, it's not real"(31:08) The coffee test(35:44) Why world models fall short(43:05) Overhyped vs. underhyped(49:40) Healthy FOMO: his information diet(56:20) Advice for students in the AI era(01:05:52) Go Trojans & staying joyfulSubscribe for the earliest signal on the founders, and the ideas, worth watching.— Connect with Prof. Saty Raghavachary —USC Viterbi Faculty Page: https://viterbi.usc.edu/directory/faculty/Raghavachary/SatyLinkedIn: https://www.linkedin.com/in/satychary/— Mentioned in This Episode —Cortical Labs (organoid intelligence): https://corticallabs.com/World Labs & AMI Labs (world models): https://www.worldlabs.ai/; https://amilabs.xyz/Lean (proof assistant)"Connections" by James Burke (book & BBC series)— Connect with 18VC —18VC Website: https://www.18-vc.com/18VC Podcast on Spotify: https://open.spotify.com/show/6b3YoAMLjZLziwqHWy3EXb18VC Podcast on YouTube: https://youtube.com/@18vcpodcast?si=l69vI6hHwLgIvw4N18VC Instagram: https://www.instagram.com/official18vc/18VC LinkedIn: https://www.linkedin.com/company/eighteenvc/18VC X: https://x.com/Official18VC

Every music startup is racing to bolt on AI. Kayla and Rila are betting on a pocket-sized synthesizer with none — a piece of hardware that gets total beginners playing their favorite song in about 15 minutes.On Episode 22 of 18VC, Philip and David sit down with the co-founders of Project Play — who also lead SparkSC, USC's entrepreneurship club — on why it had to be hardware, how two founders with no engineering background are getting prototypes built, and why they think the AI bubble in music is already popping.Highlights & Key Insights:- The 24-hour, all-women hardware hackathon Kayla almost walked out of — and the judges' comment that turned a "just for fun" build into a company- A professor told them that if they built it in a week, anyone could. Why four pitch competitions left them burnt out, and how building in public on LinkedIn flipped the script- Hardware with no electrical engineers on the founding team: a high-school friend's referral, a studio in China, and their first 10 prototypes in six weeks- The "Southwest Airlines of music tech" — $50–59 against $600–$1,000+ synths, 30–35% margins today, and the ~$20K tooling bill between them and a premium build- Why they ask for a $1 refundable deposit instead of a free waitlist- Getting rejected by Yamaha's student pitch competition, where every product was AI music — plus the "GG EZ" debate on whether AI can ever replace an artist's storyTimestamps:(00:00) Highlights & Demo(01:39) Welcome to Episode 22(03:57) How they found startups(13:16) The hackathon that started it all(16:55) Choosing a co-founder(20:35) Pitch-comp burnout → build in public(25:14) Why it has to be hardware(27:40) Music's barrier to entry(34:41) How Project Play works(38:42) Hardware without engineers(43:59) A music startup with no AI(47:39) Go-to-market & the $1 deposit(51:00) Margins & the "Southwest" play(55:01) Can AI make real art?(01:07:33) In their own words(01:10:34) Advice & who they're hiringSubscribe for the earliest signal on the founders worth watching.— Connect with Kayla —LinkedIn: https://www.linkedin.com/in/kayla-trinh-2309a6311/— Connect with Rila —LinkedIn: https://www.linkedin.com/in/rui-he-usc-marshall/— Project Play —Website & waitlist: https://projectplay.tech/LinkedIn: https://www.linkedin.com/company/project-play-tech/Founding engineer interest form: https://docs.google.com/forms/d/e/1FAIpQLSeDBQkh2oRfzDQOD-K66N5BEWBPML9lEbqYFnuHuRBcmEBAog/viewform— SparkSC —LinkedIn: https://www.linkedin.com/company/spark-sc/posts/?feedView=allInstagram: https://www.instagram.com/gosparksc/?hl=en— Connect with 18VC —18VC Website: https://www.18-vc.com/18VC Podcast on Spotify: https://open.spotify.com/show/6b3YoAMLjZLziwqHWy3EXb18VC Podcast on YouTube: https://youtube.com/@18vcpodcast?si=l69vI6hHwLgIvw4N18VC Instagram: https://www.instagram.com/official18vc/18VC LinkedIn: https://www.linkedin.com/company/eighteenvc/18VC X: https://x.com/Official18VC

More than 44% of a researcher's time isn't spent on research; it goes to chasing grants and paperwork. USC professor Mayank Kejriwal got frustrated enough to build GRAIL, an AI-agent platform that helps scientists find, write, and manage grants, with one goal: get that number down to 4%.On Episode 21 of 18VC, Philip and Corbin sit down with Mayank Kejriwal, Research Associate Professor at USC Viterbi, Principal Scientist at USC's Information Sciences Institute, and co-founder & CEO of GRAIL, on where LLMs really stand in 2026, what AI means for jobs, and what it takes to switch between professor mode and founder mode.Highlights & Key Insights:- Grants run 100–200 pages, win maybe 5–10% of the time, and take six months to hear back — and why he'd scrap the long written proposal altogether- Why the model is only ~10% of an AI product, where GRAIL's real moat comes from, and why Claude Sonnet beat newer models for academic writing- The customer-discovery finding that pushed GRAIL from selling to professors to selling to whole universities- His theory of why LLMs hallucinate — and why they're trained away from saying "I don't know"- The entry-level catch-22: if AI automates junior jobs, who becomes the next generation of domain experts?- Running two personas — humble researcher and all-in founder — sometimes within the same hourTimestamps:(00:00) Highlights(00:50) Meet Professor Mayank Kejriwal(03:23) Choosing research over Wall Street(04:51) Why a CS professor got a Marshall MBA(08:29) Knowledge graphs & fighting human trafficking(14:03) Where LLMs really stand in 2026(20:41) AI, jobs & the inevitable backlash(27:18) What students should focus on now(32:36) GRAIL's origin: the 44% problem(38:23) Picking models & building a moat(43:45) Why sell to universities, not professors(48:05) Building toward AI scientists(51:52) Will AI get cheaper or pricier?(54:56) Kill the 20-page grant proposal(58:43) Professor vs. founder: two personas(1:07:56) Who should (and shouldn't) found a startup(1:11:30) Wrap-upSubscribe for the earliest signal on the founders worth watching.— Connect with Mayank Kejriwal —LinkedIn: https://www.linkedin.com/in/mayankkejriwal/X: https://x.com/kejriwal_mayankSubstack: https://aiscientist.substack.com/— GRAIL —Website: https://enterprise.grailai.io/GRAIL at IJCAI 2026 (demo paper): https://www.ijcai.org/proceedings/2026/998— Connect with 18VC —18VC Website: https://www.18-vc.com/18VC Podcast on Spotify: https://open.spotify.com/show/6b3YoAMLjZLziwqHWy3EXb18VC Podcast on YouTube: https://youtube.com/@18vcpodcast?si=l69vI6hHwLgIvw4N18VC Instagram: https://www.instagram.com/official18vc/18VC LinkedIn: https://www.linkedin.com/company/eighteenvc/18VC X: https://x.com/Official18VC

A USC freshman is on his third startup, and this one wants your personal AI agent talking to everyone else's. No feed, no followers, no ads. Just agents making the introductions humans used to make themselves.On this episode of the 18VC podcast, hosts Philip and David sit down with Zion Zhang, founder of Humanest, an MCP network that lets personal AI agents talk to each other to engineer serendipity, broker warm intros, and collapse six degrees of separation into one conversation.Highlights & Key Insights:- Why his biggest competitive threat isn't LinkedIn. It's Anthropic or OpenAI shipping this exact network in six months, and the "model neutrality" bet he's making instead- The two hard rules baked into every Humanest agent: it never sends a message without your sign-off, and it treats anything from another agent as data and never an instruction- He built an AI wearable to capture conversations at his own events, then killed it — the one objection about it he could never engineer around- From Figbox's anti-loneliness thesis to Ponder's AI mentors to Humanest: what three pivots taught him about building only for yourself firstTimestamps:(00:00) Cold open(02:05) Welcome to 18VC & meet Zion Zhang(04:07) From Shanghai to a COVID-era Steve Jobs obsession(09:47) Startup #1: Ponder and the AI mentor that never shipped(13:45) Startup #2: Figbox and the anti-social-network pivot(18:13) What Humanest actually is: an MCP network for personal AI agents(23:28) The failed AI wearable — and finding the real moat(33:10) Why Humanest refuses to chase retention(41:08) Agent guardrails: no message sent without human approval(50:03) Why Anthropic or OpenAI — not LinkedIn — is the real threat(55:25) Inside the numbers: 30 daily users, 500 on the waitlist(01:16:41) The Lonely Chapter, being young and stupid, and when to pivotSubscribe for the earliest signal on the founders worth watching.— Connect with Zion Zhang —LinkedIn: https://www.linkedin.com/in/zionzhang/— Humanest —Website: https://humanest.ai— Connect with 18VC —18VC Website: https://www.18-vc.com/18VC Podcast on Spotify: https://open.spotify.com/show/6b3YoAMLjZLziwqHWy3EXb18VC Podcast on YouTube: https://youtube.com/@18vcpodcast?si=l69vI6hHwLgIvw4N18VC Instagram: https://www.instagram.com/18venturecapital/18VC LinkedIn: https://www.linkedin.com/company/eighteenvc/18VC X: https://x.com/Official18VC

Every mixer in SF ends by 6 PM, but the most committed builders are just getting started. Cheyoung Ahn and Bekhruz Malikov built the Night Owl Network for them: a late-night community that planned for 50 people and pulled 120+ signups in two weeks.On Episode 19 of 18VC, we sit down with the two Stanford co-founders on why showing up at midnight is itself a filter for commitment, how a lost third panelist turned into the event's best moment, and where a "special builder space" goes after one sold-out night.Highlights & Key Insights:The gap they found on Luma: every AI and founder mixer in SF ends by 6 PM, and why that leaves the hardest workers with nowhere to go"Sober night owls": why they built the event around deep 1 AM conversations instead of another partyThe two-week sprint from a text message to a 120-person event, including the vibe-coded website that scraped event data to pick the exact dateTheir third panelist dropped out, so Bekh asked the audience for a volunteer, live. It completed the investor–employee–founder triangle"Fund It or Flee It": the AI-generated pitch game invented 15 minutes before showtimeWhy anyone can fake a resume, a portfolio, or a GitHub history, but can't fake timestamped work (the thesis behind Cheyoung's commit.social, which just launched)Cold emailing as "the best way to change your life", the exact outreach that landed Anti Fund and Cresca as sponsorsIs it a club? A nonprofit? Neither and the Llama Lounge playbook they're chasing: 50 friends in a pizza shop to 800+ attendeesEvent #2 is 2–3 weeks away. One hint: StanfordTimestamps:(00:00) Highlight reel(00:36) Cold open & meet the night owls(02:20) What time do you actually sleep?(04:13) What is the Night Owl Network?(06:26) "Sober night owls" & deep conversations(07:30) Cheyoung: leading 200 cadets in Guam(09:43) Inside Springcamp: what VCs & founders actually need(12:23) Bekh: museum docent, auto mechanic, NAV Lab(15:01) How they each fell into the startup world(20:29) Zero doubt: the two-week sprint to event #1(22:38) Late night as a commitment filter(26:32) Splitting the work & the vibe-coded website(29:05) The surprise panelist story(31:58) "Fund It or Flee It" — the 15-minute game(33:28) Proudest takeaway: nobody wanted to leave(36:51) commit.social: verifying real human work(42:29) Bekh's work at Cephla: AI microscopes & robotics(45:32) Balancing internships, RA life & the network(47:36) The cold-email playbook behind the sponsors(49:37) Club, nonprofit, or something else?(50:38) How 50 invites became 120 signups(51:44) The one-year vision: an 800-person night(55:14) Event #2 teaser(56:31) Philanthropist & believer: beyond the builder titles(58:31) Advice for the 3 AM listenerSubscribe for the earliest signal on the founders worth watching.— Connect with Cheyoung Ahn & Bekhruz Malikov —Cheyoung's LinkedIn: https://www.linkedin.com/in/cheyoungahn444/Bekhruz's LinkedIn: https://www.linkedin.com/in/bekhruz-malikov-7ab3881a3/— Night Owl Network —Instagram: https://www.instagram.com/nightowlnetwork_/commit.social (Cheyoung's startup): https://commit.social— Connect with 18VC —18VC Website: https://www.18-vc.com/18VC Podcast on Spotify: https://open.spotify.com/show/6b3YoAMLjZLziwqHWy3EXb18VC Podcast on YouTube: https://youtube.com/@18vcpodcast?si=l69vI6hHwLgIvw4N18VC Instagram: https://www.instagram.com/18venturecapital/18VC LinkedIn: https://www.linkedin.com/company/eighteenvc/18VC X: https://x.com/Official18VC

He failed ten startups before turning 19. Six weeks into his eleventh, he has a $250K pipeline, a $1B client, and a prediction: software will be dead in two years.On Episode 18 of 18VC, we sit down with Yurii Tovarnytskyi, co-founder & CTO of Aximon, the AI operating system for commercial real estate that turns a broker email into a full underwriting and IC memo in five minutes, not 40 hours.Highlights & Key Insights:Ten failed startups by 19 and the edtech idea he got into Antler with, then killedHow he sold a product that didn't exist: cold-calling CRE executives, building the demo in one day, and the MVP in a weekThe zero-hallucination rule: AI extracts, it never creates a number, and every cell shows its sourcePretending to be DoorDash to get into real estate offices (yes, really)Why "until Series A, no one has a moat – your only moat is speed"The FDE future: hire a salesperson who knows the industry, teach them a little codeTimestamps:(00:00) Highlight reel(00:53) Meet Yurii & Aximon(03:20) Seven failed startups by 19(05:19) Why startups, not Google(07:47) The 5:30 AM routine(09:14) A co-founder who raced Marines(12:08) Aximon in plain English(13:05) 40-hour underwriting & $600K broker fees(19:33) Small teams, exploding workload(21:34) Broker email → IC memo in 5 minutes(24:56) Zero-hallucination underwriting(26:20) Selling trust (and the DoorDash stunt)(32:36) Data + relationships vs. general AI(38:24) "Software will be dead in two years"(41:48) Selling it before building it(43:51) Traction: $250K pipeline in 6 weeks(49:19) Deal cycles & decision makers(53:32) The FDE era: sales guys who code(1:00:15) Inside Antler & SH1P(1:03:25) The road to $1M ARR(1:04:24) No one has a moat – only speedSubscribe for the earliest signal on the founders worth watching.— Connect with Yurii Tovarnytskyi —LinkedIn: https://www.linkedin.com/in/yurii-tovarnytskyi/X: https://x.com/z1ro_devIns: https://www.instagram.com/z1ro.dev/— Aximon —Website: https://www.aximon.ai/— Connect with 18VC —18VC Website: https://www.18-vc.com/18VC Podcast on Spotify: https://open.spotify.com/show/6b3YoAMLjZLziwqHWy3EXb18VC Podcast on YouTube: https://youtube.com/@18vcpodcast?si=l69vI6hHwLgIvw4N18VC Instagram: https://www.instagram.com/18venturecapital/18VC LinkedIn: https://www.linkedin.com/company/eighteenvc/18VC X: https://x.com/Official18VC

Most investors chase Silicon Valley. Abdi Sherif went home. After returning 100x in 18 months managing money for friends and NBA athletes, raising a $5M fund out of Harvard Business School, and helping legal-AI startup Eudia land a $105M Series A, he now runs gBETA Inland Empire – a free, no-equity accelerator built on one belief: talent is evenly distributed, but opportunity is not.On this episode of 18VC, we sit down with Abdi Sherif, Program Manager of gBETA Inland Empire at gener8tor, on finding diamonds in the rough where VCs never look, the framework he uses to pick founders, and why coachability beats everything – even a killer product.Highlights & Key Insights:How teaching his friends (and future NBA players) to invest snowballed into a 100x return in 18 months, and a $5M fund raised as an MBA studentThe Eudia story: spotting a CEO who'd spun a product out for $500M, then helping the team break into DHL, Duracell, and Cargill on the way to a $105M Series AFree, no equity, no investment: the actual business model behind gBETA, and why cities and universities foot the billPOTC: the four-bucket framework he built at HBS to assess any startupThe one trait that gets founders rejectedHis first cohort raised $78K+ in non-dilutive capital in under three months, likely a gener8tor record The five startups pitching at the August 11 showcase: jet-ski sports infrastructure, AI for chambers of commerce, air-quality monitoring for vet clinics, and moreTimestamps:(00:00) Cold open: highlight reel(00:29) Meet Abdi Sherif & gBETA Inland Empire(04:38) From software engineer to a $5M fund(11:15) The Eudia story & coming home(18:34) gener8tor in plain English(21:29) Talent is everywhere, opportunity isn't(26:22) How a free accelerator makes money(30:45) The Inland Empire ecosystem(37:31) Bias to action: what separates the best(39:44) The POTC framework & coachability(46:42) Advice for student founders(53:23) Inside the seven weeks of gBETA(57:49) Founders who quit their jobs(1:04:31) Demo day August 11: the five startupsFounders: applications and office hours are open – reach out to Abdi directly. Investors: the gBETA Inland Empire Showcase is Tuesday, August 11, 5–7 PM at Moreno Valley College.Subscribe for the earliest signal on the founders worth watching.— Connect with Abdi Sherif —LinkedIn: https://www.linkedin.com/in/abdusamed-sherif/Email: [email protected]— gBETA Inland Empire (gener8tor) —Program page: https://www.gener8tor.com/gbeta/inland-empiregener8tor: https://www.gener8tor.com/gBETA Inland Empire Demo Day: https://www.eventbrite.com/e/gbeta-2026-inland-empire-showcase-tickets-1992960559683— Connect with 18VC —18VC Website: https://www.18-vc.com/18VC Podcast on Spotify: https://open.spotify.com/show/6b3YoAMLjZLziwqHWy3EXb18VC Podcast on YouTube: https://youtube.com/@18vcpodcast?si=l69vI6hHwLgIvw4N18VC Instagram: https://www.instagram.com/18venturecapital/18VC LinkedIn: https://www.linkedin.com/company/eighteenvc/18VC X: https://x.com/Official18VC

Gorest turns a napkin sketch into a playable, interactive 3D greybox in minutes, with zero AI-generated assets and full copyright ownership.On this episode of 18VC, we sit down with Heroza Zhang, Co-Founder & CEO of Gorest, published novelist, and 5th-place finisher at the World Xiangqi Championship, on why prompts will never replace human taste, and how a Cornell capstone became a company.Highlights & Key Insights:The insight from interviewing game devs: creators hate AI doing what they're good at – but happily hand over "the general rest"Why Gorest started with greyboxing, the one part of game dev nobody loves – and what a "grey fox" in Zootopia has to do with itThe 20-year analogy she uses on skeptics: C++ existed, but everyone still needed Unity. Foundation models exist, but you still need a new engineHow a high-school hackathon in 2020 (as an accidental "team leader") introduced her to VC before she could codeAre we all just NPCs? The philosophy of simulating the whole world that drives her founder journeyTimestamps:(00:00) Cold open & highlight reel(01:16) Meet Heroza & Gorest(02:03) "I want to understand everything"(05:42) Games as interactive worlds(08:56) The Cornell capstone that started it all(12:45) Novelist, chess master, engineer: which built Gorest?(16:51) The high-school hackathon that revealed VC(23:17) Gorest in one sentence: your AAA kingdom, from your mind(24:15) Why prosumers first(26:11) Greyboxing explained (via Zootopia)(29:20) Why not just a Unity plugin?(32:03) Sketch in, playable level out(35:12) No AI-generated assets — the copyright line(35:27) Where AI stops and human taste begins(40:11) Business model: B2B, subscriptions, the Unity playbook(41:14) Defensibility vs. foundation models(48:16) Competitors, Cursor-for-Blender & the blue seaSubscribe for the earliest signal on the founders worth watching.— Connect with Heroza Zhang —LinkedIn: https://www.linkedin.com/in/herozaw/X: https://x.com/HerozaZhan73432Ins: https://www.instagram.com/heroza_white/— Gorest —Website: https://gorest.ai/X: https://x.com/Gorest_AIIns: https://www.instagram.com/gorest.ai/— Connect with 18VC —18VC Website: https://www.18-vc.com/18VC Podcast on Spotify: https://open.spotify.com/show/6b3YoAMLjZLziwqHWy3EXb18VC Podcast on YouTube: https://youtube.com/@18vcpodcast?si=l69vI6hHwLgIvw4N18VC Instagram: https://www.instagram.com/18venturecapital/18VC LinkedIn: https://www.linkedin.com/company/eighteenvc/18VC X: https://x.com/Official18VC