The 19-Year-Old Who Raised $6.2M on Zero Work Experience

Opening

Reader, here’s a résumé for you. Education: dropped out of high school in Kazakhstan. Work experience: none. Age: 19. Any ordinary recruiter would have screened this out at the paperwork stage. But the person behind this résumé, Arlan Rakhmetzhanov, is now running Nozomio, a San Francisco startup managing $6.2 million (about ₩8.5 billion) in funding. He started coding at 15, cold-messaged Y Combinator¹ alumni on LinkedIn, and landed his first angel check at 17.

So here’s the question: what exactly did investors see in a teenager with zero lines of work history, to hand over that kind of money? The short answer: the currency for verifying people has changed — from a résumé listing where you belonged, to a record of what you’ve built. Today I want to break down how this exchange happened, and what shape it’s taking as it arrives in Korea’s hiring market.

---

Seed Rounds for Dropouts

This isn’t just Rakhmetzhanov’s story. On July 31, TechCrunch published a piece looking into the world of founders under 20, and every profile in it follows the same pattern. Pranjali Awasthi dropped out of high school to start an AI company, enrolled at Georgia Tech, dropped out again, and built Slashy, which bills itself as “Cursor for email.” She’s 19 now, and recently revealed she’s quietly working on yet another company. Aidan Guo, 20, raised about $1.6 million for an AI desktop-assistant startup.

Awasthi’s own recollection is telling. At 14 or 15, meeting investors meant fielding the question “why do you even want to start a company?” first. Now that she’s past 18, that question has simply disappeared. The numbers point the same way. The median age of Y Combinator founders dropped from 30 in 2022 to 24 in the 2025 batch. One analysis found the number of accepted founders aged 18 to 22 jumped 110% in a single year.

What’s interesting is that this contradicts Y Combinator’s own long-standing philosophy. In a 2018 interview, founder Paul Graham named the late 20s as the ideal founding age and warned that founding too young amounted to “premature optimization.” Sure enough, the average participant age hovered around 29 from 2015 through 2022. Didn’t Silicon Valley always love young dropout founders, you ask? Yes — but with conditions attached. You needed a technical co-founder to pair with, or at least one line of Big Tech experience on your résumé.

What filled the gap left by those conditions? Ashley Smith, a partner at early-stage investor Vermilion, tells us. These days, screeners look at a founder’s GitHub activity, open-source contribution history, communities they’ve built and grown themselves, and fluency with the latest AI tools. Her reasoning: young developers learn software by contributing to open source and tinkering with new tools, and they simply have more time for that than someone with a full-time job and a mortgage. In fact, founders under 30 make up a meaningful share of Smith’s portfolio, and a few are under 21.

Here’s the summary: the résumé has come off the screening table, and an activity log has taken its place.

---

The Résumé Was Always a Proxy

Why did this swap become possible? We need to start by pinning down what a résumé actually is.

Résumés and pedigree were never direct evidence of ability — they’re proxy indicators. Economist Michael Spence’s signaling theory², formalized in 1973, explains why. A degree’s value lies less in the knowledge actually acquired than in the signal that its holder was capable of clearing that particular gate. The reason companies and investors leaned on proxies is simple: directly observing someone’s actual ability was too expensive. You can’t know until you put someone to work, so they borrowed someone else’s filter — Stanford’s, or Google’s.

But over the past few years, two of those costs collapsed at once.

First, the cost of building. Thanks to AI coding tools, a teenager can now build a product alone and put it in the market. That means you can manufacture your own real-world track record without ever setting foot inside a Big Tech office. Rakhmetzhanov’s Nozomio started exactly that way — as an AI coding agent that reads an entire codebase, and it’s now building an API index that helps AI agents discover and use outside software.

Second, the cost of observation. Everything you build now gets logged publicly — GitHub commit logs³, open-source contribution history, app metrics, community size. Screeners no longer need to borrow someone else’s filter; they can pull up the raw data themselves. When direct indicators get cheap, proxies lose their premium. The résumé’s demotion is a direct consequence.

None of this is actually new — the idea that open-source activity works as a job-market signal isn’t a fresh discovery. Economists Josh Lerner and Jean Tirole already identified career signaling as one of the core reasons programmers contribute to open source for free, in a 2002 paper. What’s changed is scale. What used to be a workaround for a handful of hackers is now the default screen for investment due diligence.

But this new currency has a dark side: as verification got faster, forgiveness disappeared. Smith says the market no longer carries the tolerance that early-stage startups used to get — the assumption that if you iterate enough, you’ll eventually find product-market fit. Even though that growth curve is an obvious outlier, she says, everyone’s out hunting for the next Cursor. Timothy Chen, an investor at Essence, makes a similar observation: startups used to worry about incumbent giants, but now they worry about the startup founded by someone their own age, one desk over. The flashy launch-video arms race that didn’t exist 3 years ago is proof of that. In Guo’s words — he’s 20 — the fear of failure is always there in the back of your mind, because everything can go wrong at once, and the moment it does, people pile on and tear you apart. He adds that this kind of hostile social ecosystem simply didn’t exist when Zuckerberg was building Facebook.

A generation discovered through its records is a generation judged by its records, continuously. When Rakhmetzhanov says “build a company as valuable as Google, or end up on the street,” that’s not teenage bravado — it’s a precise internalization of a market where the outlier has become the baseline.

---

A New Currency Arrives in the Land of Spec-Building

What about Korea? The grammar of hiring here has been moving in the same direction for a while now.

Start with the shift from large-scale, once- or twice-a-year mass hiring (gongchae, Korea’s traditional periodic corporate recruitment cycle) to year-round, job-specific hiring. Hyundai Motor scrapped its regular gongchae cycle in 2019, and LG and SK followed. Of Korea’s 4 biggest conglomerates, Samsung is now the only one still running it. Samsung is also the company credited with introducing the country’s first open recruitment system back in 1957 — meaning the company that opened gongchae’s door is the last one still holding it open. In a 2025 survey by the Korea Enterprises Federation (KEF), 70.8% of companies with 100+ employees said they hire only on a rolling basis, and 85.8% said they recruit whenever a need arises, with no fixed hiring season at all. A separate survey of Korea’s top 500 firms by the Federation of Korean Industries (FKI) found that rolling recruitment now accounts for 63.5% of hiring — up 5 percentage points in a single year — and that 28.1% of last year’s new college graduate hires were so-called “seasoned newcomers” (junggo sinip), meaning they already had some work experience.

There’s a reason gongchae’s decline connects to today’s story: gongchae was the institutional perfection of the proxy indicator. To filter tens of thousands of applicants at once, you needed standardized signals — school pedigree, GPA, language test scores. What replaced it, job-specific hiring, asks what you’ve actually done. This shift has moved fastest in development roles. Coding tests have replaced document screening, and it’s already standard practice for a GitHub link and portfolio to get opened before a cover letter does.

But there’s one more distinctly Korean scene here: when the signal changes, an industry springs up to sell that signal. Coding bootcamp ads promise a future where one GitHub link gets you hired, and job-seeker communities pass around tips for “planting grass” — padding your commit-history calendar with green squares. It’s the old spec-building toolkit for job seekers, just relabeled as a track-record toolkit. This is Goodhart’s Law⁴ in action: the moment an indicator becomes the target, it stops being a good indicator.

You can check the current score of this signaling arms race in the U.S. Cluely founder Roy Lee got suspended from Columbia University after building Interview Coder, a tool that secretly helps candidates pass technical developer interviews. That history became a talking point rather than a liability, and Cluely went on to raise a reported $20 million cumulatively, including a round led by Andreessen Horowitz. Investors armed with the new verification tools gave money to the person who built a tool to defeat them. Verification and forgery grow up together, each feeding the other.

---

Oswald’s Lens

I’m glad about this shift, and worried about it, at the same time. Let me unpack both in order.

The glad part comes from a data perspective. When I teach data courses, there’s a line I repeat to students often: a snapshot and a time series are entirely different kinds of evidence. A résumé is a snapshot — you can dress it up right before you submit it. A commit log, by contrast, is a time series. 3 years’ worth of consistency can’t be crammed the night before. What investors actually bought isn’t a website called GitHub — it’s a verification method built on time-series data that’s expensive to fake. I think the direction itself is right.

The worry comes from my consulting experience. Working on go-to-market strategy, I watched companies select talent up close, and organizations that only evaluate what’s measurable inevitably underprice capabilities that can’t be measured. Collaboration, judgment, and accountability don’t show up in a commit log. And as grass-planting shows, any observable signal eventually gets commodified. When that happens, what a screener will actually need to read isn’t the count of green squares, but the direction of the record — what problems someone solved, why, and in what order.

So here’s what I tell my students: don’t plant grass, plant questions. A single repository that reveals what you’re genuinely curious about beats 10 repositories padded with 100 commits each. And this isn’t only homework for developers. The proxy value of a company name is losing its power at an accelerating rate, regardless of profession. Building a time series of output that reads clearly even without an institutional affiliation attached — that’s how you exchange your currency in advance for the new one.

---

Closing

Let me sum up. Teenagers with zero work experience raising millions of dollars isn’t a feel-good anecdote — it’s a signaling-system replacement. As AI simultaneously collapses the cost of building and the cost of observing, the currency of verification is moving from proof of affiliation to a record of action. The decline of Korea’s gongchae system sits on the same map. But a generation selected by its record can never stop producing one, and the new signal is already being commodified.

Here’s one thing I’d suggest trying this week: write a 3-line introduction of yourself with your company name and title removed. If you can’t fill it in, what you need to build isn’t more credentials — it’s a record.

Which side are you on, Reader? Whether you’re a hiring manager who opened a candidate’s actual work before their résumé, or an applicant who got judged by what you built — tell me about that moment in the comments. Once enough stories come in, I’ll put together a Korea-focused rundown in a future issue.

---

💬 Tell me in the comments about a time you evaluated — or were evaluated by — “what you built” instead of a résumé. I’ll fold it into the next issue. 📨 If you have a colleague job-hunting or considering a career move, pass this along.

---

References & Further Reading

Primary sources

• TechCrunch, “Build in public, fail in public: what it’s like to be a founder under 20 right now”, 2026.7.31. Link ··· This is where today’s piece starts. It carries the voices of Rakhmetzhanov, Awasthi, and the investors, straight from source.
• Inc., “Y Combinator Bet It All on AI. Now Founders Are Wondering If They Should Bet on YC”, 2026.5. Link ··· The source for Y Combinator’s 2025-batch median age of 24 (down from 30 in 2022).
• Euclid Ventures, “No Country for Old Founders”, 2026.5. Link ··· A data-driven piece on Y Combinator’s pivot, from Paul Graham’s “premature optimization” warning to the 110% jump in accepted founders aged 18–22.
• Michael Spence, “Job Market Signaling”, The Quarterly Journal of Economics, 1973. ··· The original source for the idea that a degree is a signal, not knowledge itself. This research won the 2001 Nobel Prize in Economics.
• Josh Lerner & Jean Tirole, “Some Simple Economics of Open Source”, The Journal of Industrial Economics, 2002. ··· A paper that identified “career signaling” as a motive for open-source contribution — a 20-year-early preview of today’s story.

Background

• Korea Enterprises Federation, “2025 Survey on New Hiring Practices”, 2025.2. Link ··· The source for “70.8% hire only on a rolling basis.” Surveyed 500 companies with 100+ employees.
• Federation of Korean Industries, “H1 2025 Survey on Large-Company Hiring Plans”, 2025.3 / “H2 2025 Survey on Major Company Hiring Plans”, 2025.9. Link ··· The source for the 63.5% rolling-recruitment figure and the 28.1% “seasoned newcomer” figure.
• Seoul Economic Daily, “SK Hynix Declares ‘Degree-Blind Hiring’… Samsung Already Did It 31 Years Ago”, 2026.6. Link ··· Covers the current state of affairs, in which Samsung is the only one of the 4 major conglomerates still running regular gongchae.
• TechCrunch, “Cluely, a startup that helps ‘cheat on everything,’ raises $15M from a16z”, 2025.6. Link ··· Traces the growth of the signal-forgery industry, from Interview Coder to Cluely.

Past issues worth reading together

• A Perfect Score on the Math Olympiad, and No Proctor in the Room ··· A story about institutions verifying machines’ scores. Today’s piece is its mirror image — institutions verifying humans’ track records.
• Why Does the Rooftop-Room Developer (oktapbang, a small informal rooftop apartment common in Korea) Deliver Food Every Night? ··· A look at the same AI era from the opposite side. Worth reading alongside this piece, to see the gap between people whose track record becomes an asset, and people whose skills get priced by the hour.

---

📝 Glossary

The author, Kwangseob Ahn, is a professor of business administration at Sejong University and lead consultant at OBF (Oswarld Boutique Consulting Firm). He teaches statistics and data analysis, including business data management and business analytics, while leading GTM and AI strategy consulting in the field, designing the seam between technology and business. He has published academic research on a memory architecture for AI dialogue systems (HEMA) and runs Daily Arxiv, a daily curation of global AI papers. He holds a master's from Korea University's Graduate School of Technology Management and a KMBA. He is the author of Homo Brainless: The People Who Outsource Their Thinking.

Footnotes

1. Y Combinator: America’s leading startup accelerator, founded in 2005. Airbnb, Dropbox, and other major companies passed through its program, and acceptance alone functions as a powerful stamp of approval for early-stage startups. ↩

2. Signaling Theory: An economic theory describing how one party conveys otherwise-unobservable ability to another through an indirect signal. Degrees, certifications, and brand names are classic examples. ↩

3. Commit Log: The record left behind every time a developer edits and saves code. Stacked chronologically — what changed, when, and why — it functions as that person’s work diary. ↩

4. Goodhart’s Law: The principle that once a measure becomes a target, it ceases to be a good measure — because the behavior aimed at scoring well on the measure ends up undermining its original purpose. ↩