Over more than a dozen Y Combinator batches we've refined a repeatable way to invest in the program before Demo Day. People often ask us to explain it end to end — not the pitch, the actual playbook. So here it is: the thesis, the process, and how our portfolio is changing with the latest batch.
The Thesis: Why YC, and Why Before Demo Day
YC produces unicorns at a rate the rest of venture cannot match — roughly 6.25% of companies per batch, and up to ~12% in the strongest batches, versus ~2.5% across broader early-stage venture. When a single program reliably mints outliers at multiples of the market rate, the question stops being "should we invest in YC" and becomes "how do we get the best access to it."
Our answer is to invest before Demo Day. By Demo Day the best rounds are visible to everyone and oversubscribed, which spikes valuations. We commit during the batch — typically weeks earlier — when the traction signals are already there but the price hasn't moved. We covered the full math of why this works in a separate piece on pre-Demo Day return math; the short version is that a disciplined earlier entry on the same companies still targets a venture-scale multiple, and the power law does the rest.
Two non-negotiables sit underneath the thesis. First, we do not invest in just ideas — every company we back has moved beyond concept to a real product with real customers. Second, our Demo Day positions are structured to meet QSBS criteria, which can qualify investors for up to 100% federal capital-gains exclusion on qualifying positions held five-plus years. (Consult your tax advisor — but it materially changes the after-tax math.)
The Process: How We Actually Pick
We meet roughly a hundred companies a batch and invest in about ten percent of them. That funnel runs on a tight, repeatable cadence tied to the YC calendar:
- Week 2 — Shortlist the batch and reach out to founders to introduce the fund.
- Week 4 — Research each company, its landscape, and its industry, and scout for breakthroughs at the weekly Launch Live demos. Our team is on the ground adjacent to YC in San Francisco, which gives us in-person, pre-Demo-Day access remote-only funds cannot replicate.
- Week 6 — Meet the founders and work to be the first check into the breakout companies.
- Week 8 — Invest in the top ~10% of the batch, per our research and thesis, at the Demo Day valuation or lower.
- Week 10 — Demo Day. By the time it arrives our capital is already committed. For founders, an Eight Capital check is an early quality signal and takes the fundraising pressure off their Demo Day.
Speed is the constraint that makes this hard, so we built our own software to beat it. Our team runs the funnel on a custom AI platform that auto-generates research notes, blurbs, and founder data, then monitors every portfolio company after we invest — tracking revenue, burn, runway, and growth so we can spot breakouts early and back them again. Most YC seed investors never track their portfolio at this granularity given the sheer volume; we treat monitoring as infrastructure, not a spreadsheet.
What We Look For in a Founder
Sectors rotate; founders are the constant. We screen every company against the same founder-market-fit questions:
- Deep domain expertise — founders who understand their vertical iterate faster and navigate the nuances on the path to product-market fit.
- Relentless grit — startups fail when founders quit; the ones who persist through setbacks are best positioned to win.
- Ability to raise — a founder should be able to raise at least $2M, enough runway for two real pivots before they run out of time to find a defensible business.
We also deliberately avoid consensus bets. Companies everyone tags as the obvious top pick rarely become the outlier successes — so we back conviction over crowd, and we spread that conviction across the batch rather than concentrating on the names already getting fought over.
How the Portfolio Is Changing: The W26 Shift
The thesis and the process are stable. What's evolving is the lens. Starting a few batches ago we began stress-testing every prospective investment against a single question: when far more capable AI arrives, does this company's value increase or evaporate? We call it the AGI-resiliency test, and it has visibly reshaped what we back.
The companies that pass tend to share a trait: their moat deepens as AI improves, rather than being a thin wrapper a better model erases. That pushes the portfolio toward defensible, often 'atoms-meets-bits' categories — AI agents and automation, the infrastructure those agents run on, and deep tech where domain depth and real-world systems are the barrier. You can see the shift clearly in our latest cohort:
- AI agents & automation — TesterArmy (vision-based QA that needs no scripts), Cignara (B2C support and sales agents trusted by Fortune 500s), Standout (an autonomous hiring marketplace), and Arzana (ERP automation that erases manual data entry for manufacturers).
- Infrastructure & compute — Project X (a browser-based GPU computing layer where every app runs on its own compute), Hub.xyz (training-data infrastructure with a 200K-contributor network), Arga Labs (digital-twin environments for testing AI agents), and Chronicle Labs (production simulation for enterprise agents, built by ex-NASA and Boeing engineers).
- AI-native verticals & hard problems — Prototyping.io (CAD-to-machine-code for autonomous manufacturing, already profitable and trusted by Tesla, Rivian, and Zipline), Astraea (clinical-trial data automation), and Manicule (an AI growth engine for developer tools).
A few years ago a YC seed portfolio was mostly pure software — the thinner the better. Today the most durable bets increasingly touch the physical world or sit deep in the stack, precisely because that's what a more capable model cannot trivially replace. The AGI-resiliency lens didn't change our discipline; it sharpened where that discipline points.
The Through-Line
Thesis, process, and portfolio evolution are the same idea expressed three ways. Get in before the crowd, on companies with real traction and founders built for the problem, run a fast and disciplined process to find them inside the batch — and keep adjusting the filter so the portfolio stays on the right side of where technology is heading. That last part is why the W26 portfolio looks different from the one we were building two years ago. The method is the constant. The frontier is the variable.
