After 18 Years and Bets on Uber and Shield AI, This VC Has Seen True Founder Chemistry Just 4-5 Times

Source: 20VC with Harry Stebbings | Published: 2026-08-08T14:00:12Z

With 18 years of seed investing and wins like Uber and Shield AI under his belt, David Frankel says genuine founder chemistry is vanishingly rare — he's witnessed it only four or five times in his entire career.

He's not talking about smooth collaboration. He means something harder to pin down: a CEO and CTO who don't need to finish each other's sentences, who can even disagree, but who trust each other's judgment at the critical moments and know the other is irreplaceable in their domain. Over eighteen years at Founder Collective — investments in Uber, PillPack, Shield AI, Suno, time spent with Jack and TJ, with Mikey and Georg — the word he reaches for is "alchemy." Then he pauses. "Maybe four or five times."

That's an honest number.

Every Team Meeting Starts With One Question

David opens team meetings with a prompt: "I love it because..." — then waits for people to finish the sentence. If you can't, you don't invest. It's not a ritual. It's a filter.

In a recent conversation with Harry Stebbings, he described what he looks for: every time you push back, the founder doesn't dodge, doesn't say "we're the only ones," always tells you the problem is harder than you think — and they're obsessed with it. He said, "I love it because they're obsessive."

This framework has cost him, too. PillPack founder Andrew came to him once, and the framework said no — mainly on valuation. He calls that a "terrible mistake." The framework protects them most of the time, but it has hurt them as well. In their entire history, they've never led a subsequent round, never tried to grow their position in a portfolio company beyond their initial equity stake.

The CEO's Only Real Job

A few years ago, David was invited to a lunch where Jeff Bezos was present. Someone asked Bezos where his time went. He said 50% went to finding people. David has never forgotten it.

Not long ago, he had coffee with Suno founder Mikey Shulman and asked what he'd been up to. Mikey said 30 to 40% of his time was going to recruiting.

Two founders. Different eras. Different scales. Same answer. David's observation: he's seen too many CEOs and CTOs diverge in their trajectories — before the company hits 50 people, a strong CTO is invaluable; past that inflection point, if the CTO hasn't grown with the company, they either get redeployed as a wildcard in a new direction, or they become the bottleneck. The CEO, meanwhile, has spent the whole journey learning to manage, to recruit, to persuade, to go head-to-head with reality again and again. "They're changed individuals."

The founders he backed — TJ Parker at PillPack, Jack at SeatGeek — both made that journey. He's held his SeatGeek position since 2010. Never sold a share.

His Favorite Bet: Someone Who Grew Up Inside a Vertical

"Nepo baby" usually means a well-connected second-generation kid. David uses the phrase entirely differently — he means founders who were immersed in a specific vertical from an early age.

TJ Parker was working in his father's pharmacy at fourteen or fifteen. When he started PillPack, his understanding of pharmacy operations ran deeper than anyone's. That kind of edge can't be purchased.

Suno's Martin Camacho spent years at Kensho before this. So did Mikey and Georg. They didn't pivot into music AI from somewhere else — they grew up in the space. At a public event, someone asked Martin: if a large language model could do what you do but better, would you swap out your own model? This is a Suno co-founder, someone who built the audio model from scratch. He didn't hesitate. He said he wouldn't think twice about it.

Then there's a founder named Evan, building HVAC quoting software — the kind mechanical engineers use to bid on construction projects. Evan went looking for an AI solution for his uncle's company, found nothing, and built it himself. He'd watched his uncle grind through this industry. He knew there were more than 100,000 mechanical engineers in the US, each one manually working through blueprint bids every day. David says founders like this have more edge than they even realize.

Seed Is Expensive Now, But David Still Thinks It Works

Harry Stebbings pushed on this: funds in the $50–100M range are the hardest to run — too big for the $100–250K friend-round check, too small to lead an $8–10M seed.

David didn't argue. He reframed it. Over the past 25 years, fewer than 100 companies have sustained market caps above $10 billion. Meanwhile, the median valuation of Fortune 500 companies in that same era was around $2.6 billion. If you own 5% of a company that ends up worth $260 million, you've already returned the fund. You don't need to pick OpenAI every time. The math still works.

But he acknowledges the real problem: uncapped notes everywhere, valuations grinding higher every round, and the hot AI companies that have raised enormous sums haven't actually proven their capital efficiency yet. His words: "the jury's still out."

His actual thesis: the real opportunity comes when others give up, when a company gets stranded by a big fund that's moved on. "Sometimes the best entry point is a seed extension round — when the original investors have already fallen out of love."

They occasionally take $500K–$1M stakes in larger rounds as an "insurance policy" — not for the investment logic, but because smart founders know: the partners at big funds leave, get poached, start their own funds. The people at Founder Collective are still there.

Why They Don't Want to Get Bigger

The GPs at Founder Collective are the fund's largest LPs. That's the real reason they won't scale up — they're chasing DPI, actual realized returns, not management fee income.

They made money in Uber, reportedly exiting a little early. They didn't follow on in Suno's Series A, didn't follow on in Shield AI's Series A — "by that stage, we're no longer that kind of investor. We're looking for the next seed." Every time the internal conversation turns to scaling up, it comes back to the same question: who actually wants to do this? And David's answer is always the same — he loves the early part. Finding someone when nobody else is paying attention.

That may not be the optimal return structure. He admits it himself: if he'd followed Uber all the way through, the absolute dollars would have been bigger. But the fund multiple might not have been. He describes it as closer to an addiction — finding that person is the reward in itself.

The SaaS Reckoning: He Thinks It's Overblown

SaaS valuations have cratered over the past two years. David thinks the concern is partly valid but mostly overdone.

His distinction is specific: if you're Olo, processing billions of restaurant transactions in real time every day, you can't be ripped out. If you're Veeva, embedded in biotech research, mission-critical means nearly irreplaceable. But if what you do can be replicated with a Claude prompt, you're in real trouble.

The more deeply a SaaS company is woven into core operations, the more the threat has been exaggerated. The more surface-level the product, the more real the danger.

He also ran a contrarian trade on himself: put his own money into Palantir, and calls himself a "momentum surfer" on the position — a complete contradiction, with full self-awareness. "Roy Driscoll bought a basket of beaten-down SaaS stocks and did well. I bought Palantir and made more."

He's Genuinely Worried About China

When the conversation turned to AI geopolitics, David said something direct: "I'm terrified about China right now."

He wrote the first check into Shield AI — back in 2016, with that name right there in plain sight, and real internal debate about whether to back a defense drone company at all. He's watched it grow from there, and he has a visceral sense of how AI is playing out in defense. His read: Chinese open-source models are already formidable. America's lead is not secure.

He cited a historical data point: in the 1820s, China accounted for 25% of global economic output — the largest economy of its era, before being displaced by Britain, then by the US. He said history doesn't stop turning.

He also brought up photonic computing — the next technology that might upend Nvidia. Every interconnect layer in today's data centers already runs on fiber optics. The chips themselves still use electrical signals. If photonic chips break through, energy consumption and performance will both undergo step-change improvements. He's not a thematic investor, but he said he's "deeply convinced the status quo always changes" — that applies to Nvidia, to OpenAI, to Anthropic. Disruption is a question of when, not if.

Impatience Is His Biggest Weakness

Harry asked a personal question: once you had money, what did you learn that most people don't know?

David's answer: he has less patience than he used to.

When you're accustomed to high-quality service, food, and execution, things that fall short become intolerable faster. As expectations rise, satisfaction becomes harder to reach. He has a formula: satisfaction equals perception minus expectation. Walk into a five-star hotel and your expectations are already set high; walk into a three-star hotel with low expectations and you're easily impressed. Now every dimension of life has been calibrated by apps to an impossibly high baseline — the flight is delayed, and Uber has already rerouted you to the train station within two seconds. He says this isn't something he's proud of.

Then he turned the logic toward other people. In his fifties, he cares increasingly about one thing: when you leave someone, how do they feel? Do they feel heard, or do they feel like you were there in body only? At dinner, no phones on the table. At night, no phone in the bedroom. Not because phones are distracting — he says he'd find something else to be distracted by anyway — but because it's a gesture. A way of saying: I'm here. I'm present.

What he says he's learning is simple: leave people well.