What It Takes to Build a Startup | Andrew Chen & Matt Perault

1. Key Themes

"Little Tech" Is Literally Tiny — and Invisible to Policymakers

The core framing of the episode is that the earliest startups are radically smaller and more fragile than most policy or investment conversations assume. Andrew Chen repeatedly emphasizes that these are not "companies" in any formal sense at the moment they're funded. "This is truly little tech... I think the average team is two to three people. They are running their companies not in their office, not in a co-working space, they're running it at the kitchen table." 00:00:00 This matters because policy, regulation, and even venture strategy are often designed around an assumption of organizational capacity that doesn't exist yet.

Speedrun's Model: Get to Founders Before They Know They're Founders

Speedrun's differentiated strategy is investing before a company technically exists — sometimes before incorporation. "Of the 70 [companies in the last batch], there was probably a dozen where... they had full-time jobs and they had to turn in the laptops and their badges. And we couldn't even wire the money over because they didn't have a company to even incorporate." 00:08:09 The stated goal is not just capital deployment but literally inciting company formation that might otherwise have taken years to happen, or never happened at all: "so many folks that maybe would have waited another five or 10 years... maybe they would have taken that jump. And what an amazing time in the AI era to encourage them to jump in at the right moment." 00:08:57

AI Is Collapsing the Cost of the Technical Co-Founder

Chen describes a structural shift in how two-person startups divide labor, enabled by AI coding tools replacing what used to require outsourced engineering or junior hires. "The technology and product co-founder is typically using a lot of AI right now, a lot of AI coding. So rather than going out and needing to add a lot of cost and maybe outsourcing... you can just focus on... how effective can one or two people be right now? And like, let's just make this business stable and survive." 00:11:46

Regulatory Burden Is Cumulative, Not Incremental — and Invisible Until You Hit It

Matt Perault surfaces an internal a16z policy insight (credited to state policy lead Kevin) that regulatory analysis usually asks the wrong question. "Often policy teams, legal teams will think about just incremental burden... it's not just the incremental burden. It's all the things that a founder has to face from the moment they're building with the technology... it's all cumulative." 00:18:56 For a two-person team, this stacks of paperwork is often designed for a different scale of company entirely: "a lot of these laws, a lot of the paperwork is designed for companies that are much, much larger than them... and have the ability to comply because they have teams of lawyers." 00:20:38

Startup Geography Is a Choice, Made by Founders With No Loyalty to Place

Because these teams are two or three people with no fixed infrastructure, where they choose to incorporate and build becomes one of their few truly strategic early decisions, and it's highly mobile. "There is also an interesting thing where that almost becomes part of one of the most important choices that you can make as a startup, which is where are you going to plant roots?... the two guys, two gals startup, they're very mobile... they can pick and choose where they want to go, and they do." 00:25:48 Chen notes the historic migration pattern — Peninsula (2007) → San Francisco proper → now spreading to New York, London, and other hubs — as proof that geographic dominance isn't permanent.

Founders Have No Time or Capacity to Participate in Policy

A structural imbalance exists: the people most affected by regulation are the least able to show up and advocate for themselves. "For these founders, they are so mission focused... they just don't have time to participate... they don't have lobbyists. They're not really represented in all these ways because frankly, they just don't have time. They may not even have time to shower." 00:00:00 This creates a skewed policy input problem: "you get a very skewed view because you don't hear from little tech, but you might hear a lot from big tech." 00:23:57

Founder Failure Is a Feature of the System, Not Just a Cost

Chen frames the venture power law not as a tragedy to be minimized but as the expected mechanical output of the model, and describes concrete mechanisms a16z uses to keep failed founders in the ecosystem. "Historically... half the companies don't work out at all... then you'll have another two or three where you make a little money. And then all the money is made in that kind of top decile." 00:14:31 The response to failure includes re-funding founders directly: "we just funded a founder again on a better idea than his first idea, even though he ended up just returning a little bit of the capital." 00:16:38

Community and Ecosystem-Building via Tech Week

Beyond individual company investment, Chen describes Tech Week as a deliberate ecosystem-seeding tool across cities (SF, LA, NY, and new in Boston), including a mechanism for finding founders before they identify as founders. "If you're the type of person... finding yourself attending a bunch of startup events, you know, like the probability of you starting something... you're probably startup curious." 00:30:29

2. Contrarian Perspectives

Investor Advocacy Is Structurally the Wrong Messenger — But Necessary Anyway

Perault admits an uncomfortable dynamic: when a16z shows up to represent little tech's interests in policy debates, they are inherently the wrong voice, yet founders structurally cannot show up themselves. "We show up, we represent little tech, but we're not... I'm not a founder. And so when we show up, people are like, well, we would rather talk to the startup." 00:22:28 This is a quiet admission that venture-backed policy advocacy has a legitimacy problem that isn't easily solved — the "correct" advocates are, by definition, too busy surviving to advocate.

"All of Industry Agrees" Usually Means Big Tech, Not Startups

A specifically contrarian point: policymakers conflate industry consensus with big-company consensus, and a16z explicitly positions its disagreement as intentional, not incidental. "When they say all of industry, what they means is big tech companies... they'll say you have a different view and we need you to get on board... it's by design, we're representing a different part of the ecosystem." 00:24:23 This challenges the common Overton framing that tech speaks with one voice on regulation.

Regulation Often Functions as Incumbent Protection, Not Consumer Protection

Chen makes a pointed claim that startups experience most regulatory encounters as protective moats for incumbents rather than safeguards for the public: "generally their interaction on kind of the regulatory side of the world is usually negative because these are things that... actually create protections often for either big tech or for other parts of the industry that they end up trying to disrupt." 00:21:04

Silicon Valley's Dominance Is Not Guaranteed to Persist

Rather than treating the Bay Area's startup dominance as a permanent structural fact, Chen frames it as a fragile, policy-contingent outcome that could erode — a genuinely contrarian stance inside a firm headquartered there. "It's not certain that it'll last forever. It's not certain that California will always be the best place to start companies." 00:26:47 He specifically flags the proposed wealth tax as a mechanism that could relocate the investor base that makes the ecosystem work: "I worry about things like the wealth tax... its ability to potentially relocate a bunch of family offices and investors... as one of the potential negatives." 00:29:04

Nearly Half of Venture-Backed Founders Are First-Generation Immigrants — and the Rest Aren't Locals Either

Chen throws out a striking demographic claim suggesting the Bay Area's success is almost entirely a function of imported talent, not native population. "Nearly 50% of venture backed startups are first generation immigrants... if you were to say, okay, well the other 50%, how many of them are actually native San Franciscans versus people who move here? The answer would be... probably approaching a hundred percent." 00:26:47 The implication: Bay Area dominance is fundamentally about mobility and immigration policy, not inherent regional advantage — a fragile input that policy can disrupt.

3. Companies Identified

Slack — Team messaging platform. Cited as the canonical example of idea pivots in early-stage startups. "A product we use every day, Slack, originally started out as a browser based video game company." 00:02:52

Uber, Microsoft, OpenAI — Cited collectively as examples of major companies whose founders had prior ventures/false starts before their breakthrough success, supporting the thesis that founder failure and re-backing is a feature, not a bug, of the ecosystem. 00:16:13

Vercel — Mentioned as one of the notable companies whose founder has spoken to speedrun cohorts as a guest speaker. 00:12:14

Zynga — Mentioned as a company whose founder has spoken to speedrun batches. 00:12:14

4. People Identified

Andrew Chen — General Partner at a16z and lead of the Speedrun program. Identified as the architect and operator of a16z's earliest-stage investing motion, notable for personally building the infrastructure (welcome guides, co-working access, weekly programming, demo days) that takes founders from idea to funded company in 12 weeks. "My first couple of years at the firm, I spent more in series A, series B... For us, one of the really magical things... is the idea that we are not just waiting for entrepreneurs to show up at our doorstep... but we're actually helping create the companies in the first place." 00:07:10

Matt Perault — Host/interviewer, leads a16z's AI policy substack conversation; co-author of the "Greens from a Garage" piece examining California's cumulative regulatory burden on startups. Notable for surfacing the internal policy insight about cumulative vs. incremental regulatory burden. 00:18:26

Kevin (head of a16z's state policy team) — Credited by Perault as the source of the key insight that regulatory burden should be analyzed cumulatively rather than incrementally. "The insight that our head of our state team, Kevin had... is like, it's not just the incremental burden." 00:18:56

Mark Andreessen and Ben Horowitz — Mentioned as recurring guest speakers in Speedrun's weekly programming for founder cohorts. 00:12:40

5. Operating Insights

Structure the Cap Table Decision Around Trust, Not Headcount

Chen offers explicit guidance on team size at inception: cap founding teams at 3-5 people maximum specifically to preserve equity density and avoid premature payroll obligations. "You probably need to... some people's life situations will be such that you'll need to actually pay them... it's a two or three person thing, typically." 00:06:41 The operating principle: don't dilute ownership pre-product-market-fit; small trusted teams working for equity-only preserve optionality.

Split Roles Along an Inward/Outward Axis From Day One

Chen identifies a repeatable founding-team structure across the Speedrun portfolio: one "business" co-founder handling customer discovery and deals, one "technical/product" co-founder building with AI tools. "There's often this dichotomy that we see of kind of like the outwards facing business co-founder and then the inwards facing product and technology founder." 00:12:14 This is a template worth deliberately assigning rather than letting emerge organically.

Re-Fund Failed Founders on Their Next Idea Immediately

Rather than treating a failed first venture as a black mark, a16z's stated practice is to reinvest in the same founder on a new idea quickly, even when the prior investment only partially returned capital. "We just did actually just last week, we just funded a founder again on a better idea than his first idea, even though he ended up just returning a little bit of the capital." 00:16:38 The operating logic: you've already paid the "education cost" of teaching them how to start a company — capture the compounding return.

Use Community Events (Tech Week) as a Founder-Discovery Funnel Before People Self-Identify as Founders

Chen describes reverse-engineering founder discovery: rather than waiting for pitches, watch behavioral signals (repeated attendance at startup events) as a leading indicator of future founding activity. "If you are... finding yourself attending a bunch of startup events... the probability of you starting something... you're probably gonna get there sometime in the next few years." 00:30:29 This is a scalable, low-cost top-of-funnel mechanism other investors or corp-dev teams could replicate.

6. Overlooked Insights

Failed Founders Become an Acquisition Arbitrage for Portfolio Companies

Buried in the discussion of what happens to failed Speedrun companies is a quietly significant labor-market dynamic: a16z portfolio companies are absorbing failed founders as high-leverage employees, creating a feeder mechanism between failed startups and successful ones within the same ecosystem. "A lot of these founders become very hot commodities because that means that you can get somebody that works hard, works on unstructured problems, is completely current on all of the newest technology... a lot of teams will end up hiring each other... maybe they'll lead some new initiatives and new products in your business. And then a few years later, they spin out and they're ready to start a new company." 00:17:28 This describes an internal talent recycling system inside a16z's portfolio that effectively de-risks the "failure" outcome for the fund itself — failed founders don't leave the ecosystem, they get absorbed into other portfolio companies as high-agency operators, then re-emerge as founders again later, creating multiple shots on goal for the same underlying talent without needing new sourcing.

Robotics Founders Are a Structurally Different (and Growing) Cohort With Unrepresented Policy Needs

Chen mentions almost in passing that the composition of Speedrun's cohort has shifted meaningfully toward robotics, and that this cohort has entirely different infrastructure and policy needs than software founders — a signal that current policy frameworks (built around software/AI privacy debates) may be missing an emerging category entirely. "This year we are seeing way more robotics companies than in any year past. And they have very different needs compared to... they have supply chain needs, they have real estate needs, they have a different set of investors, they have potentially a different set of partners." 00:33:51 This is a specific, dated data point (2025 cohort composition shift) that signals where startup formation — and unaddressed regulatory/zoning/real estate policy gaps — is heading next, distinct from the AI-privacy-law conversation dominating the rest of the episode.