Y CombinatorHow Founders Find Ideas That Nine in Ten People Reject
Flock Safety, Coinbase, and DoorDash were all seen as bad ideas; contrarian bets on overlooked verticals consistently beat derivative plays that gain traction.
CHAPTERS
- 0:00 – 1:15
Avoid hot markets: why unpopular ideas are often the only survivable ones
Garry frames the core thesis: chasing what’s trendy leads to derivative products and crowded markets where most entrants die. The right approach is to find something people desperately need, even if it sounds risky or “crazy” at first.
- •Hot ideas create 5–100 competitors; only a couple survive
- •Being contrarian is often necessary to escape commodity competition
- •Unpopular ideas can feel scary because failure costs years
- •Focus on urgent human needs as the anchor, then figure out the rest
- 1:15 – 3:35
AI’s early gold-rush is ending: verticals are crowded and the idea space isn’t expanding as fast
Harj and Jared compare AI to past platform shifts (internet, smartphone): a brief window where obvious ideas are plentiful, followed by a period where founders must dig for deeper secrets. They argue the AI landscape is moving into the ‘picked over’ phase, making unique insights more important.
- •New platforms create a ~two-year surge of obvious startup ideas
- •AI vertical agent startups have done well, but many verticals now have multiple competitors
- •Fewer recent model step-changes means less new greenfield
- •Standing out increasingly requires a real contrarian bet and unique insight
- 3:35 – 6:02
What “non-obvious” really feels like: dangerous, reputationally costly, and easy to talk yourself out of
The group distinguishes ‘non-obvious’ from merely ‘not widely known’: it often feels personally risky and socially discouraged. Garry shares an office-hours example where an “AI makes it possible now” wedge is obscured by narrative baggage like “tarpit idea.”
- •Non-obvious ideas often trigger fear: ‘I could waste 10 years’
- •Media/social consensus can distort founders’ decision-making
- •Past failures in a space can be a signal of opportunity if capabilities changed
- •Early PMF signals can be ignored due to external mental models (friends, TechCrunch, X)
- 6:02 – 9:10
Case study: DoorDash, Lyft, and the hidden unlock—smartphones created a new on-demand labor model
They revisit smartphone-era winners that weren’t predicted at the time (Uber, DoorDash, Instacart). Lyft’s pivot from long-distance ridesharing (Zimride) to short-haul, phone-coordinated rides illustrates how platform shifts enable entirely new behavior and markets.
- •Big winners are often second-order consequences of a platform shift
- •Zimride/RideJoy show the pre-smartphone ‘email coordination’ limitation
- •Smartphone penetration enabled daily, short-haul coordination and a mobile workforce
- •Crowded categories can still produce massive outcomes when the unlock is real
- 9:10 – 10:41
Regulatory gray areas: when users win big enough, the rules can change
The conversation turns to a recurring contrarian pattern: products that start in regulatory ambiguity because laws lag technology. They emphasize this is not advice to break laws, but to notice outdated frameworks and understand why society might update them once benefits are clear.
- •Founders feared jail (Lyft/Uber) due to unclear/hostile rules
- •Many great ideas live in legal ambiguity rather than clear illegality
- •First-principles view: some laws were built for a pre-smartphone world
- •User benefit and safety/accountability improvements can drive legal adaptation
- 10:41 – 16:11
Coinbase’s contrarian bet: compliance and banking partnerships when the early market wanted the opposite
Coinbase is presented as contrarian not by rejecting regulation, but by embracing it early. Brian Armstrong’s thesis required extra work (KYC/AML, banking partners) that made the product worse for early cypherpunk users, betting instead on mainstream adoption later.
- •Early crypto culture valued anonymity and anti-state ideology
- •Coinbase pursued the opposite: mainstream access via banks and regulators
- •KYC/AML increased friction and angered the initial market
- •Contrarian bets can be about distribution and legitimacy, not just product
- 16:11 – 18:46
A live example: open banking, regulatory capture, and why democracy can still open markets
Garry connects the regulatory theme to open banking and data access (e.g., Plaid). He argues incumbents often invoke “safety” to justify barriers, but over time policy can shift if enough users benefit and push for change.
- •Open banking debates: who owns/controls access to consumer financial data
- •Incumbents may use fees/ToS as moats (regulatory capture)
- •Safety arguments can mask switching-cost preservation
- •Over time, scaled user benefit can drive legislative/regulatory updates
- 18:46 – 20:02
Framework for contrarian ideas: flip the default startup playbooks
Harj proposes a practical method: identify emerging consensus playbooks in AI startups and consider taking the other side when the playbook becomes dogma. DoorDash is used as an earlier example of rejecting the prevailing ‘full-stack’ meme.
- •Look for ‘default’ playbooks that may have become outdated
- •DoorDash vs SpoonRocket/Sprig: avoided full-stack ghost-kitchen model
- •Contrarianism can be choosing simplicity over fashionable complexity
- •Ask what today’s AI startup memes are—and whether they’re wrong now
- 20:02 – 21:20
Compound startups and AI-native suites: when building the whole product beats point solutions (Campfire vs NetSuite)
Diana highlights a contrarian move for certain markets: build an integrated suite instead of a narrow SaaS point solution, even if it seems too heavy for an early-stage team. Campfire is cited as an AI-native CFO platform competing with NetSuite by making ‘suite replacement’ feasible sooner.
- •Compound startup strategy is hard but sometimes required for adoption
- •Some incumbents (e.g., NetSuite) can’t be displaced with a point tool
- •AI may reduce time-to-ship enough to make suite builds viable earlier
- •Campfire’s traction suggests the ‘don’t build a suite’ rule has exceptions
- 21:20 – 22:57
Codegen collapses switching costs: enterprise sales cycles and migrations get dramatically shorter
Garry explains how code generation and automation change enterprise dynamics: migrations and integrations that took months can shrink to weeks or less. This creates new openings for startups to replace entrenched systems by reducing time-to-value and implementation risk.
- •Historically, schema/data migrations required weeks of brittle custom scripts
- •Bad migrations lead to failed deployments and churn risk
- •High-quality demos plus codegen can shorten buying + implementation cycles
- •Lower switching costs expand what small teams can credibly sell into enterprise
- 22:57 – 25:13
Forward-deployed engineers may be overused—so automate them (GigaML’s ‘AI FDE’)
They question whether the human forward-deployed engineer model has become an overused default playbook. GigaML is presented as flipping the model by using AI/codegen to perform the customization work in minutes, turning ‘services’ into product speed.
- •FDE model (Palantir-style) blurred software and consulting; now it’s mainstream
- •Overuse risk: becoming a default rather than a strategic exception
- •GigaML replaces human FDE work with an AI-driven customization engine
- •Faster implementations become a decisive competitive advantage
- 25:13 – 30:07
Flock Safety: ‘unfundable’ on paper, inevitable from first principles (hardware + local government GTM)
Garry recounts investing in Flock after experiencing a neighborhood car-break-in and learning police needed license plates to act. Despite VC objections (hardware, small TAM, Atlanta, government sales), the acute customer need made the bet compelling and created a low-competition wedge.
- •Personal experience clarified the problem: without plates, police can’t act
- •Product: solar + edge computer vision camera system
- •Why VCs said no: hardware, “small market,” local government sales, non-SF origin
- •Lesson: TAM math and investor ‘rules’ can hide massive emergent markets
- 30:07 – 33:44
Distribution and impact: how Flock scaled from neighborhoods to police departments and viral proof via news
They describe how Flock iterated go-to-market and messaging based on real-world outcomes—solved crimes became publicity that drove adjacent cities to adopt. The company’s growth came from working backward from goals, then evolving sales from neighborhood groups to official municipal buyers.
- •Human impact created undeniable value (including serious crimes)
- •PR loop: solved crimes + B-roll on evening news drove viral awareness
- •Customer pull from neighboring cities/police chiefs accelerated adoption
- •Strategic pivot: neighborhood groups weren’t enough; police departments unlocked scale
- 33:44 – 36:25
The ‘sci-fi founder’ pattern: impossible ideas, long timelines, and enduring skepticism (OpenAI, SpaceX)
Diana and Jared discuss founders who pursue extremely hard, frontier ideas that most people dismiss as impossible. OpenAI and SpaceX faced heavy negative press and expert skepticism for years, requiring unusual conviction and tolerance for public failure.
- •Sci-fi founders tackle problems that seem beyond current science/engineering
- •OpenAI looked like a research side-quest before the pieces connected
- •Critiques: lack of papers/peer review; ‘kids can’t build AGI’
- •SpaceX: reusable rockets were seen as blasphemous; failures amplified negativity
- 36:25 – 37:42
Closing principle: truth comes from users and lived reality—not feeds, fame, or consensus
Garry closes by reframing contrarianism as a filtering mechanism: most people will doubt you, but the few who share the belief become your team and customers. He urges founders to ground decisions in direct user signals and personal verification rather than doomscrolling or external authority.
- •Expect 9/10 people to call you crazy; the 1/10 becomes your magnet
- •Re-examine where your beliefs come from; validate via direct contact
- •N=1 opinions (including theirs) are less important than user reality
- •Obsess over real problems, attract aligned people, and iterate to solutions
