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
Why chasing “hot” ideas leads to crowded, derivative startups
Garry frames the core thesis: if you only work on what’s trendy, you’ll end up in hyper-competitive markets where most startups fail. The antidote is to pursue contrarian ideas rooted in real user needs, even if they initially sound “crazy.”
- •Hot markets create dozens of competitors; only a few winners survive
- •Contrarian ideas often sound wrong to most people at first
- •The goal is to find what people desperately need, then work out the rest
- •“Competition is for losers” as a lens for idea selection
AI’s early greenfield window is closing—competition is back
Harj explains that a year earlier it was easier to find AI startup ideas due to greenfield verticals and rapid model step-changes. Now many vertical agent categories are crowded and model progress feels more incremental, making unique insight more important.
- •Earlier: abundant greenfield AI verticals + frequent model breakthroughs
- •Now: multiple startups per vertical; fewer “easy” workflow-automation picks
- •Founders need a unique insight/contrarian bet to stand out
- •The idea space isn’t expanding as fast as before
How to find a ‘secret’: non-obvious ideas feel risky, not neutral
The group distinguishes “non-obvious” from merely uncommon—truly non-obvious ideas feel dangerous and career-risky. Garry highlights how founders can misread promising signals due to social/media narratives about ‘tarpit ideas.’
- •Non-obvious ideas often trigger fear of wasting years
- •Mental models from peers/media can suppress good instincts
- •Dead competitors can be a sign of difficulty—or new timing with AI
- •Early customer pull can coexist with external skepticism
Lessons from past platform shifts: the two-year gold rush, then deeper secrets
Jared compares AI to earlier tech shifts (internet, smartphones): there’s a brief window where obvious startups are easy to spot. After that, the best opportunities require deeper, contrarian insights—and can look unrelated to the platform at first.
- •New platforms open a short ‘gold rush’ for obvious ideas
- •After ~2 years, obvious categories get picked over
- •Big winners are often unexpected consequences of the platform
- •Founders must search for the next layer of secrets
Zimride/Lyft and the gray-area advantage: users can force laws to change
The conversation uses ridesharing’s early days to show that some breakthrough startups live in legal gray areas created by outdated rules. Founders who focused on user benefit took the risk; others held back due to fear of illegality.
- •Early ridesharing was widely viewed as potentially illegal
- •Some teams hesitated; Lyft/Uber launched despite fears (even jail)
- •If user value is overwhelming, regulation often adapts over time
- •Many great ideas sit where law hasn’t caught up to technology
First-principles regulation: outdated laws, safety tradeoffs, and open banking
They clarify they’re not advocating ‘do illegal things,’ but rather to examine whether laws reflect current technological realities. Garry uses open banking/Plaid as a live example of regulation, incentives, and potential regulatory capture by incumbents.
- •Think from first principles: who benefits, what harms exist, how to mitigate
- •Some laws were written for a pre-smartphone/pre-crypto world
- •Incumbents may use “consumer safety” narratives to block competition
- •Open banking as a current battleground over data access and fees
Framework for contrarian bets: flip the current startup playbook
Diana asks what contrarian bets founders should look for now; Harj proposes a method: identify emerging ‘consensus’ playbooks and consider taking the other side. DoorDash is introduced as an example of rejecting the then-fashionable ‘full-stack startup’ meme.
- •Contrarian search method: invert today’s default advice
- •DoorDash vs the ‘full-stack’ food startups (ghost kitchens)
- •Consensus playbooks can become wrong as conditions change
- •AI founders should question the last year’s “standard operating model”
Compound startups in AI: when building the whole suite beats a point solution
Diana discusses the “compound startup” concept and when it might now be viable with AI. Campfire is presented as an AI-native competitor to NetSuite, where a full product is necessary to win adoption despite the usual early-stage bias toward narrow scope.
- •Compound startups are hard but may be more feasible with AI
- •Some markets require an end-to-end suite (not a point tool) to displace incumbents
- •Campfire example: AI-native CFO/ERP product taking on NetSuite
- •Traditional ‘ship fast’ advice may flip in certain categories
Codegen changes enterprise switching costs—and reshapes go-to-market speed
Garry explains that code generation can dramatically reduce integration and migration friction, compressing time-to-value from months to weeks (or less). This enables startups to win enterprise deals faster and makes previously infeasible sales motions workable.
- •Codegen can reduce schema conversion and integration effort
- •Time-to-value can drop from ~1 year to <1 month in some cases
- •Better demos + faster implementation compress sales cycles
- •Creates new possibilities for complex enterprise startups
Forward-deployed engineers: from contrarian tactic to overused default
Harj and Garry revisit the forward-deployed engineer (FDE) model: once a Palantir-style contrarian move, now widely copied. They note Bob McGrew’s skepticism that it’s being over-applied and propose it as a candidate for contrarian inversion.
- •FDE model blurs consulting and software to speed adoption
- •It’s become a default enterprise AI playbook
- •Risk: overuse can mask weak productization and hurt scaling
- •Contrarian move may be to minimize or replace FDE usage
AI ‘forward-deployed engineer’ as product: GigaML’s automation twist
Diana and Harj describe GigaML’s approach: replacing human FDE work with an AI-driven system that transforms customer specs quickly. This shifts customization from weeks to minutes and reframes what ‘deployment’ means—more product, less services.
- •GigaML replaces human FDE effort with codegen/AI automation
- •Faster implementations help win deals against slower competitors
- •Turns customization into a near-instant product workflow
- •Example of flipping an entrenched enterprise playbook
Flock Safety case study: ‘unfundable’ hardware + government sales → massive impact
Garry recounts investing in Flock Safety after experiencing a neighborhood break-in and learning police needed license plates to act. Despite VC antipatterns (hardware, small initial market, Atlanta, selling to governments), the product’s user-need intensity drove adoption and growth.
- •Personal pain made the first-principles case obvious (crime solving)
- •Product: solar + edge computer vision license-plate cameras
- •VC objections: hardware, small TAM, non-SF geography, gov sales
- •Focus on real need led to viral-like spread via solved crimes
Scaling Flock: distribution discoveries, go-to-market pivots, and TAM fallacies
They draw general lessons: VC-style TAM math can be misleading early, and founders can expand markets through execution and pivots. Flock evolved from neighborhood associations to selling into police departments/city governments and used media amplification to accelerate inbound demand.
- •Do not over-index on early TAM calculations
- •Distribution can emerge from real-world outcomes (newsworthy solves)
- •Pivoted GTM from neighborhoods to city/police buyers
- •Business model and market size can expand after product proves value
The ‘sci‑fi founder’ pattern: impossible ideas, negative press, and persistence
Diana highlights founders who pursue extremely hard, long-horizon ideas that most people dismiss as impossible. OpenAI and SpaceX are used to illustrate enduring ridicule, repeated setbacks, and the need to remain steadfast until the world catches up.
- •‘Sci‑fi founders’ tackle ideas gated by science/engineering difficulty
- •OpenAI: early skepticism, criticism about lack of papers, unclear path to product
- •SpaceX: reusable rockets seen as blasphemous; failures amplified by press
- •Persistence through years of doubt is often required
Closing advice: verify reality through users, not discourse; contrarian magnets win
Garry concludes that being contrarian is about grounding beliefs in direct experience and customer truth, not internet consensus. If many call you crazy, the small minority who agree can become your early adopters and collaborators—if you build for real needs.
- •Nine people may reject your idea; the tenth can be your core customer
- •Re-examine where your beliefs come from; prioritize verifiable user input
- •Ignore doomscrolling and status games—even from famous voices
- •Be a ‘magnet’ for people who share the contrarian belief by executing
