samhuleatt.com

Startup Ideation

I spent a full year as an EIR trying to find a startup idea before I joined On Deck. This is the synthesis of the frameworks and heuristics I collected and tried to apply. Just as useful for investors as for founders.

One note on AI before we start. As of 2026, I do not think models are much help with startup ideas. They lack creativity and lived experience. If anything they drift toward consensus, when the best approaches are often deeply human: personal, orthogonal, or taste-driven. Those are the areas models seem to struggle with most.

Start with a hair-on-fire problem

Great startup ideas are not invented. They are felt, usually by someone living so close to a problem they cannot let it go. You do not think your way to them at a whiteboard. You notice the thing that has been bothering you for years, or obsess over a problem you feel uniquely capable of solving. As Paul Graham put it in How to Get Startup Ideas:

The way to get startup ideas is not to try to think of startup ideas. It’s to look for problems, preferably problems you have yourself. The very best startup ideas tend to have three things in common: they’re something the founders themselves want, that they themselves can build, and that few others realize are worth doing.

The thing that matters most in both judging and forming ideas is the quality of problem fit: the right person solving the right problem. Before that, decide whether you are chasing a venture-scale problem or not, because the two are different games.

Most businesses do not need venture capital. So-called lifestyle businesses get a bad rap, but they are often the better choice and frequently produce better outcomes for founders than venture-funded ones. A venture-scale problem is different. It should feel like a race against time, a hair-on-fire problem for a massive set of customers, where your only path to winning is dominating the market. Choosing the venture path is an agreement: you are going to the moon, or you are blowing up on the launch pad.

One trap worth naming: most founders who feel they need to raise money in order to “validate” a problem will fail. Often they are really trying to validate themselves. The highest-odds version is to raise only after you have proven to yourself that the problem is hair-on-fire and that solving it now, at scale, is imperative.

Action: Before anything else, decide if you are building a venture-scale company or a great business. Both are valid. They are not the same path.

Find your earned secret

Even a great idea gets discounted heavily if you are not the right person to execute it. The good news is that the best ideas you will have tend to be ones you are uniquely suited to solve, because you have an earned secret. A16Z’s definition is the one I use: you tried to solve some hard problem in your past and learned something about the world that few other people know.

Rather than hunting for individual ideas, hunt for what YC calls idea spaces: clusters of related problems where you have unique insight. Map your team’s skills and earned secrets to the spaces where you can see what others cannot.

  • What are my actual superpowers?
  • Do I have an earned secret, something I learned the hard way?
  • Do I have an unfair advantage, usually in distribution or access?
  • Is there a problem I am obsessed with solving?

The last question is the one I weight most. A good idea is usually the problem you literally obsess about. The best founders have turned it over from every angle, and within minutes it is obvious they could talk about it for a week straight. If you have to manufacture that energy, it is the wrong problem. David Zhou and I ask a version of this when we invest in funds: is this the last job this person will ever have? When the answer is yes, they have gone to unreasonable lengths to be the right person for it. Founders Fund makes the same point in Choose Good Quests: pick something worth years of your life.

This is also the part AI cannot do for you. A model can map a market, but it cannot have been in the room where you earned your secret. The intersection below is the only place worth starting.

EARNEDSECRET ↑mildhair-on-firePROBLEM INTENSITY →WORTH YOUR LIFEyour secret + a real problemSOLUTION LOOKINGFOR A PROBLEMSKIP ITSOMEONE ELSE’SCOMPANY
The only quadrant worth your life

Action: Write down your earned secrets before you write down ideas. The ideas worth pursuing fall out of the secrets.

What makes an idea great

Once you are in the right space, here is what separates a great opportunity from a fine one:

  • You have a secret: a non-obvious solution to an obvious problem.
  • Your solution is 10x better and cheaper, democratizing access to something previously elite.
  • There is a clear why-now: a regulatory change, a new platform, a behavior shift, a technology change, or a new business model.
  • It confers status, which creates mimetic desire and network effects.
  • It has market pull: a problem so severe the market drags the solution out of you.
  • It targets a small but rapidly growing niche, often a market that looks boring.

Uber is the cleanest case study. The secret was that black cars were underutilized assets. It let drivers self-actualize by earning and let riders self-actualize through status. It was a 10x better experience than yellow cabs, and cheaper. It was not possible before the iPhone, with its apps, payments, and GPS. And it induced mimetic desire: my friend takes black cars, now I want that too.

Action: Run your idea against this list. If it only hits one or two, keep looking. The best ideas hit most of them at once.

Heuristics to generate ideas

Ideation is a muscle. You have to practice and give ideas time to marinate and cross-pollinate. Steve Jobs said everything in the world was built by people no smarter than you, and he was right. Don’t dismiss ideas that sound a little crazy either. As Sam Altman put it, the best ideas are fragile, and most people never even say them out loud because they sound silly.

Here are the heuristics I keep coming back to:

  • If a task takes you more than three minutes online, there is probably a business in removing it (Kevin Ryan).
  • Steal the key feature of a Series B company and build a focused version of it (TK Kader).
  • Find an emerging trend and take it to its end conclusion (commissions falling, then Robinhood goes to zero).
  • Find a commonly used spreadsheet and turn it into software (cap tables became Carta, per David Sacks).
  • Find a large, fragmented, low-NPS industry and vertically integrate to simplify it (Rabois).
  • Come for the tool, stay for the network (Chris Dixon).

YC’s startup recipes compress most of this: start with what your team is great at, build the thing you wish existed, pick something you would happily work on for ten years, look for what recently changed in the world, find new variants of recently successful companies, ask people what they want solved, and look for industries that seem broken.

A word on AI here, because it is where founders get lazy. A model will gladly generate fifty of these on command. That is exactly the problem. If a prompt can produce the idea, assume a thousand other founders got the same one, which is how you end up in a tar pit. Use AI to research and frame, not to source the insight. The opportunities I find defensible right now are vertical AI, going deep into one industry where your earned secret actually lives, and physical AI, robotics and hardware where the hard part is the real world, not the model. The thin horizontal wrapper a model hands everyone is the opposite of an edge.

Action: Keep a running idea list. Treat AI output as raw material to react against, never as the answer.

Pressure-test before you build

Once you are generating ideas, the hard part is deciding which ones deserve a deeper look. The trap is the feeling that a better idea is always around the corner. The fix is a system: a cheap, repeatable way to decide whether an idea earns more of your time.

Before you ever talk to a customer, build enough personal conviction to know the problem merits exploration:

Watch for tar pit ideas: the ones that sound great but have structural traps that have sunk founder after founder. Personal CRMs are the classic. Anything whose main pitch is “productivity gains” usually qualifies. In 2026, add the obvious one: an AI wrapper with no earned secret and no distribution edge.

Two reminders that keep you honest: solve hair-on-fire problems rather than inventing products in search of one, and favor problems where you have founder fit.

One more common mistake is starting at the market. Founders think they need a multi-billion-dollar TAM to attract investors. I would push back. Starting at the market is a step removed from the customer and the problem, and many of the best markets start small and grow fast. Kevin Kelly’s 1,000 true fans is the better mental model: worry less about TAM and more about whether five to ten customers you know truly want this and would pay for it. The best ideas expand their market over time. Severe problems tend to indicate large markets anyway.

MANY RAW IDEASPASS THE FILTERSFEW DEEP DIVESVALIDATEALL INhair-on-fire?earned secret?why now?
A system beats waiting for the perfect idea

Action: Pick one conviction exercise above and finish it this week. If you cannot write a crisp PR/FAQ, you do not understand the idea yet.

Validate cheaply without fooling yourself

I will not go deep on validation here. At a high level, get specific feedback from potential customers using a solution-asset: a landing page, deck, or memo that lets you collect real reactions and ideally sign someone up.

The thing to guard against is confirmation bias, hearing what you want to hear. The cleanest antidote is paying customers. People who are not just willing but happy to part with money are the best signal that you are onto something.

  • Surveys: useful but overrated. Easy to ask leading questions and to poll a friendly network, which feeds confirmation bias.
  • LOIs (B2B): get businesses to pre-commit. Payment in advance is even better. Investors still discount these versus real paying customers.
  • Paid acquisition (B2C): spin up a landing page and run traffic to gauge real conversion. It costs something, but it gives the least biased read.
  • Paid betas: get customers to pay a discounted but meaningful amount before the full product exists. Offer to refund them if they do not love it.

This is where AI genuinely changed the game. Prototyping used to be the expensive step. Now you can ship a working version in a weekend, which means a landing page and a waitlist is no longer proof, it is the floor. The bar moved to a real product in front of real users, with usage and retention. Cheap to build cuts both ways: the cost of testing your idea collapsed, and so did everyone else’s, so the edge is no longer the prototype. It is your earned secret and how fast you learn.

COST → 0 (AI DID THIS)GENERATING IDEASRESEARCHPROTOTYPINGVALUE LIVES HERE NOWEARNED SECRETTASTELEARNING VELOCITY
AI made ideas cheap and insight expensive

Action: Get one real customer to pay you something, however small, before you believe your own pitch.

Ideation is a muscle

A few honest things about the psychology, because nobody warns you.

Ideation is an emotional roller coaster. You swing between convergence and divergence, highs and lows, and it gets worse under a time-box, which is exactly what limited runway creates. Like any creative work, it rarely goes well under deadline pressure. The best ideation happens when your mind is relaxed, not when it is panicking.

Previously successful founders hit a particular wall, what Alex Furmansky calls the Goldilocks paradox: every new idea feels either too hard or too small, so nothing seems worthy of the next act. The way out is lowering the bar from “build the next unicorn” back to “build something my friends use.”

A few more things I have come to believe:

  • Take bigger swings. Both Jerry Neumann and NFX argue that for venture-scale problems, most founders should pursue riskier ideas, not safer ones.
  • Luck and timing matter, but they are not the whole story. Studios like Atomic, AlleyCorp, and Sutter Hill repeatedly incubate winners because they follow a rigorous process and draw on earned secrets and unfair networks.
  • The “aha moment” is mostly a myth. Where the idea came from matters far less than whether it is the right problem for you. I have rarely been the one with the initial idea.
  • Do not overthink it. Often your first idea is your best one.

There is also nothing wrong with joining a company already attacking your problem with a head start. Consider joining forces rather than starting from zero.

When Sam Altman was asked what defined the best startups, he said the degree to which you succeed approximates the degree to which you built a product so good that people spontaneously tell their friends about it. That is both as simple and as hard as it sounds. Focus maniacally on customer feedback and traction until you feel the urgency to go full-time, bootstrap until your shipping velocity actually needs outside capital, and only then sign the venture-backed founder agreement.

Action: Stop waiting for a perfect idea. Start writing ideas down today and put in the reps.

Resources

The best of what I read while working through this, if you want to go deeper:

I generally would not read a stack of books on this. Founders should be biased toward action. If you have read even half the links above, you know enough to start. Just start ideating and validating.

← all writing share on x →