Investing in Ideas: A Five-Step Discipline for Choosing What to Build
Updated
Knowledge on this page was mainly distilled from the following articles: Knowing Is Already Doing, Critical Demand Signals Nobody Talks About, Most Things Are Black and White, The AI Moat You Can't Buy, The Idea Is the Product Now.
Why Idea Investment Is a Discipline
When AI compresses execution to hours, the highest-leverage activity is choosing what to build. But most builders skip straight to shipping. Investing in ideas isn't a eureka moment — it's a repeatable discipline that looks quiet, deliberate, and nothing like hustle culture.
The Five Steps
1. Live Inside the Problem
Use the broken tools. Do the manual workflow. The frustration is the data. Reading about market gaps from a distance produces ideas that sound right and feel hollow. Direct contact with reality — either friction with an existing problem or sensing patterns in how people actually behave — is where high-leverage ideas emerge.
2. Kill Ideas on Paper Before Building Them
Write the idea in one paragraph. Who is this for? What pain does it address? Why hasn't someone solved it already? If the answers aren't clear in writing, the idea isn't ready. Building won't fix that. This is where using AI as a thought compiler can add structured friction to expose weak assumptions.
3. Look for Signal Before Writing Code
This is harder than it sounds. A landing page with no traffic proves nothing. A tweet can get engagement from people who'll never pay. Conversations mislead when you ask leading questions. Every validation method has a way of telling you what you want to hear. The discipline is in designing tests that can actually say no — and then listening when they do.
4. Stay With the Problem Longer Than Feels Comfortable
The first idea is almost never the best one — it's the obvious one. The better idea usually hides behind it, visible only after the obvious one has been examined and found wanting. Resist the urge to ship the first version that comes to mind.
5. Make the Risky Bet Explicit
If the idea requires a market that doesn't exist yet, name that bet out loud: "I believe people will want X because I'm seeing Y." Risky insights produce some of the highest-leverage ideas — think Apple's bet on what a tablet should feel like — but only when the risk is conscious and named rather than hidden behind optimism.
Two Types of High-Leverage Ideas
- Pain-driven ideas: Solve a problem people already feel. They have a built-in audience, are specific enough to be opinionated, and can't be replicated by anyone who types "give me startup ideas" into a chat window.
- Category-creating ideas: No existing market is asking for them. Nobody was asking for an iPad. These emerge from noticing patterns in behavior and betting that a new category wants to exist. Microsoft had the engineering to build a tablet. Apple had the sharper idea of what a tablet should feel like.
Both types share one thing: they emerge from direct contact with reality, not from prompting an AI for startup ideas. AI generates ideas at ninety per minute, but they pattern-match on what already exists and lack the texture that comes from living inside a problem.
Q&A
Why can't AI generate high-leverage ideas?
AI pattern-matches on what already exists. It generates ideas from the outside looking in. High-leverage ideas emerge from living inside a problem or sensing behavioral shifts before they have a name — texture that statistical prediction can't replicate.
What is the difference between pain-driven and category-creating ideas?
Pain-driven ideas solve a problem people already feel and have a built-in audience. Category-creating ideas bet that a new category wants to exist — like the iPad — based on behavioral patterns, not existing demand. Both require direct contact with reality.
How do you design a validation test that can actually say no?
Avoid tests where engagement is ambiguous — landing pages with no traffic, tweets from non-buyers, conversations with leading questions. Structure the test so a negative result is clear and unmistakable, then commit in advance to accepting that result.
Why is the first idea usually the wrong one?
The first idea is typically the obvious one — the solution that comes to mind before the problem is fully understood. Better ideas hide behind it, becoming visible only after the obvious idea has been examined and found wanting.
How should fast feedback change the way you pick ideas?
You should favor ideas where usage quickly reveals whether the product is helping or failing. Fast feedback lets you improve the product through repeated contact with reality, which is now more valuable than simply shipping version one. In practice, this means looking for frequent workflows, measurable outcomes, and users close enough to the pain to react quickly.
Why can a smaller niche be a better starting point than a bigger market?
A smaller niche can be better because it often concentrates repeated behavior and clearer signals, which helps the product learn faster. In a broad market, feedback is noisier and iteration takes longer, even if the upside looks larger on paper. The best wedge is often the market where learning compounds first, not the market with the biggest headline number.
What kind of idea is best suited to building a product loop?
The best ideas for a product loop involve recurring decisions, visible outcomes, and a user who can notice improvement quickly. Think of workflows where people repeat the task often enough to generate real behavioral data, not just survey opinions. If every use produces a signal you can learn from, the product has a better chance to compound.
How do I know if I'm validating or evading?
If your validation process keeps producing ambiguous results and you keep running more tests, you may be evading a verdict rather than seeking one. A well-designed validation test can say no clearly. If you have run several tests and still cannot state whether the idea works, ask whether you are avoiding the answer because it would require you to kill the idea.
What is the duct tape test for idea validation?
It is a method for confirming demand by finding people who have jury-rigged solutions from tools never designed for the job. You look for three things: multiple people using different wrong tools for the same problem, solutions that barely work, and independent invention with no coordination. All three together indicate urgent, unmet demand rather than a mild preference.
Why are workarounds a stronger demand signal than complaints or wishes?
Workarounds prove someone crossed the activation-energy threshold from annoyance to action. Wishes cost zero effort and signal almost nothing. Complaints often point toward feature requests for existing tools. But when someone has invested real time building a brittle solution from mismatched tools, they are already behaving like a customer paying with time and frustration instead of money.
How do adjacent-space workarounds strengthen validation?
When people in related but different fields independently hack together the same kind of solution, the problem is structural rather than personal. For example, Zapier's founders found the same integration pain across Evernote, Salesforce, and Dropbox forums. No single community asked for Zapier, but dozens independently cobbled together the same brittle workflow. Adjacent-space patterns confirm the problem transcends any one niche.
How do you tell productive research apart from avoidance during idea investment?
Ask one question: "What will I do differently because of this?" If you can name a concrete action the research enables, you are inside a productive cycle. If you cannot, and the research feels comfortable rather than challenging, it is likely avoidance disguised as preparation. The distinction is purpose, not volume.
Should each idea investment loop produce a visible artifact?
Yes. Each cycle of knowing and building should produce something that can disappoint you, whether that is a prototype, a landing page, or a structured conversation with a potential user. Without an artifact that touches reality, the loop spins without generating real feedback, and learning stays theoretical rather than compounding into better decisions.