Our 2021 version of this post said: look for problems, not ideas; scratch your own itch; and live in the future. All three came, without much credit, from one essay. This rewrite credits it, adds what it leaves out for a small software company in 2026, and takes in our short post on the four types of products, which made the essay's best point as a two-by-two grid.
The essay everyone paraphrases
In November 2012 Paul Graham wrote "How to get startup ideas". Its opening line is the advice in one sentence: "The way to get startup ideas is not to try to think of startup ideas" (Graham). His argument, in short:
- Ideas you sit down to invent are usually bad. He calls them "made-up" or "sitcom" ideas: they sound plausible, and friends say they could see themselves using them, but nobody actually does.
- Good ideas grow out of your own work and life, especially if you're at the leading edge of a field that's changing fast. His summary: "Live in the future, then build what's missing."
- Depth beats breadth. Better a small group of people who want something badly than a big group who want it a little. He compares it to a well: narrow and deep.
- Turn off the filters that make you skip tedious problems (he calls this "schlep blindness") and unglamorous ones. Stripe, which handles the tedium of card payments, is his example.
- Competition matters less than founders think. Startups rarely die from competitors; they die because nobody needs the thing.
The last point is backed by data. In CB Insights' March 2026 study of 431 venture-backed startups that closed since 2023, poor product-market fit was a factor in 43% of the cases where a reason could be identified, and running out of money in 70% (CB Insights). The two are linked: money runs out while the team builds something not enough people want.
The four types of products
Our 2021 post sorted products by two questions: is the problem big or small, and do many or few people have it?
| Few people | Many people | |
|---|---|---|
| Big problem | 1. Best place to start | 2. Large incumbents, hard to enter |
| Small problem | 3. No business | 4. A feature of someone else's product |
- Big problem, many people: payroll, accounting, email. The need is real, and so are a dozen well-funded products. A newcomer needs a sharp angle.
- Small problem, many people: usually solved as a feature inside a bigger product. A small company can win here only with a free tool or an app-store listing, and it's hard to charge.
- Small problem, few people: not a business.
- Big problem, few people: the best place to start. These people are easy to find, they talk to each other, and they'll pay for something that works even if it's rough. Graham's "well" is this square.
The two-by-two is a simplification: "few people" has to mean enough to pay you. Our micro-SaaS guide works out how many customers a target needs at $9, $29 and $99 a month. At $99 a month, about 50 customers already make $5,000 MRR.
A source the 2021 post missed: rules with a date
Graham's "live in the future" works if you're at the edge of a fast-changing field. For most small founders there's a more ordinary version: a new rule with a deadline. When a rule changes, thousands of small businesses must do something new by a fixed date, the big products are slow to add it, and buyers search for exactly the words in the rule. Three recent examples:
- Email sender rules. Since 1 February 2024, Gmail requires anyone sending more than 5,000 messages a day to Gmail users to set up SPF, DKIM and DMARC, offer one-click unsubscribe and keep the spam rate below 0.3% (Google). Every small SaaS that sends email had to learn three acronyms at once.
- EU VAT in the Digital Age (ViDA). Adopted on 11 March 2025, it lets EU countries require e-invoices and brings digital reporting for cross-border business sales from 1 July 2030 (European Commission). Millions of small sellers will need software that produces and checks a format they've never heard of.
- Accessibility and AI rules. The European Accessibility Act has applied to many consumer services since 28 June 2025, and the EU AI Act's transparency duties apply from 2 August 2026, with its high-risk rules moved to December 2027 by the 2026 Digital Omnibus. We covered what they mean for a small SaaS in does the European Accessibility Act apply to you and the EU AI Act for a small SaaS that calls an LLM API.
Rules make good ideas because they pass all of Graham's tests at once: the people affected want a solution badly (there's a penalty), they're easy to find (the rule defines them), and the work is a schlep that big companies put off. The risk is that the window closes when the big products catch up, so the product must also be good at the job after the deadline passes.
Where else to look, in order of how often it works
- Your own work. A process you've done by hand for years, in an industry you know. You know the words customers use, where they gather and what they already pay for. This is still the most reliable source.
- Other people's spreadsheets. Ask people in a trade to show you the spreadsheet they run part of their business on. Every complex spreadsheet is a product someone hasn't built yet.
- Rules with a date, as above.
- Negative reviews of an existing product. One- and two-star reviews on software marketplaces name the missing feature in the customer's own words. A cluster of the same complaint is a niche.
- A new capability. Something that became cheap recently: running AI models in a browser, reading documents automatically, translating well. Ask which tedious job it makes affordable for a small business that couldn't afford it before. Our free tools do many such jobs in the browser, and seeing what people use them for is a good way to spot repeated work.
Sitting down with a blank page to "think of a startup idea" is not on the list, for Graham's reason: it produces ideas that sound good and nobody needs.
A five-question filter before you build
Run every idea through these, in order. Stop at the first "no".
- Can you name ten people who have this problem, and reach them this week?
- Do they already spend money or hours on it? A workaround (a spreadsheet, a freelancer, a manual process) is good evidence; "it would be nice" is not.
- Is it big for them? Would they be in trouble, lose money or miss a deadline without a solution?
- Can you build a first version in weeks, not months? Our post on when to launch an MVP covers what "first version" means.
- Would they pay before it exists? A pre-order, a paid pilot or a signed letter of intent. Our idea validation guide covers how to ask without getting polite yeses.
An idea that passes all five is worth a few weeks of building. One that fails the first two is a sitcom idea, however good it sounds.
What changed since 2021
The 2021 post ended with "have an eye on the future and create fun projects". That's still how many good ideas start, but it's no longer enough on its own: building got cheap, so obvious ideas now get built many times over. The edge has moved to knowing one group of customers well, and to boring, dated problems such as rules and formats that others skip.
Sources
- Paul Graham: How to get startup ideas (November 2012).
- CB Insights: Top reasons startups fail (5 March 2026).
- Google Workspace Admin Help: Email sender guidelines (read 29 October 2026).
- European Commission: VAT in the Digital Age (ViDA) (read 29 October 2026).
- Our earlier posts on the European Accessibility Act and the EU AI Act, which cite the regulations themselves.
Building a software product?
Talk it through with us on a free 30-minute call, or email [email protected]. You can also try our free tools.
Suggested Articles
Here are a few more articles you might like.
