How to start a startup in 2026: a 12-week path for one or two founders

What are the steps to start a startup in 2026?

A startup, for a one- or two-person team, is a small product sold to a defined group of people, with a plan to grow it past what you can do by hand. It is not a funding round, an office or a company name. Those are things you may get later. The path below is built around the one thing that cannot be skipped: someone gives you money or a firm commitment for something you made.

1. Pick a problem with a buyer. Name who has it, what it costs them today, and who in their organization or household pays.
2. Talk to ten people who fit. Ask about the last time the problem happened and what they did about it.
3. Pre-sell or get a firm commitment. A signed letter of intent, a deposit, or a booked paid pilot counts. A compliment does not.
4. Scope one outcome. Write the smallest product that delivers one result for one kind of customer.
5. Build it with AI coding tools. Use a written brief, a prompt, and a short list of checks you can run yourself.
6. Set a price and a way to collect it. Charge from the first user, even if the price is low.
7. Handle the legal basics. Terms, privacy notice, and a decision about when to form a company, based on the rules where you live.
8. Launch to the people you already know. Then ask each of them for one introduction.

Everything that follows expands one of these steps. If you do not have an idea yet, start with how to find a startup idea, then come back here.

How do you choose a problem worth a company?

A good first problem has three properties: it recurs, it already costs the person money or time they can name, and they currently solve it badly, with a spreadsheet, a stack of email or a manual routine. A problem people only complain about, and never work around, is a weak signal. If nobody has built a workaround, nobody has tried hard to solve it.

Pick a narrow buyer. “Small business owners” is not a buyer. “Owners of independent physiotherapy clinics who book patients by phone” is one. The narrower the group, the easier it is to find ten of them, to write a message that speaks to them, and to build only what they need.

Your own history is the cheapest source. A task you did for an employer, a tool you built for yourself, a client who kept asking you for the same thing. You already know the vocabulary and you know where to find the other people who have it. If you want raw material, you can browse ideas other people posted on the board and check which problems you could describe from experience.

How do you find first customers before you build anything?

Go where the buyer already talks: a forum, a professional community, a niche newsletter’s replies, a trade group, or the comments under the tools they use. Send short, specific messages. The goal of the first message is a 20-minute conversation, not a sale.

Hi [first name], I’m looking into how [role] handle [specific task]. I’m not selling anything yet. Could I ask you three questions about the last time it came up? It should take 20 minutes, and I’ll share what I learn from other [role]s afterwards.

In the conversation, ask for a story, not an opinion: when did it last happen, what did you do, what did it cost, what did you try. Do not describe your idea until the last five minutes. Then ask the question that matters: “If I built this and it worked, what would you pay, and could we start with a trial next month?” Write the answer down as spoken. The method is explained in more detail in how to validate an app idea.

Count signals, not smiles. Strong signals: they show you the spreadsheet they use, they name a budget, they introduce you to a colleague, they agree to a paid pilot. Weak signals: “interesting”, “send me a link when it’s ready”, “I’d definitely use that”. If after ten conversations nobody has given a strong signal, change the problem or the buyer before you write code.

How do you build the first version with AI coding tools?

AI coding tools such as Claude Code, Cursor, Codex and Lovable can often produce a working first version of a narrow product quickly, if you give them a precise brief. That moves the hard part away from typing code and toward deciding what to build. A vague prompt produces a vague product.

• Write the scope first. One user, one outcome, a short list of what is out. How to scope an MVP walks through the boundary, and the MVP scope planner gives you a structured start.
• Write the brief as a build prompt. Include the user, the main workflow, data you store, and the failure cases. See how to write an app build prompt for a template.
• Build in small steps. Ask for one screen or one workflow at a time, run it, and fix what breaks before adding more.
• Write your own acceptance checks. Five things a customer must be able to do, which you test by hand before each release.
• Keep the stack boring. A hosted database, hosted authentication if you need accounts, and a payment provider you did not write yourself.

Review what the tool produced, especially anything involving logins, payments, or personal data. AI-generated code can look right and be wrong in ways that only show up when a real customer does something unexpected. Treat the output as a fast first draft from a contractor you have not worked with before.

How should a new startup set its first price?

Charge from the first user. Free users tell you whether something is pleasant to try. Paying users tell you whether it solves a problem worth money, which is the only question that decides whether you have a business. Start with a price that makes your first ten customers think “that’s easy to approve”, then raise it for new customers as you learn.

Tie the price to the value the customer gets, not to your costs. If the product saves a clinic two hours of phone calls every day, a monthly price that is a fraction of that time is easy to justify. For subscription tools, how to price a micro SaaS covers value metrics, tiers, trials and annual plans in detail.

Use an established payment provider rather than building billing yourself. Which one fits depends on where you and your customers are, and on who ends up responsible for sales tax or VAT on each sale. Read the provider’s documentation on that point and check the rules where you sell.

What are the legal basics for a new startup?

Rules differ by country, and often by state or canton, so this is a general checklist, not legal advice. Talk to a local accountant or lawyer before you rely on any of it. Whether you can take early payments before a company exists, and who is liable and taxed when you do, depends on where you live, so find out before the first payment instead of after.

Do not copy another company’s terms word for word. They were written for a different product, in a different jurisdiction. Use a template from a reputable source as a starting point and have it reviewed once the product has revenue.

How do you launch a startup without an audience?

Launch first to the people you already talked to. Email each of them personally with a link and a specific offer, such as a founding price for the first ten customers. Ask them for feedback after the first few days of use, and ask each for one introduction to someone with the same problem.

Then add public launches: a short post explaining the problem and what you built, a listing in free directories, and the communities where your buyers already are. The free launch directory finder helps you pick places by audience, and the launch copy generator drafts the tagline and description. A fuller plan is in how to launch without an audience.

Track where each signup came from. A small spreadsheet with source, date and outcome is enough. The point is to learn which channel produces customers, so the next month of effort goes there.

What should weeks 1–12 look like?

This schedule assumes full-time focus. If you work around a job, double the calendar time and keep the order. The order matters more than the speed.

At the end of week 12, ask one question: did anyone pay, and would they be upset if the product vanished? If yes, keep going on the same buyer. If no, you have lost three months and gained a list of people and a clearer idea of which problem to try next. That is cheaper to learn now than after a year of building alone.

What mistakes slow new founders down the most?

• Building before talking. The product feels like progress and the conversations feel like risk, so people reverse the order.
• Choosing a vague buyer. If you cannot name where to find ten of them, the launch will fail regardless of the product.
• Adding features to avoid pricing. Charging is the test. More features postpone it.
• Spending weeks on a company name and logo. A plain name and a clear sentence beat a polished brand with no customers.
• Waiting for the perfect idea. Pick a good-enough one, test it fast, and drop it without drama if the evidence says no.

If you have an idea you do not plan to build yourself, you can post it on the board and let someone else try. Posting is free, needs no account, and is permanent.