What do these people have in common? No degree, no connections, no seed capital. They didn't grind out code alone for years. They talked to an agent — and the agent built. It took them months, not a decade.
Start from zero
How to make an AI agent work for you — no degree, no code, no seed money. A step-by-step plan for your first week.
Facts first
Tel Aviv. One person, six months.
Maor Shlomo launched Base44 in January 2025 — solo, no investors, no office. Six months later it had 250,000 users, and on June 18, 2025 Wix bought it for $80 million in cash. Shlomo gave $25 million of that to the eight employees who had joined along the way.
TechCrunch, Jun 18, 2025 ↗Stockholm. $100M ARR in 8 months.
Lovable launched in November 2024. By July 23, 2025 it hit $100M ARR — the fastest ramp in software history. 2.3 million users, 180,000 paying — with just 45 employees. In December 2025 it raised $330M at a $6.6B valuation.
TechCrunch, Jul 23, 2025 ↗New York. High schoolers, an app from a bedroom.
Zach Yadegari and Henry Langmack started Cal AI in May 2024, at 17. Fifteen of the eighteen colleges Zach applied to rejected him — with a 4.0 GPA. The app reached 15 million downloads and $30M a year, and on March 2, 2026 MyFitnessPal acquired it — Cal AI had crossed $50M ARR by the deal.
TechCrunch, Mar 2, 2026 ↗Your first week
- 1 Day 1
Pick one pain
Not "I want a business" — one specific problem you understand better than most, from your job, your city, your hobby. Simple test: if you can explain it in a minute, it qualifies.
- 2 Day 1
First conversation with an agent
Open any agent — Claude, ChatGPT, Lovable, Base44. Don't ask it to "write code." Describe the outcome: who the user is, what pain you're solving, what done looks like. The agent will figure out how to build it.
- 3 Day 2
A prototype in one evening
Ask for the smallest working version — one feature, one screen. Don't polish. The evening rule: by midnight it should work, not look pretty.
- 4 Day 3
Critique, don't rebuild by hand
Changes are words too: "this part is awkward, make it like this." You're the editor and the client. The agent is the builder. The moment you start rebuilding by hand, you're back in the old world where this takes years.
- 5 Days 4–5
Show it to three real people
Not friends — people who actually have this pain. Their reactions are the script for your next conversation with the agent. Two of these loops and a prototype becomes a product.
- 6 Day 6
Hand the routine to the agent
Let the agent not just build but run things: answer inquiries, collect data, write reports. You handle people and decisions — everything else gets automated.
- 7 Day 7
Measure it
Hours saved, leads, money — any honest number. If there's a number, there's a system you can scale. No number — go back to step 1 and narrow the pain.
Do the math: what your routine actually costs
Step 7 of the plan is numbers. Move the sliders — this is what routine work takes from you right now.
routine eats per month
routine eats per year
an agent gives back per year (~70% of routine)
Conservative math: 70% is a typical share of routine an agent handles on its own after the first week.
Three mistakes that kill a start
- Perfectionism. A month of "almost done" means nobody has seen it yet. Show it rough.
- Prompt-engineering courses instead of a conversation. Agents understand plain language — explain the task the way you'd explain it to a person.
- A product "for everyone." For everyone means for no one. One pain, one audience, one outcome.
Frequently asked questions
Do I need to know how to code?
No. Yadegari built the first Cal AI prototype as a high schooler, and most Lovable users never write code at all. Your job is understanding the customer's pain and describing the result clearly. Code is the agent's job.
How much money do I need to start?
$0–40 a month. Claude, ChatGPT, Lovable and Base44 all have free tiers — enough for the prototype in the plan above. Paying makes sense once real people have tried your prototype (days 4–5).
Which agent should I pick first?
The one closest to your task: a site or an app — Lovable, Base44; writing, analysis, automation — Claude, ChatGPT. Hard to get wrong: the skill of framing a task transfers between agents one-to-one.
How long until the first result?
A working prototype — one evening (day 2 of the plan). A version you're not embarrassed to show — a week or two. After that it's all about how many loops of “show → listen → brief the agent” you run.
The agent makes mistakes. Now what?
It does, regularly. That's why day 3 is critique in plain words and days 4–5 are testing on real people. You're the editor and the client: you accept the work, you don't take it on faith.
Is it too late to start in 2026?
Cal AI was acquired in March 2026, Lovable raised $330M in December 2025 — money is still accelerating into this space. Too late is when someone else does it in your niche.
Get the extended checklist
The full step-by-step plan with example phrasings for the agent. Opens right after you subscribe — we'll also send it to your inbox. No spam, one-click unsubscribe.
This page is alive: we update it as new cases and tools appear.