In 2013 I earned ৳10,000 a month as a social media executive at Bdjobs. I did not own a laptop. On Fridays and Saturdays I carried the office laptop home and learned Illustrator, WordPress and After Effects from YouTube, because nobody was going to teach me and nobody had asked me to learn.
In 2026 I have shipped six live products and three mobile apps, mostly alone, with an AI writing most of the code. An ebook marketplace with real payments, a live quiz platform, a women-entrepreneur marketplace, a service marketplace with 78 fixed-price services, an event funnel, a game.
The distance between those two sentences is not talent. It is thirteen years of learning before anyone asked, plus one expensive lesson I want to give you for free.
The lesson KajKey charged me for
Between those two dates I co-founded KajKey, Bangladesh's first native-language freelance marketplace. Year one: 60,000 users, $500K in transactions, incorporated in the USA and Bangladesh, angel and VC money. It failed to raise further and I walked away in early 2020.
Part of the reason is one I still find hard to say in public. I put a developer team to work without proper documentation. I did not have the technical depth to know what I was asking for, so I could not tell when I was getting something else. I paid for that lesson in real money. What I learned is now the first rule of how I build with AI: architecture before code, every time.
A CTO that never says "not possible"
I tell every training room the same thing. Claude Code is not a tool. It is a CTO, a system architect, a full-stack developer, a back-end and a front-end developer in one seat, and it never asks for leave, never gets annoyed, and never says "bhai, this is not possible."
Your job is to be the founder of that team. A founder does not write code. A founder says what to build, for whom, and in what order. The vocabulary that scares non-technical people, domain, hosting, database, API, auth, payment gateway, you do not have to memorise. You ask your CTO to explain it in plain language until the decision is yours.
The founder-CTO loop
Here is the loop I run for every product. Four steps and one rule.
- Architecture first, in plain language. My first message is never "build me X". It is: "You are the CTO and system architect. I am a non-technical founder. Before writing any code, explain the whole system to me in simple terms, suggest the stack with one line of reasoning each and where the costs will be, and break the work into small milestones I can test with my own eyes."
- Small milestones. The single biggest mistake I see non-technical founders make is asking for the whole software in one prompt. One milestone, one thing I can click.
- Test with your own eyes at every milestone. Not "does it compile". Does the form actually store the data. Does the payment go through in test mode. Does the page look right on a phone.
- Ask what it needs from you before each milestone: content, accounts, decisions. The queue of blockers is on you, not on the CTO.
The rule: anything that touches money or personal data gets a human security review before it goes live. AI-built does not mean safe. That review is like insurance. The day you need it, it is too late to buy it.
Where it gets dangerous
Around milestone two or three something strange happens. You catch yourself writing "keep the orders table separate, reporting will be easier later" and you realise three weeks ago you did not have one of those words. Sitting next to a CTO, a founder learns engineering. That is the real free course of this decade.
The danger is the opposite feeling: trusting a green tick. More than once this year I had a build that "passed" and a feature that did not exist for the customer, because I read the summary instead of clicking the button. The CTO is brilliant, and it is also confident when it is wrong. Milestone testing with your own eyes is not optional. It is the job.
Practice
What to do on Monday
- Take the software idea that has been sitting in your notes for a year. Open a coding assistant and send only the architecture prompt above. Stop after the architecture and the stack. No code today.
- Read the architecture back. If you cannot explain it to a colleague in plain language, ask again until you can.
- Write your milestone list, each one something you can click.
- Book a human developer for a security review before anything with payments or user data goes live. Put the date in the calendar now.
What is the one product you would build this quarter if you had a CTO who never said no?