The Year Amara Carried an Idea, and the Six Weeks She Built It
By Yemi Ogundairo on
The idea arrived in the back of a tailor's shop in Surulere, on a Saturday, with a wedding three weeks away. Amara watched her tailor turn a notebook sideways and work backwards through two years of pencil measurements, looking for hers. He did not find them. He measured her again, apologised twice, and wrote it on a fresh page that would also, eventually, be lost. On the bus home she had the whole thing worked out: a small app where a tailor keeps every client's measurements, and the client can pull hers up on her phone. Then she spent a year not building it, until the evening she sent us a message that read, more or less, can I build an app with AI if I cannot code? Amara is not one person. She is about forty of them, and their story is the same story, so we have given the forty of them one name.
The year of almost starting
The year was not lazy. It was three false starts.
The first was a course. Thirty-two lessons, bought on a Sunday with real enthusiasm, abandoned at lesson four because lesson four was about arrays and her tailor was still losing measurements.
The second was a quote. A developer she was introduced to listened, nodded, and named a figure. It was not an unreasonable figure for the work he was describing. It was simply the end of the conversation.
The third was an afternoon. She typed build an app with AI into Google, found her way to a chat window, and asked it to build her a tailoring app. It gave her forty ideas, three feature lists and something that looked like a plan. By six o'clock she had a document. She still did not have a product, and she quietly concluded the problem was her.
What the AI did, and what it refused to do
It was not her. She had asked a generic question and received, correctly, a generic answer.
Here is the honest division of labour. AI writes the code, drafts the layouts and produces the copy, and it does all of it in seconds. That part is real, and it is why someone with no programming background can now build an app with AI across a few weeks of evenings, when the same product used to need a team and a budget she did not have.
What it does not do is decide. It does not know who the product is for, which feature earns its place, or when something is good enough to put in front of a person who is not you. Ask it for an app and you get an essay. Ask it for one named screen, holding one set of information, letting a person do one thing, and it is startlingly good.
So the work was never the typing. It was the deciding. What follows is the order she did it in.
How to build an app with AI, the way it actually went
Six weeks. The first two produced nothing you could look at, and they are the reason the last four worked.
Week one: one sentence, and twenty-three things crossed out
She wrote her list first: twenty-six features, including a fabric marketplace, a delivery tracker and a rating system for tailors nationwide.
Then she wrote the sentence. Who uses this, and what one thing do they get done. Hers was: a tailor saves a client's measurements, and the client can see her own on her phone.
Twenty-three of the twenty-six features did not survive that sentence. Nothing was lost, they went on a second page marked later. What remained was four screens. If your version one is fifteen screens, you have not finished cutting, and the reason to be strict is not modesty. An unfinished product teaches you nothing. A small finished one teaches you everything, because real people touch it and do things you did not predict.
Week two: the screens nobody sees
She drew before she built. Paper first, then Figma. It felt like a detour when the AI could generate a page in seconds, and it saved her the most time of anything she did.
Nobody films week two. Every video about how to build an app with AI opens at the moment code appears on a screen, which is why so many people believe the drawing is optional.
Draw each screen three times: empty, when someone new arrives and nothing is in it yet, full, which is the case everyone designs for, and wrong, when the connection drops or a person types something odd. Drawing the empty state is what caught her first real mistake. A brand new client has no measurements at all, so her main screen, the one built around a list of them, was the wrong first screen entirely. Fixing that with a pencil took a minute. Fixing it in week four would have cost an afternoon and broken two other things.
She also renamed every button to the words her tailor would use. Save Measurements, not Submit.
Week three: one path, end to end
One route through the sentence, working from start to finish, and nothing else. Open the app, add a client, save six measurements, see them listed.
Four habits made the AI useful rather than impressive. Ask for one screen at a time, describing what it shows and what the person does on it. Give it the shape of the information, so a measurement has a client, a date, a body part and a number in inches, because vague input produces vague code. When something breaks, paste the actual error text, never your description of it. And keep a short note of every decision made so far, then paste it back in when you open a fresh conversation, because the model remembers nothing about yesterday.
She was approving code she could not read, and that is workable. Judge it by behaviour, not by how it looks. Does the measurement still exist after a refresh. Does a blank name get refused. Does the list show the right client.
Weeks four and five: the parts that make it real
Then the Thursday. She showed her tailor the working thing on his counter, he typed in a client, closed the tab to answer a call, and it was all gone.
You can build an app with AI in a weekend. You cannot make it survive a refresh in a weekend, and three things are what separate a demo from a product.
Information that survives, which means a database, in practice one service you connect to and a short setup. Her Thursday failure was exactly this: beautiful screens attached to nothing.
A way for people to be themselves, which means sign-in. Her second scare came here, when a friend testing it saw another client's measurements. Do not build this part yourself, use a service made for it, because accounts and passwords are the one area where your mistake lands on someone else.
A real address, which means a domain and hosting, so it becomes a link you can send to a person. Hosting a first version is usually free at that size, and the domain is the small recurring cost.
The week most people quit is week two
Amara nearly stopped in week two, and not for a technical reason. Week two is the week you have nothing to show, the drawing feels like procrastination, and the idea you have carried for a year seems safer in your head than in front of somebody.
What saved it was not motivation. It was that she had already told her tailor Thursday.
Three things do that job, and none of them require a course. Pick a finish date and say it out loud to someone whose opinion you mind. Book a demo with one real user before you have built anything, because nothing sharpens scope like a person expecting to see something on a specific day. And work in fixed sessions instead of when inspiration turns up, because two hours twice a week finishes products and waiting to feel ready does not.
Hold yourself to that and you can build an app with AI on your own. Plenty of people do, and it costs them nothing but the evenings.
If you would rather not do it alone
Some people want the Thursday built in, and someone who has done this before looking at the work every week. That is what DSGN Launchpad is: six weeks to build an app with AI from your own idea, two live sessions a week, with a designer and an engineer checking what you make as you make it. You finish with a live product, a database, sign-in and a real address, and it stays yours. The fee is ₦20,000 and applications are on the DSGN Launchpad page. If you are still comparing routes, the trainings page lists the full range of Gr8QM programmes.
Either way, start where Amara started. One sentence, and a notebook that keeps losing measurements.