Chaitanya Malik
FOUNDER, AGE OF AI
A while back I posted on LinkedIn that AI has made coding so easy that even people who can't code can now ship real software. A lot of people disagreed, loudly.
So I made a video to prove it. I built a working app and put it live on AWS in about half an hour, and I didn't write a single line of code myself. The AI wrote all of it. I just told it what to do and checked its work.
This post is the how-to version. Do these steps and you can build and ship your own app the same way. Here's the full build:
In the video I build Fintech Insights — a small mobile app where people drop short fintech takes, tag them by category, like them, and filter the feed. Don't get attached to the app though. The method is what matters, and it works for almost any small thing you want to build.
You'll need:
Read these before the steps. Get them right and the rest is easy.
Don't just say "build me a social media app." Tell it exactly what features you want, lock down the tech, and tell it not to add extra stuff. This is the actual prompt I used:
lets create a social media app with these features:
- User enters a display name
- User writes a fintech insight
- Category dropdown: Payments, Banking, Lending, Trading, AI in Finance, Startups
- Post appears in the feed
- Like button
- Filter by category
- 5 default sample posts
It should be mobile-first, in Next.js v16, DB = SQLite.
Keep the UI world-class. Do NOT go beyond the features scope.
Let's plan this. ultrathink
A tight feature list stops it from over-building, and naming the tech (framework, database, mobile-first) keeps it on track. That "do not go beyond the scope" line matters more than you'd think.
This is the step most people skip, and it's the one that saves you the most trouble. Don't let it start coding yet. First, make it write a plan:
Write a PLAN.md first. Don't build anything yet.
Then make it go back and check its own plan:
Verify the plan. Did you miss anything? Fix it.
When I did this, it found a couple of features it had missed and added them before writing any code. A mistake in the plan takes seconds to fix. The same mistake in a half-built app takes hours.
Tell it to build in stages and test after every stage:
Build it phase by phase. After each phase, run the build and the tests,
then do a visual check before moving on.
Get the small things working before you build the big things. If you skip this with AI coding, the errors pile up fast and you end up in a mess.
Also tell it to make a fresh folder for the project so nothing else gets in the way.
You shouldn't be typing anything here. A good agent will write all the code, write its own tests and run them, run the build and lint checks and fix whatever breaks, and even open a browser to look at the app on a phone-sized screen to make sure it looks right.
Your job is to watch, keep it pointed at the goal, and make the call when it asks you something. You can jump in and change direction any time.
Before you deploy, run it and use it like a normal user would:
Start the dev server.
Open it (an incognito tab helps), add a name, post something, like a post, filter the feed. Make sure the whole thing works. The AI checks itself, but you should check it too.
Anyone can run something on localhost. The real thing is getting it live, and I let the AI handle the deploy as well.
First, keep your keys safe. I used AWS credentials that expired in one hour, and I didn't let the agent see the actual access key and secret. Keep those in files it can't read.
The agent doesn't know every detail of every cloud service, so let it look things up:
Search the internet for how to deploy a Next.js app on AWS, then do it.
Use region ap-south-1 (Mumbai).
And expect to change your plan partway through. I started with AWS Amplify, but it's serverless and its filesystem is read-only, so SQLite wouldn't run. So we tried other things:
Which service you land on isn't the point. The point is that when you go down a wrong path, the AI catches it and fixes your direction if you keep talking to it. That back-and-forth is honestly most of the job.
When the AI is running commands on your real cloud account, you're the one responsible. If you don't understand a step, ask before you approve it. One wrong move can cost you real money.
A real app, built and live in about half an hour, with no code written by hand. The job isn't really "writing code" anymore. It's telling the AI what to build, checking what it did, and making sure it actually works.
If you haven't started using these tools yet, start now. The people who have are moving a lot faster than the people who haven't.
Can you really build an app without writing code? Yes. In the video I build a working app and put it live on AWS in about 30 minutes, and the AI writes every line. Your job is to give clear instructions and check the result, not to type code yourself.
What tools do I need? An AI coding agent (I use Claude Code), Node.js, and an AWS account for the deploy step. That's it. A clear idea of what you want to build matters more than any tool.
How much does it cost to deploy? In this build, hosting on AWS Lightsail with CloudFront came to about $10–11 a month. Serverless options can be cheaper, but they didn't fit the SQLite database I used here.
Is AI-generated code good enough for real use? For a small, well-scoped project, yes, as long as you make it write tests, run the build, and verify each phase. For a real production system, go slower, phase by phase, and review the code more carefully instead of rushing.
Which AI agent should I use? I used Claude Code, but the method works with any capable coding agent. What matters is that it can plan, write and run its own tests, and use tools like a browser to check its own work.
▶ Watch the full build: youtu.be/MOddCradhK8
More at ageofai.co.in.