How I build 5 apps a day

I'm pretty sure I build faster than 99.9% of the world right now.

Subtle flex lol. 💪

So I'm going to do two things today:

So I’m evolving Ajay’s Quest into AJI Labs. It feels like a truer manifestation of the one man innovation lab I want to create for myself. How much can one human being create in this world?

(Before some of you ask: AJI is a play on AGI → Artificial General Intelligence)

I build SO many things. I’ll release a few as weekly drops. Mostly useless. Hopefully entertaining. Occasionally useful by accident.

The first drop: Ajay’s List, a self-improving, auto-updating list of things I use and consume.

Super simple, and that’s the point. An idea doesn’t have to become a startup to deserve to exist. (I’ve been making trailers for each launch too, because it’s fun.)

JUST TODAY (a Saturday), I built an anime dating sim, remote team games, a text-based game, Ajay’s List, the AJI Labs site, a Whatnot stream overlay and more. The only limits right now are compute, intelligence and whether my wife wants to hang.

The speed comes from AI. But getting something I actually want to put my name on takes more than asking it to “build an app.”

Here’s my system:

Most of my ideas start as one sentence.

“It would be funny to make a dating sim where you date AIs.”
“Make a web app so remote teams can play games together.”

That’s enough to start, but I ask for ten different directions: different formats, different audiences, different reasons someone might share it. Then I pick. AI is pretty good at going far when you ask for divergence. I’ll often research, scroll Pinterest and feed it more inspo.

Same with design. Before building the actual app, I get four to ten visual directions on one page. I can point at one and say, “That one, with the colours from this one.”

In essence, I’m building a decision tree of hundreds of possibilities to find the ‘BEST ONE’.

You might struggle to describe your taste, but you can usually recognise it pretty quickly when you see it.

Every app needs some of the same things. A newsletter signup. A “Made by AJI Labs” badge. Social previews. Deployment setup. A check that it works on your phone.

I’ve started putting those into a shared kit. The design changes for each app, but I don’t need to work out how an email signup works five times a day.

The tenth app should take less effort than the first. If it doesn’t, I’m repeating work I should have saved.

In practice: shared environment variables for my local builds, shared MCP tools, and one-click deploys to Vercel, Fly and whatever else I use. I can sometimes go from idea to deployment in minutes.

Some actual feedback I’ve given AI:

“Stop doing black backgrounds.”
“Too Claude.”
“Make a simple website, then a button to go 3D as optional.”
“Keep most of my copy verbatim unless there are grammatical mistakes.”

If I keep saying the same thing, it belongs in the instructions the AI reads before starting. Otherwise I’m teaching it the same preferences every session.

AI has defaults. Leave it alone and your apps start looking suspiciously related. I want each drop to have its own identity, and writing down what I like (and what makes me immediately go “nope”) gets there faster.

The agent that built the app shouldn’t be the only one deciding whether it works.
I use separate agents to check it with fresh eyes.

One tests the experience: click the buttons, play the game, try it on a phone, open the forms, try to break things. Take screenshots and actually look at them.

Another checks security and costs: are any keys exposed? Can someone spam a paid endpoint? Are there limits on usage?

ALSO TEST THINGS YOURSELF. Nothing worse than shipping slop you’ve never seen. I test on my laptop and phone, and most of my input now is visual feedback.

Sometimes building the app took an hour, and getting it live took another.

Wrong account details. Deployment errors. Domain nonsense. Some tiny setting I’d forgotten existed. So when something goes wrong, I write down the fix. There’s now a repeatable order for getting the code online, adding production keys, connecting the domain and checking the live site.

That live check matters. Working on my laptop is lovely, but that’s not where you’re going to use it.

One drop burned through $400 of AI credit before I noticed. Not a disaster, but enough to make me think about what happens if people actually use these things.

So anything calling a paid AI model gets usage limits: a cap for the whole site, limits per person or room, and limits on how much someone can submit.

I also estimate what 1,000 or 10,000 visitors might cost before launching (IMPORTANT). I’d quite like people to enjoy the stupid little game without it becoming an exciting financial event for me.

I choose what’s worth making. I pick the direction. I decide whether it looks good, feels fun, and deserves to go out into the world. That’s where I want to spend my time.

Every repeated correction, forgotten check and painful launch is a chance to make the next build easier. Over enough builds, that’s how I get to five apps a day.

And AJI Labs is where I’ll find out what happens when I keep going.

Until next time,

HAPPY BUILDING

Ajay