I Led a Team of 5 AI Assistants to Build an App in 28 Hours

Hey,

You might remember that I’ve written before about how I’m building an app that helps me become a better person. Back then I called it “the thoughts app”. 🙂

Well, this weekend I was building again.

On Saturday, I only had a concept in my head. And of course, I had to buy the domain first. 🙃

By Monday morning, it had been 28 hours.

During that time, I led a team of five AI assistants, each with their own role.

And today? I have a working mobile app on my phone.

My assistants will probably demand longer vacations now… It exhausted me too, I went to bed at 3:00 AM on Saturday. But hey… I’m satisfied.

A New Kind of Team

How is this different from leading a team of people?

It’s different.

Not better or worse. Just different.

Maybe with AI you don’t manage personalities and don’t need psychology knowledge, which helps you understand or execute many things. 😊

But the same applies here as with people:

Clarity and context.

If your instructions are unclear, you get unclear results. If you’re specific about roles and outcomes, the work gets done quickly and with quality.

This is how I structured the team:

Main assistant (ChatGPT) - Project manager who understood the bigger picture
Designer (Google Stitch) - Created all visual screens
Three developers - Each with different strengths: - Claude Opus 4.5 for complex architecture - Claude Sonnet 4.5 for reliable execution - Gemini 3 for fresh perspectives

Process:

Main assistant improved my concept through conversation
Once we aligned on the vision, the Designer joined
Designer created screens while developers provided feedback
Developers converted designs into working code
I orchestrated everything, made decisions, and changed direction when needed

The Pivot

Late in the evening, things started getting complicated with server complexity.

I knew that if I persisted, it would extend the execution. So I wouldn’t finish by Sunday 24:00.

I could have persisted, but I didn’t.

Instead, I listened to my gut feeling and changed the entire approach.

New strategy: 100% local operation.

I assessed that despite changing a large part of the code, this would be better.

No servers. Everything runs on the device.

This was the right decision. The app works better, faster, and more reliably.

What I Learned About Collaborating with AI

1. Specialization matters

Don’t use one AI for everything. Different models have different strengths. Use them strategically.

2. Cross-checking improves quality

Having developers review the designer’s work (and vice versa) caught some issues I would have missed.

3. You’re still the director

AI can execute work brilliantly, but you must provide the vision, make judgments, and know when to change direction.

4. Speed is real

What would take weeks in traditional development happened in one intense weekend.

The Future Is Already Here

Well, this is how I often feel the future that is undoubtedly coming.

The tools are available. The models are capable.

The only question is:

Are you ready to think like a team leader instead of a tool user?

You can try this:

Choose a small project you’ve been postponing. Break it into three roles (strategy, design, execution). Assign each role to a different AI tool. Test what happens when you lead more specialized “agents”.

If you need help anywhere, don’t hesitate to write to me. I’m happy to help.

P.S. I’m already fully testing my first “flows” and assistants in the new app. Now I need to test whether the app and UX actually help me in everyday life. If it’s useful, I’ll share…

Stay in touch! ✌️

Primož