I'm a father of two, 7 and 12. They are obsessed with Roblox, especially Rivals.

Like a lot of parents, we did not love it. We tried the usual things: block it, limit it, set timers." It became a daily battle, a lose-lose situation.

So I flipped the problem.

Instead of fighting what they loved, I decided to lean into it, but with a twist.

Why just play an FPS when you could build one together?

My kids became the PMs. Claude and I became their engineer.

I was shocked by how fast we moved. We picked a name, Cooked and started building.

Get in. Lock in. Get cooked!

In the first 3 hours, we had an actual FPS we could play together in the browser.

Next day I invited my friend's two kids over. Suddenly, I had four PMs.

We had 3 laptops, 1 iPad and 1 iPhone.

So in 30mins we added mobile support and could play all 5 together.

The kids would ask for features:

"Can we have a rocket launcher that knocks people off the map?"

"The knife should stab faster when you hold it."

I know almost nothing about FPS design, so they were the experts. They explained what they wanted. We turned that into prompts. Claude would implement it.

We would reload the page and play it 10 mins later.

It's the most fun I've had building software in years. My kids are proud of it.

# Here's how Cooked is cooked.

The architecture

- It's just a web page. TypeScript + Three.js. - No game server. Multiplayer runs over a P2P WebRTC mesh. - Supabase is only the matchmaker. It handles presence and the WebRTC handshake. - It is very cheap to run. Static site plus a few Cloudflare edge functions for TURN relay. - Bots fill empty rooms. Each bot gets a random loadout and play style.

# What worked and what didn't

I have been writing software for 20 years. For the last 9 years, as CEO of Krisp.ai, I have not had much time to program myself. I missed it.

Claude changed it. I obsessed with this side project for the last 2-3 weeks.

First, building a project with your kids is a great way to bond. It gives you shared interests and real quality time.

Second, I wanted to understand the limits of Claude today and how we can apply this at Krisp. This helps me stay grounded.

Opus 4.8 is extremely good at architecting application systems, researching references, and implementing algorithms.

I would tell it things such as "research how the weapon system is designed in Roblox, Diablo, Fortnite and come up with a proposal on how to do this in Cooked". It would come up with an impressive backlog and I would just say "Go".

Where it was really weak was UI design. It simply couldn't detect/see obvious design problems.

I’ve tried to implement a loop with a game designer agent, UI designer and system architect. While the research, reasoning were top-notch, the system failed due to bad design taste and the inability to see images properly and spot problems there.

Interesting fact. It really failed miserably when I challenged it to design and draw "hands" that hold the weapons. I also have struggled with building decent maps. It lacks imagination and fails even after providing a lot of guidance.

Another interesting fact. This morning, I gave the same problem of “drawing a hand” to Fable and it drew it from first-shot, for all weapons. Very impressive.

# Why I'm sharing this

Two reasons.

One: if you're a parent fighting the same screen-time battle, consider flipping it. Building the thing they love, with them, turned a source of conflict into a shared creative project. My 12-year-old now understands feature scoping. My 7-year-old has strong opinions about explosion sizes.

Two: AI coding tools have crossed a line. A dad who is not a game developer, together with four kids, built a real multiplayer browser FPS that their friends actually play.