And The Winner Is…

I ran the test everyone's been arguing about all week.

Same simple prompt...

Three flagship models.

Codex GPT-5.6 Sol Ultra. Claude Code Fable 5. Grok 4.5.

The task: build me a multiplayer browser game.

Codex won.

And it was not even CLOSE.

Sol Ultra built me EMBERBOUND...

A 1-4 player online co-op brawler.

Think Golden Axe with sorcery.

Landing page.

Lore.

Character creator.

Combat.

Multiplayer warbands.

It designed the character sprites...

All four heroes with genuinely different weapon silhouettes, not palette swaps.

And watching it work is the wild part.

The new Codex Desktop app runs FLOATING windows with a live view of everything it's doing...

Mine was working in 2 browsers at the same time.

You literally watch your agent play-test its own game. 🤯

The thing that used to scare people about agents was the black box.

"What is it doing right now?"

Now the answer is...

Look.

Honestly?

I think I'll just release the game for free if you guys want to play it. 😂

But here's the takeaway for YOUR stack...

Benchmarks are marketing.

Tests are truth.

One same-prompt bake-off taught me more than every leaderboard published last week.

Steal the method: one real task from your business, run it on each model you're considering, compare outputs side by side.

Twenty minutes. Zero guessing. 🫡

Did You Know?

Golden Axe — the game Emberbound is inspired by — came out in 1989...

meaning the AI that rebuilt the genre in a weekend is younger than the save files. 😅

🗞️ The Tip AI News 🗞️

Now... a confession and a rant 😅

Yes, I spent my weekend using frontier AI to build a video game.

Do as I say, not as I do. lol

Because last week I saw a post that summed up everything wrong with the agent community...

A guy proudly scaled his setup to 1,400 agents.

Screenshots.

Cache math.

Genuinely impressive engineering. 🤝

Then one comment under it said everything:

"If it doesn't make dollars, it doesn't make sense."

That's the post.

The AI community has a tinkering problem.

Everyone wants to scale to 1,000 agents...

Build a "mission control" to watch agents that don't make money...

Or build a "Jarvis" so their computer takes 5 minutes to tell them something they could see on screen in 5 seconds.

But ask one simple question...

"What's your ROI?"

Suddenly it gets real quiet.

You don't need 1,400 agents.

Not 100. Not even 10.

You need ONE.

One agent that understands your business, moves the needle, and generates value while you sleep.

1 agent making you $10,000/month beats 1,400 making you nothing.

So here's what I'm thinking about building...

A prompt that analyzes your Claude, ChatGPT, or Gemini chat history...

figures out who you are, what you actually do all day, and where you waste the most time...

Then tells you The One Agent YOU need.

And builds it for you.

Over to You...

That's it for today.

Go run your own bake-off...

And then ask your agents what they earned last month. 😂

Jeff J. Hunter 
Founder, AI Persona Method | TheTip.ai

“AI Money Group” to Learn to Make Money with AI