artan

Full-Stack Engineer & AI SaaS Architect | Building scalable products & AI-powered tools @razegrowth

[artan @shoupw_]

@alejmaestre Exactly. The interesting part isn’t the game itself it’s compressing weeks of development into roughly two hours. The real question now is whether this speed is repeatable across different projects or only works for specific demos.

[artan @shoupw_]

I built this playable FPS with GPT-5.6 Sol in ~2h.
Not a mockup: movement, shooting, weapon switching, health/ammo, capture points, animations, physics and real time multiplayer.
Next: test the same with grok 4.5 and fable 5
I’ll publish the prompt, cost, interventions and code.

[artan @shoupw_]

We have way more builders than actual users right now.
Everyone’s building agents to generate content, code, and environments… but who’s left to use it?
AI writing for AI, coding for AI. Meanwhile real people are just doomscrolling.
Where’s the real demand supposed to come from?

[artan @shoupw_]

@aiseomastery Exactly. The impressive part isn’t one feature, it’s getting movement, combat, physics and game state working together in roughly two hours. Next I’m giving other models the identical brief and time limit.

[artan @shoupw_]

@shiplogged Fair criticism. The “AI default aesthetic” is becoming a fingerprint. This run prioritized functionality within 2h, for the next comparison, every model gets the same visual reference and I’ll score UI originality separately.

[artan @shoupw_]

@chrislaupama Thanks, sounds interesting!

[artan @shoupw_]

@ScottFree369 Yeah I’m planning to test Kimi K3 too, especially after seeing the frontend results. Appreciate it bro 👍

[artan @shoupw_]

We have way more builders than actual users right now.
Everyone’s building agents to generate content, code, and environments… but who’s left to use it?
AI writing for AI, coding for AI. Meanwhile real people are just doomscrolling.
Where’s the real demand supposed to come from?

[artan @shoupw_]

I built this playable FPS with GPT-5.6 Sol in ~2h.
Not a mockup: movement, shooting, weapon switching, health/ammo, capture points, animations, physics and real time multiplayer.
Next: test the same with grok 4.5 and fable 5
I’ll publish the prompt, cost, interventions and code.

[artan @shoupw_]

It uses a lightweight Node WebSocket server on the local network. Clients send movement, aim and fire events; the server owns lobby, teams, health, respawns and combat validation, then broadcasts state to everyone. Remote movement is interpolated client side to keep it smooth. Testing is done with multiple browser/device clients connected to the same LAN server.

[artan @shoupw_]

Source code for the GPT-5.6 Sol version:
https://t.co/MSQG81EGHH
Clone it, break it, improve it. I’ll share the identical brief before testing the other models.

[artan @shoupw_]

@MrAhmadAwais These toy game comparisons test who produces the prettiest demo, not who solves the hardest engineering problems

[artan @shoupw_]

Kimi K3 is cooking
#1 in Frontend Code Arena, ahead of Claude Fable 5. Open weights dropping July 27 too.
Moonshot AI just made things interesting again 👀

[artan @shoupw_]

The biggest AI coding shift is not “AI writes more code.”
Across ~400k Claude Code sessions, people made most planning decisions; Claude made most execution decisions.
The senior engineer role is shifting: set constraints, design evals, review outcomes, own production risk.

[artan @shoupw_]

AI coding agents are not killing expertise. They are compounding it.
Anthropic studied ~400,000 Claude Code sessions. Expert sessions reached verified success more than 2× as often as novice sessions.
AI lowered the cost of execution.
It did not lower the value of judgment.

[artan @shoupw_]

Three frontier models. Nearly identical context limits.
GPT-5.6 Sol: $5 / $30
Claude Opus 4.8: $5 / $25
Claude Fable 5: $10 / $50 Fable costs 2× more than Opus on input and output.
The question: does it produce 2× fewer failures on your workload?

[artan @shoupw_]

The biggest AI coding shift is not “AI writes more code.”
Across ~400k Claude Code sessions, people made most planning decisions; Claude made most execution decisions.
The senior engineer role is shifting: set constraints, design evals, review outcomes, own production risk.

[artan @shoupw_]

AI coding agents are not killing expertise. They are compounding it.
Anthropic studied ~400,000 Claude Code sessions. Expert sessions reached verified success more than 2× as often as novice sessions.
AI lowered the cost of execution.
It did not lower the value of judgment.

[artan @shoupw_]

“Best AI coding model” is now a meaningless sentence.
GPT-5.6 Sol leads the Coding Agent Index: 80 vs Fable 5 at 77.2. But on SWE-Bench Pro, Fable 5 wins: 80.0% vs Sol at 64.6%. The model didn't change. The test did.
Benchmark the workflow, not the logo.

[artan @shoupw_]

As a software engineer who has built countless products and softwares, this is the first time I’m interacting with an AI tool that genuinely feels more intelligent than me.
GPT-5.6 Sol is on another level.
The way it takes a task from 0 to 100 managing agents, deep reasoning, testing, debugging, fully understanding the problem and delivering complete solutions is actually scary good.
This isn’t just a better model.
It’s the kind of thing that makes you seriously think about the future of our industry and humanity.
Absolute masterpiece.
#GPT5 #OpenAI #AI