Moaz1

AI directed engineering. Game development.

[Moaz1 @MoazUnique]

GPT 6 Astra High on the same bpydev MCP took 8m 02s and 0.98M tokens, and it was the only one that used the MCP's own render tools.
https://t.co/cvfy7CJLAg

The results are genuinely great, I don't think I will ever use the classic MCP again lol Thanks to @michaelgold

[Moaz1 @MoazUnique]

This MCP is crazy. Same Blender brief, Opus 5 med in Claude Code vs GPT 6 Astra High in Codex.

bpydev + Opus: 5m 49s, 1.01M tokens
lab + Opus: 8m 02s, 1.68M tokens

Both Opus runs: 15% of Claude Pro 5h limit
Astra alone: 22% of Codex Plus

Full results: https://t.co/9OQtGoctQD

[Michael Gold @michaelgold]

Building on my enhanced Blender MCP (as lot of you asked for): https://t.co/fw370ohMsM is open source.

Give an agent a reference image and it drives @Blender from the CLI

Write Python, render headless, inspect, repeat.

Astra hit >99% image similarity w https://t.co/fw370ohMsM on BlenderBench

[Moaz1 @MoazUnique]

@michaelgold mate. whatever you did to cook that repo, I am very grateful you've done it I will share some interesting results shortly :)

[Moaz1 @MoazUnique]

@chetaslua gotta yoink this idea real quick

[Moaz1 @MoazUnique]

If the Trump Admin is actually stopping AI people from doing bad, why didn't they stop OAI-HF?

Or can they only suspend models after release?

And how fast would they stop a rogue swarm taking over the internet via botnet?

[Chubby♨️ @kimmonismus]

Holy, sh*t: Trump is really mad at Dario!

There is now slow-down or pause going on with him.

To quote him: "'Treasonists, Traitors, and Leakers, BEWARE!'".

Id say that settles it. No one is going to slow down now.

[Moaz1 @MoazUnique]

@khushiirl I believe this has always been the case prior to AI being good at coding. solo devs would never easily compete against organisations with big budgets

[Moaz1 @MoazUnique]

@__soragoto__ a very good example of when to use hardcoded conditions as well

[Moaz1 @MoazUnique]

@craigweiss @H_K_8472 this web traffic should all be humans as agents would hit API endpoints directly instead of rendering the web page - unless some bot scrapping the page did load it up though. but it should be a fraction of the traffic I believe.

[Moaz1 @MoazUnique]

@grimcodes @joshinflo can you try crosstalk? it lets agents from different harnesses coordinate.
set one model as leader, others as peers, they’ll collaborate on larger tasks. just keep the scope big so they don’t over do on small stuff.
link in bio

[Moaz1 @MoazUnique]

@grimcodes @joshinflo it will be non-interactive (claude -p)
this will be a headless CLI session that codex should add a monitor for and wait for it’s output.

[Moaz1 @MoazUnique]

I'll stop using t3 code because @theo never resets the usage limits.

[Rana Hanocka @RanaHanocka]

Claude + Thrixel Skills built this entire Roblox game from ONE prompt.

Everything was built autonomously:
• Every 3D asset
• Every game mechanic
• All logic & scene setup

Today, we’re expanding Thrixel Skills to include Roblox. Excited to see what you build! 👇

[Moaz1 @MoazUnique]

@RanaHanocka @amarjit_uk I've got to this a go!

[Moaz1 @MoazUnique]

Token mogged

[dax @thdxr]

"the largest"

[Moaz1 @MoazUnique]

@thsottiaux you would eat the competition if you actually fix the usage limits that everyone is complaining about

[Moaz1 @MoazUnique]

Spent 12 hours with Opus 5 on ultracode trying to fix a bug with an UE game, and it couldn’t. It claimed it was fixed multiple times & visually confirmed. Gave it to Luna Max and it took 2 hours to get it fixed I want to fully switch to Codex Pro but I’m afraid of limits.

[Moaz1 @MoazUnique]

++ @thsottiaux SOS

[Moaz1 @MoazUnique]

Spent 12 hours with Opus 5 on ultracode trying to fix a bug with an UE game, and it couldn’t. It claimed it was fixed multiple times & visually confirmed. Gave it to Luna Max and it took 2 hours to get it fixed I want to fully switch to Codex Pro but I’m afraid of limits.

[Hybrid 🏃🏾 @markfenner]

We live in bubble on this app.

A lot of us are running agents, building workflows, creating skills, comparing harnesses, connecting tools, and etc..

Go ask one of your friends what Kimi or GLM is. They will probably look at you like a madman.

OpenAI’s consumer research found that 49% of ChatGPT messages are still people asking for advice or information.

Another 40% are requests to complete a task.

Meanwhile, 70.2% of sampled Codex users had assigned at least one task estimated to take a person more than an hour.

I’m sure your timeline makes it feel like everyone is running agents and orchestrating workflows.

They are not.

Do not confuse being surrounded by early adopters with being late.

You are early. Keep learning and keep building.

[Rana Hanocka @RanaHanocka]

Built this chaotic kitchen simulator entirely with Claude + Thrixel, an agentic 3D creation platform.

Thrixel: Generated the editable 3D assets
Claude: Assembled the scene in three.js & wrote all game logic.
1 /goal prompt + Goal to Game Skills + 3 follow-ups
More👇