
AI/ML Engineer | Building AI Agents,Deep into CS, Math & Apple APIs #AI #MachineLearning #VibeCodeKing #Python #Supabase #SaaS #LLMs #TechEntrepreneur
Joined August 2015
- Tweets230
- Following853
- Followers443
- Likes317
Pinned Tweet
Been quietly rebuilding my 2D fighting game since the GPT-6 Astra demo. I added myself into the game the black guy teleporting 😭
The physics are better.
Combat is faster.
More collisions.
More destruction.
More blood.
It actually feels like a fighting game now.
But the next version is where this gets out of this world!
Since TypeSafe released Jev, I’ve been experimenting with a different architecture:
You describe the fighter you want:
“Fast lightning swordsmen that fights like Sasuke Uchiha. Teleports behind enemies and fights like Minato from Naruto. Heavy aerial combat who fights like avatar Aang.”
The game figures out the character archetype, abilities, animation/physics requirements and which generation skills it needs then compiles that into an actual playable fighter.
Not just an AI image.
A generated character you can actually fight with.
And I want to take the same architecture into story mode:
Generate your fighter.
Generate your abilities.
Generate your story.
Play it.
Basically: a fighting game that can expand around the player.
What a time to be alive where i am building my dream game ( from my childhood) while working full time with 2 under 2 kids.
😭😭.
Demo( Wanna bet i cant make it... see waitlist and find out 😏):
veilbreakgame.online/
Yooo I just used Codex Cloud to change an iOS app LIVE on my physical iPhone 🤯
No rebuild. No restart. The UI changed in seconds.
I’ve been dreaming of building mobile apps on the road. Just watched it happen.
𝗔𝗻 𝗔𝗜 𝗵𝗼𝘁𝗲𝗹 𝗰𝗼𝗻𝗰𝗶𝗲𝗿𝗴𝗲 𝘀𝗵𝗼𝘂𝗹𝗱 𝗸𝗻𝗼𝘄 𝘁𝗵𝗲 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝗰𝗲 𝗯𝗲𝘁𝘄𝗲𝗲𝗻 𝗮 𝗿𝗲𝗾𝘂𝗲𝘀𝘁 𝗮𝗻𝗱 𝗮 𝗰𝗼𝗻𝗳𝗶𝗿𝗺𝗲𝗱 𝗯𝗼𝗼𝗸𝗶𝗻𝗴.
A guest asks for late checkout. Before saying “done,” the system must check hotel policy, live room availability, any fee, and the guest’s consent then verify the reservation actually changed.
That requires more than a prompt and some tools.
• Configuration sets defaults and mandatory rules; live availability is a runtime fact.
• A compiler checks workflow contracts and versions the workflow before it runs.
• Events carry the request through checking, consent, booking update, and confirmation; “Update requested” is different from “booking updated.”
• Jev’s Choice, Noul, and Score can identify intent, route the request, or prioritize it they’re decision inputs; the runtime still decides permitted actions.
• The runtime checks authority and idempotency and reconciles before retrying if the booking provider times out.
• The system records workflow version, decisions, and tool outcomes to explain what happened.
The same foundation can handle housekeeping or restaurant requests by adding workflow configuration, tools, and event handlers.
I’d judge it by requests completed correctly, latency, cost, duplicate changes, unresolved outcomes, and how safely the workflow can be changed.
For a hotel, would you start with late checkout or housekeeping?
🤖 Made with AI
I think now we are at a time where we can build an agentic fighting game generator using unreal engine , custom engine on top of for fighting mechanics, Blender mcp . Creativity is the moat now
Grok 4.7 is out now!
x.ai/news/grok-4-7
Benchmarks claim it’s really skilled at electrical engineering not surprising considering that is majority of elons minions
I’m testing my new 2D sprite game engine and codex literally told me this to keep hands off the keyboard while it works? The heck they think they talking to like I don’t pay you.
The more autonomous coding agents become, the less permissions can be a prompt. Permissions have to become infrastructure.
unity.com/blog/unity-plugin-…
Yep… let’s go crazy yall Astra for demo Claude for masterpiece
This demo from astra 6 is just the beginning i will use weaker models to enhance the rest and apply my new physics engine to it! what a time to be alive! finally get to create my 2d sprite fighting game generator let me know if something like that sounds fun!
Chika Ohaya retweeted
Fable 5.1 vs GPT-6 Astra for a Pringles ad
Round 2
Made on HeyOz
Fable 5.1 vs GPT-6 Astra for a pringles ad
Made on HeyOz
Readers added context they thought people might want to know
Large language models like Fable 5.1 and GPT Astra are purely text generating transformers and cannot generate videos or images this post is purely for engagement farming.
platform.claude.com/docs/en/models…
developers.openai.com/api/docs/model…
GPT-6 Astra helped me build this custom fighting game.
( inspired by smash bros! )
Then I packaged its physics engine as a reusable Codex skill. ( By standing on the shoulders of genius PHD physics legends! )
Air combos. Energy blasts. Platforms that retain their vibration.
Still experimental. Next: bringing the engine into my anime-inspired game.
Below shows the outputs after one shot prompt! imagine after refining ! Yoo we are the next era its about to be fun as heck!
i keep thinking AI is becoming less of an ML problem and more of a software engineering problem
like the models are already crazy good. the hard part now is everything you gotta build around them so they actually work in the real world
what context do they get? what can they touch? what happens when they mess up? can they recover? can you verify what they did?
that’s basically the harness
and i think that’s where software is going
by 2027 the UI might honestly be the smallest part of the product lol. most of the actual software is gonna be this giant harness underneath it controlling models, agents, tools, state, all of it
lowkey harness engineering might just become software engineering
I really think this token maxing mess getting overhyped 😂
Like yeah these AI coding agents hard, but dang, we still gotta build good software.
AI software still gotta be good software at the end of the day.
I thought I was behind until I have people younger than me like siblings and cousins asking me. What is Claude? That’s when I was like oh I’m I’m really up here.
I think software in 2027 is going to get visually simpler while becoming insanely more complex underneath. We’re moving from apps with AI features to full AI harnesses running agents, events, memory, permissions and verification in the background.
The user might literally just see: “Everything is good. 2 things need you.”
Vertical ai SaaS → vertical agents → vertical harnesses. I think that’s where this is going.









