Shayan · /post build sessions^

✈︎ We're live on Product Hunt. "Build in public" accidentally became a second job. Build. Explain the build. Package the build. Distribute the build. We built The Spiel Engine to turn work sessions into publishable content automatically. Check ↓ producthunt.com/products/the-s…

@ChainZenit @omarsar0 He was running on treadmill thinking: a treadmill is just running with great floor design.

@omarsar0 that is a spicy take, how did you land on that?

Loop engineering is just prompt engineering with great system design.

A treadmill is just running with great floor design.

@adonis_singh ^Trust me bro > Polymarket

I doubt china even has a 4.8 competitor, let alone mythos

@marconoiz Efficiency!

Sell me your startup in one word

@emollick We don't even know your use-case. I've had builds where Minimax 3 outperformed GPT-5.5, but I wouldn't use that alone to claim it's the better model. One person's workflow is not a model comparison. There are too many variables involved.

GLM-5.2 is good but it is not GPT-5.5/Opus 4.8, and even further from Mythos. Yet it is solid & it demonstrates that the open models continue to chase the frontier What is happening is that open weights crossed into GPT-5.2 territory & capabilities at that point are considerable

@ivanfioravanti Model use cases have too many variables to say this with certainty. Benchmarks may put GLM 5.2 in the same intelligence tier, but usage changes everything. A model can underperform in CLI/tooling work, while being excellent at design systems and UI...

Let me iterate this once again: GLM 5.2 is NOT as good as opus 4.8/gpt 5.5 I think is more a little below opus 4.7/gpt 5.4. It's a great model, I use it as daily coding driver, but we need to be honest.

A GPS is just asking for directions with great map design.

@omarsar0 Depends entirely on what you mean by "loops"! A big orchestration agentic workflow can be a loop. A simple Codex prompt "Don't stop until you hit this GOAL with these metrics..." is also a loop. Loop this, loop that, is more hype-based marketing. The meaning spectrum is vast.

in OpenCode v2 all instances of the tui and desktop and web share the same backend so everything is synced by default and resource usaged is minimized no matter how many windows you have open

@_tenZdhon_ lol Tenzin is there 2 types of users for the product? I just looked at the website quickly. Looking at 2 types of CTA's (Download app / book demo), and different pages, i assume there are 2 types of users. But i think you can significantly improve the copy and positioning.

Why is my Claude glazing itself lmao "This is a genuinely sophisticated system" "The architecture is genuinely strong" "This is a remarkably mature AI system" bro wrote the whole code

SpielOS plugin cooking... Stay tuned! V2 is a completely new version... With full rebranding.

@adamheimann I guess that needs a central communication/context hub, as the single source of truth for context and communication management.

Brother. I’m telling you… :) for me I manage and build and debug etc… on an ecosystem not just a single VPS lol. A lot of the time different pieces of the ecosystem are on different remote set ups to suit the need. When im focusing on 1 part locally that plugs into another part i often dont feel like pulling in the other repo setting it all up locally when all need is for the model to have awareness or context of it so it considers it when planning changes. I’ve been having the local model use the terminal to ssh into the server to get context or I act as the “BUS” between the 2 which makes me feel like cuck. lol.

The next billion dollar startup idea: → LangGraph for IDE agents. Orchestrating agents in Cursor / Claude Code / Codex is still messy as fuck. You try chain tasks, split roles, pass context… it's BROKEN. Subagents? Not enough. Skills? Nah... Missing layer = Opportunity^

@kylegawley Yeah all you need to stay poor but pumped!

ChatGPT is all you need to validate your startup idea

@adamheimann Mine is full local for now (MVP), haven't tested remote communications yet. But seems like another solid idea to me for a real app. Communication between local and remote agents! We're onto something here^

@ShayanSpiel Can local and remote ssh agents converse? lol I need that.

@adamheimann Yea^ I'm literally creating an orchestrator for IDE Agents right now for SpielOS! (not a LangGraph for agents, but i had to do it to make my app work)... STILL removing bloat and optimizing after 5th FULL refactor! Developing properly for IDE agents is a nightmare.

@ShayanSpiel Homie. I literally have made an entire system for this it’s insanely bloated and a pain in the ass but it works. This idea is solid.

just tried glm 5.2 trash

WTF did I just tell y'all? A US-Led AI future is a disaster for humanity. Open-weight all the day. And as long as China leading it, in ALL-IN China supporter. I just said this: x.com/shayanspiel/st…

@rackSpreader1 fuck... AI My 1000 followers is now on YOU🫵

Growing on x is so easy. All u gotta do is just say whats on ur mind and mention AI and youll gain a thousand followers

@jxnlco How to make a deterministic script run, or any run, with a command? Meaning it does not interpret the command as prompt first, and immediately run the script... Is there a way?

I’m top 5 Computer Use users at OpenAI Ask me anything.