Shayan

✈︎ 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…

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…

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^

@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.

Spend however much you want on marketing. Experiment and explore. But in the end, it’s word of mouth that matters most. Nothing else is as meaningful, or durable.

@haider1 10x that shit, let's get GLM 53

USA is within months of losing the lead in AI by the time openai releases gpt-5.6, we'll hopefully have one of these models ready to beat it: glm 5.3 qwen 3.7 / qwen 4 minimax 3.5 / m4 deepseek v4.5 / v5 kimi 2.8 / kimi 3 dario made the West look scared of its own technology

@supabase Etc etc.

What's the longest prompt you've ever written?

@thdxr Are we 🪿 now?

when you're down just remember the goose was not valued

Are we now just middle-men selling what AI made?

@synthwavedd All in supporting China on this. US-led AI future is a DISASTER!

All in on open-source models

@BratDotAI That's a problem. HUGE problem that i obsessively have too. But nothing good comes out of surface level executions. To succeed you need to focus deeply on 1 single idea, for long enough that it starts giving fruits. We all know that, but incapable of doing it.

Sometimes I wish I had started my indie dev journey before AI took off. Maybe I would’ve focused on one idea, written every line of code myself, and spent more time figuring out distribution. Now AI makes me want to build 10 different things… only to realize after validation that most of them may not even be worth building 😅

If it only works in your setup, it’s not a product! ^it’s a hack that works in one environment.

@petergyang That's trump^

One thing that still annoys me about even the best AI models and agents: They’ll give an overconfident opinion. Then if you push back even a little, they’ll usually reverse it right away. It makes me feel like they weren’t thinking that hard in the first place. A good human teammate might change their mind too, but they’ll at least explain why or have a more principled approach to their thinking. Maybe there’s a way to fix this with a system prompt?

@RamaswmySridhar Nice! I also analyzed GLM vs other frontier costs in different scenarios. Full breakdown here: spielos.xyz/GLM5.2-Cost-Br…

Follow-up to my GLM vs Opus thread: let's talk cost. We ran 103 dbt tasks x 3 trials on each model. Same harness, same tasks. GLM: 860M tokens Opus: 439M tokens That's ~2x. But the "why" is more interesting than the number.

@0Sunnycity2 @MiaAI_lab It really depends on the scale of the project. That's why i said i do the planning specifically with a stronger model. The more complex the project is, the more you need a stronger model for plans. But well, there are workarounds like breaking down tasks etc. To decrease costs.

@ShayanSpiel @MiaAI_lab My point is: DS flash is not good enough to plan and if you have to spend so much time futzing with your harness so that it is(OP opinion), why not just use a better model to plan and save the time futzing with your harness

Why DeepSeek v4 Flash is my #1 choice Because it gives you 90% of frontier-level performance at 35–100x lower cost than Opus 4.8/GPT 5.5. And you can run it locally quite easily. GLM 5.2 is better but DS4F is 4-5 times cheaper. Never disappointed me. Still #1 in OpenRouter

@JayneshKasliwal Don't d

Getting married in 2 days Any tips ?

@siddharthwv Need 250 for subs

Laptop + Wifi + Good coffee + Delusional optimism Can make you a millionaire.

Better harness ≠ more features. Feature count is a vanity metric. A harness is the system that controls execution: Context → Tools → State → Agents → Output The winner isn't the product with 100 features. It's the one with the best runtime and orchestration.