Lived near Munich and still miss the lakes and the Alps.
So I had Claude build a walkable forest of the region in Three.js. Fed it real geo data, photos of the shore, and the actual tree species (spruce, fir, beech) and pictures of a hut.
Then made it keep comparing its own renders against the references and fixing what didn’t match.
First serious attempt at Three.js with agentic workflows.
Still rough around the edges, working on improvements.
You can walk it here: https://t.co/xRuAhoah2r
@luckeyfaraday It’s always about the next barrier. Every step we get to think of larger, bigger problems to solve. We will know how to think at a size we never imagined before.
@MeticulAi Yes. I don’t know why people are obsessed with “one-shotting” stuff. That just signals lack of creative direction. Anyone can do that. Add some soul to your creation, even with AI. Be patient!
https://t.co/Yf5bbRunWn
One-shot-maxxing: “Fable just created this in a single prompt”…
That’s the mark of the meat proxy and I don’t understand it.
When we were building code by hand, the iterative process of it was crucial to the enjoyment. The decisions on every cycle, the steering of the creation.
With AI we can work in the same way. But yet, somehow, people like to “remove themselves” and think it’s impressive that Fable or Opus one-shotted something. Yes, it’s impressive as a technical demo but it’s not building something.
What do we want AI for? To replace us or to augment us?
Don’t one shot. Build. Enjoy the ride. Influence the flow, let your soul shine through your steering.
@Alex_Vainshtein It’s looking great already! I think the water and the shadows give away a bit but the vegetation looks amazing. How did you achieve the level of detail there?
I’ve been working with something similar but my results are quite different. https://t.co/WzYfNmo5d6
Lived near Munich and still miss the lakes and the Alps.
So I had Claude build a walkable forest of the region in Three.js. Fed it real geo data, photos of the shore, and the actual tree species (spruce, fir, beech) and pictures of a hut.
Then made it keep comparing its own renders against the references and fixing what didn’t match.
First serious attempt at Three.js with agentic workflows.
Still rough around the edges, working on improvements.
You can walk it here: https://t.co/xRuAhoah2r
@starlight_ai@that_john_doe@CtrlAltDwayne I love to see non programmers start building things like this with AI. We just need a vision, ideas and creativity and tokens to build something! Keep it up!
@Unnato_0x Man, that’s amazing! Looks really good!
AI + threejs is becoming such a great stack to create games, scenes and worlds you can interact with.
@realivann Yes! I love how AI is basically deleting the barrier of entry for these things. Creative people will have the time of their lives with these tools.
@CtrlAltDwayne Give it reference material. You can even use Codex / Gemini to generate synthetic data. Tell it to replicate it in Blender / threejs / Unreal. The secret sauce is that the model has something to mimic and understand. Without it it’s just generic AI slop.
@sonuv_dev Yes! The whole point of this workflows is that now you get a world to explore, to adjust, to tune into a story and feeling. A world you can interact with. Fable is reallly impressive.
@BenjaminPopp4 I feel the pain, mate. Some people have a kink for complex and long forms. I literally think some folks derive pleasure from designing this and looking at it. Unfortunately, they work at the Finanzamt.
@inline_bundle@hakimieiqbal What you mean? Everything. Even the agentic creation process. It takes thought and consideration to orchestrate AI agents in a systematic, reliable and consistent way.
@dani_avila7 6/10 to 10/10 on 80M rows is the number that matters. I’m hustling at a startup and these issues are our daily grind and the real cost is when a lock hits production.
A temp DB gets difficult with multi GB setups but it’s definitely an interesting approach.
@nex_ify I love this. Professionally I see this being useful in on-point setup for automatic incident monitoring, for example.
On the other hand… “Find the best time to buy and sell Nvidia stock and call me when something goes wrong”.
Would have been wild the last few days.
@RoundtableSpace Multi-agent squads are underrated. I worked a fintech and we used a similar approach to understand legal documents at a scale.
This is how you stop over-optimistic single agents high-fiving their own work all the time and actually achieve good results.