I’ve lost count of the times I typed a big, ambitious prompt into an AI tool and got back something flat and generic. You ask for a 3D experience and you get a gradient background with a spinning cube. Sound familiar?
So this post stopped me mid-scroll. An AI professional showed how to create a full 3D scroll experience with Opus 5.5, and the prompt was just “make it a scroll website”. That’s all. No essay-length instructions and no wall of technical specs.
The catch is the useful part. The original poster says 95% of the work was done beforehand. The prompt didn’t do the heavy lifting. The setup did. I think it’s one of the clearest examples I’ve seen of how working with AI agents is changing, so I’ve laid out the creator’s process as steps you can follow.
What Was Ready Before the Prompt 👇
Here’s what the author had in place before typing those five words:
- A detailed 3D model built with headless Blender
- An explorable, walkable web viewer for that model
- A set of agent skills covering 3D, web, and animation
- A rendered animation made from the creator’s own direction
Once you look at the list, the one-line prompt makes total sense. The agent wasn’t building something from nothing. It was putting together pieces that already existed. Here’s how those pieces stack up, one step at a time.
The Step-by-Step Process
- Build the core asset first. The expert started with a detailed 3D model made in headless Blender. “Headless” means Blender runs with no visual interface, driven by scripts or commands, so an AI agent can create and edit the model through code. Why it matters: the model is the raw material, and everything else depends on it.
- Make the asset explorable on the web. Next came a web viewer where you can walk through and explore the model. Why it matters: it proves the 3D content already works in a browser, so the agent never has to fight that battle later.
- Package know-how into agent skills. The creator put together a set of skills for 3D, web, and animation. Skills are reusable instruction bundles that teach an agent how to handle a certain kind of task. Why it matters: the agent doesn’t have to guess your standards or rediscover your workflow each session.
- Direct the motion yourself. The animation was rendered based on the author’s own direction. Why it matters: creative calls like camera paths, pacing, and mood came from a human. The AI carried out the vision. It didn’t invent it.
- Send the casual prompt. With all of that ready, “make it a scroll website” was enough. Why it matters: the agent had the context, the content, and the skills, so it only had to connect them.
When the context, content, and skills are ready, adding a new feature or creating a new experience is just one casual prompt away.
Why This Approach Works So Well
That quote from the post’s author is the whole lesson. Most of us spend our energy on the prompt itself. We tweak wording, add rules, and pile on adjectives. This contributor flips that around. Put the energy into what the agent can see and use, and the prompt can stay simple.
Think of it like hiring a skilled contractor. If you hand them finished blueprints, the materials, and a crew that knows your standards, you can just say “build the deck” and walk away. If you hand them nothing, no amount of clever instructions will save the job.
It also pays off over time. Once those assets and skills exist, every new idea gets cheaper. A product page, a walkthrough video, an interactive pitch deck: each one is “just one casual prompt away”, as the creator puts it.
How to Apply This to Your Own Projects
You don’t need to work in 3D to borrow this. The same pattern fits almost any creative or technical work with AI agents:
- Gather your context. Keep brand guidelines, project goals, and reference examples in files the agent can read.
- Prepare your content. Have the real copy, images, data, or models ready before asking for the final product.
- Write reusable skills. Turn any process you repeat into a written skill, like “how we format landing pages” or “how we structure animations”.
- Keep creative direction human. Decide on style, pacing, and tone yourself, then let the agent do the execution.
- Test with short prompts. If a short prompt gives you weak results, that’s a sign your setup has gaps. Fix the setup, not the prompt.
Tips & Tricks
A few practical takeaways I picked up from studying this workflow:
- Look for tools that run without a GUI. Headless and script-driven tools like Blender let agents work directly with your files, which opens up a lot more possibilities.
- Build in layers. The expert went from model to viewer to animation to website. Each layer reused the one before it.
- Make each piece work on its own. The walkable viewer was useful by itself. When each part stands alone, the final assembly is way less fragile.
- Treat skills like an investment. Writing them takes time once. After that they save you time on every project.
My Take
I was honestly impressed by how modest the final prompt was. It’s a good reminder that the most impressive AI results usually come from careful prep work that nobody sees. The prompt gets the credit, but the foundation does the work.
If you’ve been frustrated by AI giving you generic output, try spending your next session building context and skills instead of writing longer prompts. You might be surprised how much a supercharged setup can do with five simple words.
Head over to the full LinkedIn post to see the creator’s 3D scroll experience in action. It’s well worth a look.