I Built 3 Apps with Claude Opus 4.7 in Under 30 Minutes

Opus 4.7 just dropped, and I didn't want to talk benchmarks. I wanted to test something specific: can it build 3D, visual apps from just a plain English conversation?

No PRD. No wireframes. No design background. Just the CLI, some ideas, and Opus 4.7.

Here's what happened.

The Experiment: Three Apps, Three Terminals, Zero PRDs

I set up three empty project folders and spun up a Claude instance in each one using Semox (a terminal multiplexer that makes managing parallel sessions easy).

Each app got a single conversational prompt. No tech stack specified. No mockups. Just a description of what I wanted with a shared aesthetic direction: dark theme, synthwave feel.

The three builds:

• A 3D cartographer - terrain sculpting, orbit camera, save/export to OBJ
• A 3D voxel office - four AI agents communicating in a visual workspace
• A SaaS landing page - dark theme, orange neon, 3D visual elements

App 1: 3D Cartographer (Built in 8 Minutes)

I dropped a reference map I'd made with Google Flow into the folder and asked Claude to build a 3D renderer/cartographer based on it. Full functionality - terrain brushes, save, export.

Claude chose Three.js on its own. It built a height-map grid terrain with brush tools to raise, lower, and smooth the landscape. Added orbit camera controls, adjustable sun and lighting, save/load via JSON, and OBJ export.

Eight minutes from prompt to working app.

The camera zoom needed some fine-tuning (it only had two positions - way out or way in), but the core build was solid. The synthwave aesthetic with the dot grid, glowing horizon, and dark sky matched the reference perfectly.

App 2: AI Agent Office (Practically One-Shot)

This was the build that surprised me most. I asked Claude to create a 3D office environment where I could plug in my OpenClaw AI agents and visualize them communicating.

Dark theme, amber glow, voxel pixel art style. Four agents at desks in a roundtable setup.

One prompt. One shot. Working 3D environment with communication events flowing between agents.

The agents' models clipped into the furniture slightly, but the fact that it understood "3D voxel office with AI agent communication visualization" from a single paragraph is a real indicator of where Opus 4.7 sits compared to older models.

App 3: SaaS Landing Page (~30 Minutes)

The landing page took the longest of the three at about 30 minutes. I gave it a product concept (OpenClaw Builder - an OpenClaw wrapper for building autonomous AI agents), a buyer persona, and the visual direction.

I also told it to think like a UX/UI designer with 20 years of experience. No assets, no wireframes, no brand guide.

The result was a complete landing page with 3D visual elements, a gamified CTA, and solid dark-theme aesthetics. There was a minor overlap in one section, but for having zero creative assets to work with, I was impressed.

What This Means for Builders

This experiment wasn't about shipping these specific apps. It was about testing a workflow. And here's what I took away:

The idea-to-prototype gap is almost gone. Three concepts that would've taken days of design and development became functional prototypes in a single session.

PRDs aren't always the first step. For production work, absolutely write a PRD. But for validating whether an idea has legs, just describe it and see what happens.

Opus 4.7's visual capabilities are real. Older models couldn't handle this level of 3D/visual work. This isn't benchmark talk - I tested it myself.

Parallel builds are powerful. Running three Claude instances at the same time in Semox felt like having a small development team working in parallel.

Who Should Care About This

Freelancers: Build 3-5 visual prototypes in an afternoon. Show clients what's possible before they commit.

Agencies: Use this as a rapid prototyping tool. Test concepts before investing design hours.

Anyone curious: You don't need a 3D background or Three.js experience. You need a terminal and some ideas.

As I said in the video: no PRD, no 3D visual background, just Claude Opus 4.7, the terminal, some ideas, and some time.

The abilities of your portfolio just 10xed. Now go build something.

Watch the Full Build Session

I recorded the entire process - all three builds running in parallel, the real-time reactions, the debugging, everything. If you want to see exactly how this works, check out the video above.

And if you build something cool with Opus 4.7, I want to hear about it. Drop a comment on the video and tell me what you're working on.

Making AI simple and easy to use. For businesses, entrepreneurs, and creatives.

🎥 Watch the full video