Building Superfile — an AI workspace, from zero to live in 3 weeks
An AI-native workspace that unifies files, notes, and an AI assistant in one product. I led design and co-built the MVP — from zero to live in 3 weeks.
Role
Product Designer & Co-creator
What is Superfile?
Superfile is an AI-powered workspace that combines file management, notes, and an AI assistant — all in one place. Think of it as the product you wish existed when you were juggling Notion, Google Drive, and a chatbot across three browser tabs.
You can pull in files from Dropbox, Google Drive, or your local folders. Write and organize notes. Then ask the AI questions that are actually grounded in your own content — not just the internet.
It’s a small team product. Just me and one developer. We gave ourselves three weeks and shipped something real.
The Problem
Most people I know — writers, researchers, indie founders, small teams — are paying for at least four or five tools that don’t talk to each other. Files live in one place. Notes live in another. The AI you use doesn’t know about either. So every time you need to get something done, you’re the one doing the stitching.
Problem 01
Files, notes, and AI are siloed across separate tools
Problem 02
AI can't reference your own documents and context
Problem 03
Switching between apps breaks focus and costs time
Problem 04
Teams end up paying for multiple overlapping subscriptions
My Role
I was the sole designer on this. But I also went beyond that boundary — I built several pages directly using Claude Code. This wasn’t just a design handoff. It was an end-to-end collaboration between me, the developer, and AI tools, where I stayed involved all the way to production.
Defined the product concept, value proposition, and feature scope
Designed all screens — from the landing page to the in-app workspace
Used Claude for ideation, early research, and writing product copy
Built several pages myself using Claude Code during development
Worked closely with one developer to ship everything in three weeks
Made real-time design decisions during build based on technical constraints
Process
How we actually built this
Week 1
Idea to design
Used Claude to pressure-test scope. Cut ruthlessly. Clarity before pixels.
Week 2
Design to build
Figma and Claude Code running in parallel. I built pages myself. Design met code.
Week 3
Build to live
Edge cases, polish, ship. AI-native from first line to last deploy.
Product
What we shipped
We didn’t try to build everything. We focused on the three things that made the product worth using at all.
01
Unified file management
Connect Dropbox, Google Drive, and local folders. Organize with nested folders, star important files, and preview multiple formats without leaving the app.
02
Rich notes editor
Text formatting, code blocks, tables, callouts, and toggles — with an AI assistant built into the editor. Not bolted on. Built in.
03
AI that knows your stuff
The AI chat can reference your files and notes directly as context. Ask it to summarize a document, draft something based on your notes, or find connections across your content.
Key Design Decisions
Three weeks means you can’t agonize. Every decision had to be deliberate and fast.
One sidebar, three modes
Files, Notes, and AI Chat live in one persistent sidebar. Switching between them doesn't reload anything — your context stays. I tested a tabbed layout first, but it made each mode feel like a separate app. The sidebar made everything feel like one product.
AI chat references your content, not the web
Most AI tools answer from the internet. That's useful, but it's not the gap we were solving. We designed the AI specifically around your workspace — your files, your notes. If the AI doesn't know something, it says so rather than making it up from somewhere else.
Keep the editor opinionated
We could have added 30 block types to the notes editor. We didn't. Text, code, tables, callouts, toggles — that covers almost everything a real user needs. More than that and you're building a distraction, not a workspace. Constraints here were a feature.
Team permissions without the complexity
We needed sharing and access control, but we couldn't let it become a feature that required a tutorial. Spaces with role-based access — admin, editor, viewer — simple enough to set up in under two minutes, powerful enough for a small team to trust.
What I learned
The biggest shift wasn’t the timeline — it was how involved I stayed throughout the build. Using Claude Code to actually ship pages myself changed how I think about the designer-developer relationship. I wasn’t handing off and waiting. I was in the codebase with the developer, making real decisions in real time.
AI tools also changed how I approach early-stage thinking. I don’t do long solo brainstorming sessions anymore. I talk through the problem with Claude, poke holes in the concept, then design. The quality of questions I bring to design is better because of that.
Three weeks is fast. But fast doesn’t mean rough. When the tools are good and the scope is honest, you can ship something you’re proud of.
Where it landed
Superfile is live. It went from a conversation to a shipped product in three weeks — and it’s the kind of product I’d use myself, which is usually a good sign.
Shipped and live at superfile.app
Built by 2 people in 3 weeks
AI-native from ideation to deployment