DesignBrief AI
Building the AI-powered design workflow platform that turns vague client briefs into structured direction, kanban boards, and live design output, from 0 to 1 as a solo founder.
- Role
- Founder & Product Designer
- Timeline
- 2025 - Present
- Platform
- Web (Desktop-first)
- Year
- 2025
The Problem
Freelance designers lose hours before a single pixel is placed. A client sends an email that says "make it pop" or "something clean but bold", and the designer either spends two days going back and forth asking questions, or makes assumptions that send the project sideways in week three.
The brief is the most important document in any design project. It is also the most consistently broken part of the workflow. Generic AI tools like ChatGPT or Notion AI are not built for this problem. They produce text. They don't understand design vocabulary, brand personality, visual direction, or the specific gap between what a client says and what a designer needs to hear.
DesignBrief AI was built to own that gap entirely. Not just to translate the brief, but to run the entire design workflow that follows it, from structured direction to a kanban board to AI-generated page output, with human approval at every step so the designer stays in control.
My Role
I conceived, designed, and built DesignBrief AI from scratch as a solo founder. My role covered everything: product strategy, the full UX and UI design, the 21-point brief translation framework, the four-agent AI architecture, copywriting, and hands-on implementation guidance for the engineering layer. This is not a concept, it is a live, functioning product. I made every product decision, resolved every design tradeoff, and shipped every feature.
Team
Solo founder. All design decisions made independently. Engineering implementation in React 19 + Vite + Express + Supabase, with Claude API powering the AI layer.
Research & Discovery
Before building anything, I needed to understand why every existing solution was failing designers at the brief stage. I looked at how freelancers actually receive, process, and respond to client briefs, the real workflow, not the idealized one.
The brief is never the real brief
What clients send is not a brief. It is a collection of feelings, references, and half-formed ideas wrapped in business language. The actual brief, what the designer needs, has to be extracted, structured, and sometimes invented from signals the client didn't know they were giving. No tool was doing this extraction work. They were treating the client's words as the input when the client's intent is the input.
Designers waste the most time before they start
The research showed a consistent pattern: the hours before the first deliverable are the least structured and the most expensive. Follow-up emails, clarification calls, misread tone, wrong visual direction, all of it happens because the brief translation step is entirely manual and entirely informal. Three to five hours per project, on average, before a wireframe exists.
Generic AI tools create false confidence
Designers who tried using ChatGPT to interpret briefs reported a specific problem: the output sounded right but wasn't actionable. It produced design-adjacent language without design intelligence. A senior designer reads a brief and immediately identifies what's contradictory, what's missing, and what the client actually wants versus what they said. Generic AI doesn't do that. It rephrases.
The workflow breaks at handoff between tools
After the brief is interpreted, designers move to Notion for tasks, Figma for design, Linear or Jira for project management, and email for client communication. Every one of those transitions costs time and context. No tool owned the full pipeline, brief to board to build. That gap was the product opportunity.
Human approval is not optional in AI workflows
The strongest signal from talking to working designers: fully autonomous AI output is not trusted. Designers want control at every meaningful decision point. They want AI as a fast, opinionated collaborator, not an autonomous executor. This shaped the core architecture: every kanban card requires human approval before the AI builds the next one. The pipeline only moves when the designer says so.
Key Decisions
The Question
Should the brief translator be a chat interface or a form-plus-output layout?
The Decision
Two-panel layout: input on the left, live structured output on the right.
A chat interface felt familiar but created a problem: the output was buried in conversation history. Designers need to reference the structured brief throughout the project, not scroll back through a chat thread. The two-panel layout keeps the input visible for editing and the output always accessible. It also signals that this is a professional tool, not a chatbot.
The Question
Should the AI kanban builder be fully autonomous, generate everything with no interruptions?
The Decision
Human approval required before every card advances. The pipeline only moves when the designer approves.
Full autonomy was technically simpler to build but it destroyed designer trust in testing. When the AI made a wrong assumption on card three, it had already built on top of it by card seven. Mandatory approval gates cost a few extra clicks per card and save hours of rework. The product is built around this principle: AI moves fast, designer stays in control.
The Question
How should the design system be extracted: Figma import, in-app editor, or curated library?
The Decision
Automatic extraction from the translated brief, stored as a 9-section design system panel per project.
Figma import required a connection most early-stage clients don't have. An in-app editor was scope creep that would take months to build properly. Extraction from the brief itself was the right answer: color intent, typography behavior, spacing philosophy, component style, motion feel, and visual language are all present in the brief translation output. The design system panel pulls those signals and attaches a shared token set to every kanban card so AI-generated pages stay visually coherent across the project.
The Question
Should the client intake form be a standard multi-field form or something different?
The Decision
One question at a time, client-facing, delivered via a branded shareable link.
A standard multi-field intake form gets abandoned. Clients open it, see fifteen fields, and call instead. The one-at-a-time format reduces cognitive load, increases completion rates, and feels professional, not like a Google Form. The branded link means the designer controls what the client sees. The AI enrichment pass that runs after submission expands thin answers, resolves contradictions, and extracts implicit signals before the brief translation engine ever sees the input.
The Question
Should the AI use a single large prompt or a multi-agent architecture?
The Decision
Four-agent pipeline: Brief Interpreter, UX Strategist, Design Director, Presentation Engine.
A single large prompt produces comprehensive output that is not opinionated. It describes everything without deciding anything. Senior designers don't work that way: they eliminate before they add, they make bets, they name the one non-negotiable. Breaking the translation into four specialized agents forces depth at each stage. The Brief Interpreter extracts signal. The UX Strategist makes the conversion bet. The Design Director cuts first then builds. The Presentation Engine makes the output skimmable in under five minutes. Each agent is constrained to its role, no agent does what the next one is responsible for.
Flow Diagram
Final Design
The shipped product covers five core surfaces: the dashboard and project library, the brief translator with structured output, the kanban AI builder with approval gates, the client intake form pipeline, and the design system panel. All desktop-first, with the glass effect applied selectively to sidebar, modals, and floating panels.
All Screens
Key Flows
Core flow: brief to structured output
Client intake pipeline
Kanban board
Outcomes
21
point brief translation framework
Built from scratch, no existing framework to reference
4
agent AI pipeline powering the translator
Brief Interpreter, UX Strategist, Design Director, Presentation Engine
10
custom animated SVG empty states
Every empty screen has a unique illustration with brand purple palette
5
core product surfaces shipped
Dashboard, translator, kanban builder, intake form, design system panel
What I Learned
Building DesignBrief AI taught me that product design and product thinking are the same thing when you're the only designer in the room. Every decision I made on the UI had a product consequence. Every product decision had a design implication. You can't separate them. The biggest lesson: the brief translation output took three full iterations before it was actually useful. The first version was correct but unreadable. The second was readable but not opinionated. The third version, with the four-agent architecture, the brand personality quadrant, the brief quality score, and the visual structure that makes it skimmable in under five minutes, is the version that actually serves a designer on a real project. I would have invested more time in that output quality earlier. The pipeline was technically sound from month one. The output quality took until month four. That gap cost me early feedback that would have been valuable when the architecture was still easy to change.
Next Project
Web Design