SelfishPrimate

Plainify emerged from a recurring friction point I observed while working with AI coding assistants. Designers and developers often have a clear mental model of what they want to build, but struggle to communicate that vision in a way that produces quality output from AI tools like Claude, Cursor, or Bolt. I built Plainify to bridge that gap.

The Challenge

Working with AI coding tools introduced a new communication problem. Traditional design handoffs weren't built for machine consumption, and ad-hoc prompts rarely captured the full scope of a product idea. Teams would describe the UI but omit the tech stack, or nail the color palette but forget to specify navigation patterns.

During a "Vibe Designing" workshop I organized for my design team, this friction became even more apparent. Every participant spent significant time crafting prompts, only to realize they'd left out critical details. The iterative rewriting slowed down the creative momentum we were trying to build.

Research & Discovery

I started documenting what made AI outputs successful versus problematic. The pattern was consistent: well-structured, comprehensive specification files dramatically improved initial results and reduced back-and-forth corrections. Scattered requirements across Figma files, Notion docs, and Slack threads led to fragmented AI outputs.

This research led me to develop PLAIN (Product Language for AI Notation) — an open-source specification format that consolidates everything a product needs into a single markdown file. The format covers 14 distinct sections, from project overview and value proposition to design direction, component inventory, and technical architecture.

Design Objectives

• Create an interface that eliminates blank-page anxiety when defining product specifications
• Support both quick ideation and detailed customization workflows
• Maintain user privacy by keeping all data client-side
• Generate output compatible with any AI coding assistant
• Make the tool accessible without accounts or sign-ups

Solution: Two Distinct Workflows

The application offers two complementary approaches based on user context and time constraints.

Relax Mode targets rapid ideation. Users input a brief product description — as simple as "A recipe app that suggests meals based on available ingredients" — select an AI model, and receive a complete 14-section specification within seconds. This mode works well for early exploration or when time is limited.

Custom Mode provides granular control over every section. The interface walks users through curated options, contextual examples, and intelligent defaults. Design style, color palette, typography, tech stack, UI library — each choice is guided rather than open-ended. For any decision point where users feel uncertain, an "Let AI decide" option delegates the choice to the model.

Output & Integration

Both workflows produce a clean markdown file named {project-name}-plain.md. Users can preview the output, make inline edits, copy to clipboard, or download directly. The file drops into any project directory and works immediately with AI assistants — simply instruct the tool to read the spec and begin building.

Visual Reference Analysis

For users who already have design direction established, Plainify accepts image uploads. Screenshots, mockups, or wireframes are analyzed by the AI to extract color palettes, layout patterns, and component structures. These visual insights feed directly into the generated specification, ensuring alignment between existing design work and the final output.

Privacy Architecture

Data privacy shaped core architectural decisions. The application runs entirely client-side with no server-side storage. Specifications persist in browser local storage only. No accounts, no sign-ups, no data collection beyond basic analytics for usage patterns. Product ideas remain on the user's device throughout the entire workflow.

Technical Implementation

The generation engine uses Anthropic's Claude Haiku models. Users choose between Haiku 3.5 for faster output or Haiku 4.5 for more nuanced results. The markdown format was intentional — it's readable by humans, parseable by AI, trackable by Git, and editable in any text editor. No proprietary formats or vendor lock-in.

Key Features

• 14-Section Framework: Comprehensive structure covering project overview, design direction, tech stack, data models, and more
• Dual Workflow Support: Quick generation via Relax Mode, detailed customization via Custom Mode
• Visual Reference Processing: Upload mockups for AI-analyzed design extraction
• Inline Preview & Editing: Review and modify output before exporting
• Client-Side Processing: All data stays in the browser, nothing stored server-side
• Universal Compatibility: Markdown output works with Claude, GPT, Cursor, Bolt, and other AI tools

Watch the Video

See Relax Mode in action — from product idea to complete specification in seconds.

Reflection

Building Plainify reinforced a principle I've observed throughout my design career: the quality of output depends heavily on the quality of input. AI coding assistants amplify this relationship — vague prompts produce generic results, while structured specifications yield coherent, buildable applications.

The project also validated a workflow I call "Design in Code" — iterating on interfaces directly in the codebase using AI assistance rather than traditional design-to-handoff pipelines. Plainify itself was built this way, with Figma reserved only for branding assets while all UI work happened in the terminal with Claude Code.

Links

• Plainify App
• Plainify on Medium
• Plainify on YouTube
• PLAIN Format on GitHub