Designing and shipping an AI workspace for better proposals.

The difficult part

Proposals are where independent designers spend their most anxious hours for their least certain returns. The raw material already exists: past proposals, past pricing, what won and what sank. But it lives in fifteen documents across three tools, so every new proposal starts from a blank page anyway.

The design problem: make AI drafting feel like reusing your own judgement, not generating generic filler. The moment an AI proposal reads like an AI proposal, it costs you the job.

The product story

The workspace centres on the proposal editor, where AI suggests scope, structure, and pricing drawn from your own history rather than from the internet’s average proposal. Around it: a lightweight CRM for the client pipeline, an approvals step for teams, pitch-deck generation for the meetings that follow good proposals, and shared document links so you know when a client actually opened the thing.

Analytics closes the loop. Win rates by proposal type, where clients stop reading, which sections correlate with a yes. The pitch is simple: stop guessing what converts.

Design system and interaction thinking

This one is also an implementation story. I designed the product, then built it: React, TypeScript, and a Cloudflare deployment, with AI tooling in the loop from first wireframe to production. The design system stayed deliberately small, a working set of documented tokens and patterns, because a solo product punishes every component you cannot maintain.

Building it myself changed the design. Interaction details that would have died in a handoff document, like optimistic saves and inline AI suggestions that never steal focus from typing, survived because I could carry the interaction decisions directly from design into the build.

Result

A working preview with proposal generation, CRM, analytics, approvals, and document sharing, designed and built end to end.

What changed in my thinking

Shipping solo changed how I judge effort. Visual details are easy to revise; state management, edge cases, permissions, and empty states are where the product becomes expensive. I now design those parts earlier.