The Unorthodox Angle
The proposal is where creative firms lose. Not because they can't write, but because the work of writing proposals is structurally different from the work of doing the work. The AI bridges that gap — but only if the firm knows its own brand well enough to describe it to a machine.
Architecture firms have a problem that nobody talks about: the work is good, the proposals are bad, and the gap between the two is where careers go to die.
I know this because I built a tool to fix it. coDesign Compositor is a proposal authoring tool for architecture and design firms. It creates branded, dual-output RFP responses. The input is project information and firm branding. The output is a document that doesn't look like it was assembled at 2 AM by someone who would rather be designing.
I mention the 2 AM thing because that's when most proposals get written. Not during business hours. Not by the senior designers. By whoever drew the short straw, at the last minute, using a template from 2019 with the wrong logo.
The Gap
Creative firms have a structural problem. The people who do the creative work are not the people who write the proposals. The creative work is what they're good at and what they want to do. The proposal is what they have to do to get the work that lets them do the creative work.
This gap exists in every creative profession. Architects write proposals. Designers write proposals. Photographers write proposals. Every creative who wants to eat eventually faces a document that asks them to justify their vision in terms a procurement committee can score.
The proposal is where creative firms lose. Not because they can't write — many of them write beautifully. They lose because proposal writing is structurally different from creative work. It requires a different mode of thinking, a different register of language, a different relationship to the work. And it has to happen on top of the work, not as part of it.
What the Tool Does
coDesign Compositor takes the firm's brand — their visual identity, their voice, their project history — and uses it to generate proposal responses that are consistent, professional, and on-brand. Dual output means it produces two formats from the same content: a formatted PDF for the client and a plain-text version for the submission portal.
The Pinnacle Health Systems proposal — Detroit HQ Consolidation — is a real example. A healthcare system consolidating their Detroit headquarters. The RFP asks for technical approach, team qualifications, project timeline, and fee structure. The firm has all of this information scattered across previous proposals, project archives, and the senior partner's head. The tool pulls it together.
This is not AI writing the proposal. The AI is doing the assembly — pulling from the firm's existing materials, applying the brand template, structuring the content for the RFP format. The creative decisions — what to emphasize, what to leave out, how to position the firm — those are still human. The AI handles the infrastructure so the human can handle the strategy.
The Pattern
This is the same pattern I wrote about in the photography pipeline: the AI's job isn't the creative work. It's the work around the creative work. The metadata. The formatting. The assembly. The tedious, necessary infrastructure that nobody became an architect to do.
But there's a second pattern here that's more interesting: the brand as a data structure.
When you set up coDesign Compositor, you create a brand profile. The profile includes visual identity (colors, fonts, logo treatment), voice (tone, vocabulary, sentence structure), and project history (past work, client types, expertise areas). This profile is a structured representation of the firm's identity. It's a brand bible, but it's machine-readable. And because it's machine-readable, every proposal that comes out of the tool is automatically on-brand.
Most firms have a brand. Few firms have a brand that's structured enough to be enforced automatically. The tool forces the discipline of defining the brand in terms a machine can use — and that discipline makes the brand stronger, even for the proposals that don't use the tool.
Beyond Architecture
The proposal problem isn't unique to architecture. It's universal in professional services.
Law firms write proposals. Consulting firms write proposals. Engineering firms write proposals. Every professional services firm that competes for work through RFPs faces the same gap: the people who do the work aren't the people who write the proposals, and the proposals suffer for it.
The pattern generalizes. Any professional services firm with a definable brand, a project history, and a repeating proposal format can use this approach. The AI assembles the proposal from the firm's existing materials, applies the brand, and produces a consistent output. The human reviews, adjusts, and submits.
The interesting question isn't whether AI can write proposals. It can. The interesting question is whether the firm's brand is structured enough for the AI to use. And that question — what does our brand look like as data? — is one that most firms haven't answered.
The Uncomfortable Truth
Here's what I learned building this: most firms don't know their own brand well enough to describe it to a machine.
They have a logo. They have a website. They have a vague sense of who they are. But when you ask them to define their voice — their sentence structure, their vocabulary, their tone — they struggle. When you ask them to categorize their project history — by type, by client, by expertise — they have the information but not the structure.
The tool forces this structure. To use coDesign Compositor, you have to create a brand profile. To create a brand profile, you have to define your brand in terms a machine can use. And that process — defining your brand as data — is valuable whether or not you ever use the tool again.
The proposal that writes itself (almost) is the end product. The brand as a data structure is the real deliverable. And most firms need the deliverable more than they need the proposal.