I was at the gym when a potential client emailed me. She wanted to move forward, but one small detail in the proposal needed to change before she signed.
Normally, that meant waiting until I got home, opening Figma, finding the correct file, editing the copy, checking the layout, exporting another PDF, and sending it back.
This time, I opened Codex from my phone. Because the proposal had been built from source files with approved copy, reusable components, and clear design rules, Codex could find the page, make the change, regenerate the PDF, and prepare the email. I reviewed everything before it left the studio.
By the time I got home, the proposal was approved and the contract was signed.
That ten-minute revision finally clarified where AI was useful in my design studio. I didn't need it to invent the work. I needed it to carry approved decisions through a task that would otherwise interrupt the work only I could do.
A ten-minute job can take over an entire afternoon
The original proposal process wasn't difficult. That was exactly why it had survived for so long.
When a client requested a revision, I opened the Figma file, changed the copy, checked the affected pages, exported a new PDF, removed the outdated version, and replied. On a quiet day, the whole thing might take ten minutes.
But a studio is rarely dealing with one isolated ten-minute job.
A proposal revision lands between client feedback, scheduling questions, invoices, and whatever creative problem I was trying to hold in my head before the notification appeared. Each interruption feels too small to matter. Together, they can take over the day.
Design needs a different kind of attention. I am making decisions about something that doesn't exist yet, and I can't do that well in the five minutes between messages. The cost of the proposal change wasn't the edit itself. It was leaving the creative problem and having to find my way back.
That was the friction worth solving.
Why I stopped asking AI to begin with the design
Before this workflow worked, I had spent months trying to make AI useful during the most creative stage of a project.
I asked it to explore concepts, critique typography, find references, and help with brand directions. It could organize information and give me more options. The work became less convincing when I asked it to decide what the brand should be.
The beginning of a brand project is full of ambiguity. A founder may know how the business should feel without having the language for it yet. The audience, offer, positioning, and visual references are still being interpreted. There can be a hundred reasonable directions and only a few that feel right for this particular business.
When those decisions are missing, AI fills the gaps with familiar patterns. The answer can look polished while still feeling as if it could belong to almost anyone.
The proposal revision gave it a very different job. The client had identified the exact sentence. The approved layout already existed. The studio's voice, typography, spacing, photography, and export rules had been decided. Success was easy to describe: change this copy, preserve everything else, generate the correct file, and stop for my approval.
Once the direction existed, AI became much better at carrying it forward.
The first version was slower than opening Figma
The first proposal I built with Codex was probably the least efficient proposal I made that month.
I supplied the strategy, copy, photography, and page structure, then spent several hours correcting the first build.
Some instructions were too loose. Layouts that looked fine alone felt inconsistent across the document. Copy that fit one proposal became awkward when a different project needed another paragraph.
Opening an old Figma file would have been faster that day.
The return appeared later, when the same task came back with a different client, another scope, or one sentence that needed to change before someone could say yes.
That is worth saying plainly because AI workflows are often presented as instant time savings. The useful ones usually require slower work first: decide what should stay consistent, document it, test the exceptions, and correct the places where your instructions are not nearly as clear as you thought they were.
The prompt was the least important part
It would be easy to turn this story into a post about the exact prompt that revised the PDF. That prompt only worked because the business had already made the important decisions.
The workflow became useful because the important decisions were already documented:
- the proposal structure and content hierarchy
- Jaia Studio's typography, spacing, colour roles, and page rules
- reusable sections and components
- approved photography and brand assets
- the studio's voice and positioning
- examples of what a finished proposal should look like
- boundaries around what could change and what needed review
A logo, two fonts, and a colour palette wouldn't have been enough. The system needed to explain how the brand behaved when copy ran long, a service changed, or a page needed a different kind of proof.
This is the difference between owning brand files and having an AI-ready brand and design system. One gives the tool assets to find. The other gives it decisions it can reuse.
Jaia Studio now builds that missing layer for established health and wellness teams through an AI-ready brand and design-system engagement. The work starts with one repeated production problem, strengthens the brand rules underneath it, and tests a controlled workflow before a larger implementation is recommended.
What does an AI proposal workflow actually change?
Both workflows begin with a real conversation and a recommendation shaped around the client.
| Before |
With the Codex workflow |
| Decide the scope and write the proposal |
Decide the scope and write the proposal |
| Find the right Figma file |
Open the proposal source |
| Replace and reflow content manually |
Apply approved content to defined components |
| Check every page for layout changes |
Review the generated document and exceptions |
| Export and rename the PDF |
Generate a consistently named PDF |
| Attach the file and write the reply |
Review the file and prepared reply before sending |
The AI proposal workflow didn't remove the thinking. It shortened the production path between an approved decision and a client-ready file.
For a small studio, that distinction matters. Saving ten minutes is useful. Protecting the uninterrupted hour around those ten minutes is the bigger win.
What should AI handle in a design studio?
In this workflow, Codex can find the source, apply a defined change, rebuild the document, generate the PDF, and prepare the delivery email.
I am still responsible for what to recommend, how the project should be positioned, what the proposal needs to communicate, whether the information is accurate, and whether the final work is strong enough to send.
The design system carries decisions that should remain consistent. I review the moments where a rule is not enough.
That boundary matters even more when client information is involved. A working interface is not proof that a workflow is safe, accurate, or ready to operate unattended. Private information needs appropriate protection, and anything leaving the studio still needs a responsible person to approve it.
AI can complete steps. It can't own the relationship, the risk, or the judgment behind them.
Where can an AI workflow still fail?
The proposal system is useful because its limits are clear.
Missing information can produce a wrong assumption. Unexpected copy length can expose an untested layout. An outdated source can generate a perfectly formatted version of the wrong information.
A revision can also stop being production and become strategy. Changing a date is straightforward. Responding to a client objection requires understanding the concern and deciding what the proposal should say.
No amount of automation removes the need to notice that difference.
Its job stays narrow: prepare a strong, review-ready version without taking over decisions it was never qualified to make.
Is your business ready for an AI workflow?
After the gym revision, I started looking for other parts of the studio that might benefit from the same approach. I use four questions before turning a repeated task into an AI workflow.
Does it happen often enough to deserve a system?
Building a workflow has a cost. If the problem appears once a year, a manual fix may be the better decision.
Could I explain the rules to another person?
If the outcome depends on information that only exists in my head, the workflow isn't ready. The first job is documenting how the decision gets made.
Is there a reliable source of truth?
The tool needs approved copy, current assets, working components, and examples it can trust. Outdated information only creates a faster mistake.
Can someone review the result before it matters?
The safest early workflows create drafts, options, or review-ready files. They don't publish or make sensitive decisions without oversight.
The proposal passed all four tests. Asking a tool to create a brand from one vague prompt doesn't.
For health, wellness, and fitness businesses, these boundaries matter beyond design quality. Proposals, service pages, educational materials, and campaigns all shape trust. A faster draft is only useful when the source is current, the claims are accurate, and someone qualified still approves what the public sees.
This is one reason branding matters more as AI makes polished output easier to produce. More content does not automatically make a business clearer or more credible. The system underneath it determines whether that output still feels like the same business.
Start with the interruption you keep repeating
The most useful AI workflow in your business may not look impressive in a demo.
It may prepare a document, organize feedback after a call, update a recurring presentation, or assemble a landing-page draft from approved content. Its value comes from removing a real production bottleneck while leaving consequential decisions with your team.
The proposal didn't get revised from the gym because AI suddenly developed better taste. It worked because the thinking had already been turned into a system.
If your team is already experimenting with AI but every client-facing result still needs to be rebuilt by hand, the missing piece may not be another tool. It may be a brand and design system detailed enough for the tools you already have to follow.
See how Jaia Studio builds an AI-ready brand and design system around one repeated workflow.
