FEATURED AI WORKFLOW / ZENCHI AI
A brief becomes
a reusable assistant.
Inside Agent Studio: saved instructions, business context, real responses and conversations you can return to.
Public walkthrough · No sign-in required here · Development evidence captured 12 October 2026
THE PROBLEM & MY CONTRIBUTION
Stop rebuilding the same brief.
A useful business assistant needs a clear role, the right context and a way to resume work.
Through AI-assisted development with Codex, I helped shape and build Zenchi AI's Agent Studio: an interface for configuring an assistant, saving its instructions and reference context, choosing a model and returning to saved conversations.
My work spans the product requirements, user journey, implementation direction, output review and functional verification. The example below shows a Small Business Marketing Planner and its actual response to a sample business question.
- 01
Define the assistant
Set its purpose, audience, model, instructions and conversation starters.
- 02
Save the context
Keep an assistant revision and approved reference notes for reuse.
- 03
Run a real question
Use the saved instructions to guide the model's response.
- 04
Return to the work
Recover saved conversations instead of starting again.
ACTUAL SCREEN / ACTUAL RESPONSE
A planner that asks before it assumes.
The input
I run a local wellness business. Ask me the three essential questions before drafting a plan.
The generated response
The assistant asks for:
- The specific product or service being marketed.
- The ideal audience or customer.
- The available marketing budget and time commitment.
The decision behind the workflow
Collect missing commercial context before generating a marketing plan. Save the instructions so the approach can be used again.
What this demonstrates
Prompt design, task boundaries, a reusable assistant interface, real model execution and saved work. It does not establish a customer growth result or autonomous campaign execution.
FUNCTIONAL VERIFICATION
Beyond a generated answer.
A separate isolated test checked that saved context and conversations survived the workflow.
- Authenticated assistant save.
- A real provider response using the supplied reference context.
- Usage-based credit settlement.
- Replay without an additional charge.
- Conversation recovery after a new login.
Checks recorded on 12 October and reviewed in the 15 October functional audit. They are dated development evidence, not a fresh full-product test or a production launch sign-off.
View the captured persistence-test screen
The same approach, applied to growth.
Explore the audit experience, SEO reporting, interactive proposals and client results.
View supporting work ↗