Most people use AI like a blank chat window — re-explaining context every single session. Here's the architectural fix: how Projects and GPTs work together to create an AI that actually knows your work.
Here's the workflow most people repeat every day. Open ChatGPT. Type a prompt. Explain your product, your users, your constraints, your priorities. Get an answer. Close the window.
Next session: blank slate. You explain everything again.
This isn't an AI problem. It's an architecture problem — and it has a fix that most designers and product teams haven't discovered yet.
The Distinction That Changes Everything
ChatGPT has two systems that most people use interchangeably — or not at all. They're not the same thing, and they're not substitutes for each other.
A Project is the memory layer. It's a persistent workspace where you upload files — research reports, PRDs, personas, meeting notes, business metrics. The AI reads them and holds them in context across every conversation. Your history survives session close. You stop starting from zero.
A Custom GPT is the intelligence layer. It's a specialized AI persona you design — with a specific role, a specific way of reasoning, and specific output formats. Think of it as a consultant you've briefed: you defined their expertise, their communication style, their guardrails, what they should never do.
One stores your world. The other knows how to analyze it.
A GPT without a Project is a brilliant consultant who wakes up with no memory every morning. A Project without a GPT is a well-organized filing cabinet with no intelligence.
— Roy Villasana
What Happens When You Combine Both
This is where the system changes qualitatively. When a Custom GPT operates inside a Project, you get something that doesn't exist in either tool alone: an AI that knows your specific work AND knows how to reason about it with a consistent methodology.
To make this concrete: imagine a banking app called MobilBank with a UX problem — 59% of users who start onboarding never finish, and the ID verification step alone has a 43% abandonment rate. You've done the research. You have a UX report, usability test results, user personas, a PRD, a competitive analysis, support tickets, and a business metrics dashboard.
Uploaded into a Project, all of that context is available simultaneously. Paired with a Custom GPT configured as a UX Strategist — one that knows to tag evidence vs. assumptions, apply severity ratings, and structure outputs in a format you can present to stakeholders — the same prompt that produces generic advice in a blank chat window now produces cited, framework-grounded, actionable analysis.
Not because the prompt was better. Because the system was designed.
The Architecture in Practice
The mental model is simple:
- Project = Operational Memory — your product, your research, your context
- GPT = Specialized Intelligence — the expertise, the methodology, the output structure
- Both together = Contextual AI Worker — a system that compounds in value as you add more context and refine the instructions
The setup takes under 30 minutes. You don't need to write code. You upload documents, write instructions that define the AI's role and behavior, and run structured prompts designed to produce consistent, presentable output.
The outputs that come back cite specific documents. They separate evidence from assumptions. They apply severity ratings. They produce executive briefs calibrated for C-level language — in seconds, not days.
This is a workshop I ran live — you can watch the full session below.
Where to Start
Pick one real project you're working on. Ask yourself what context — if the AI had it — would make every prompt dramatically better. Your research findings, your user personas, your product constraints, your success metrics.
Upload that. Define a GPT that knows how to analyze it. Run one prompt with a structured output format.
Then compare that output to anything a blank chat window has given you.
That gap is what architecture feels like.