One of the biggest challenges in designing AI products isn't building the technology. It's maintaining a close connection to the people you're designing for.
On one project, our primary users were highly specialised professionals whose schedules made regular research difficult. User interviews were invaluable when we could conduct them, but they weren't always available at the pace product development demanded.
Rather than relying solely on documentation or assumptions between research sessions, we explored a different approach.
We built our primary personas as AI sub-agents that could participate throughout the product development process.
Understanding the Opportunity
Our organisation already had an incredible source of knowledge.
Experienced Subject Matter Experts worked alongside the product team and had years of experience performing the roles we were designing for. One senior stakeholder had also created detailed training material explaining how these professionals worked, the decisions they made, the reasoning behind those decisions, and the outcomes they were trying to achieve.
Instead of treating this as static documentation, we saw an opportunity to transform it into something interactive.
The goal wasn't to replace users.
It was to capture institutional knowledge in a format that every member of the team could access while building the product.
Creating AI Personas
Rather than writing a single prompt or markdown file, we created dedicated persona agents that could be used inside Cursor, Codex and Claude Code.
Each agent was intentionally structured around three core components.
Knowledge
Domain expertise, terminology, workflows and the practical experience required to perform the role.
Guardrails
Clear behavioural boundaries, assumptions to avoid and guidance for handling uncertainty.
Expectations
The goals, priorities, decision-making patterns and success criteria that defined each persona.
This structure made the personas easier to evolve over time while keeping their behaviour consistent across different AI tools.
Integrating Personas Into Product Development
The personas became part of almost every stage of our workflow.
During discovery, they helped challenge assumptions and identify gaps before research sessions.
During ideation, they allowed us to explore workflows from the perspective of experienced users.
During prototyping, we asked the personas to complete common tasks, explain their reasoning and highlight areas that felt unintuitive or inconsistent with real-world behaviour.
This gave the team an early validation layer before engaging with customers, allowing research sessions to focus on refining ideas rather than uncovering avoidable issues.
Most importantly, every customer interview continued to improve the personas themselves.
As new insights emerged, they were incorporated into the knowledge base, creating a continuously evolving representation of our users.
Making Expertise Accessible
The personas were hosted centrally within our GitLab repository, making them available across the organisation.
Designers used them to review journeys.
Engineers used them to validate implementation decisions.
Product managers used them to explore new opportunities.
Instead of searching through documentation or waiting for SME availability, teams could access years of accumulated domain expertise directly within the AI tools they were already using.
Institutional knowledge became part of the development environment rather than something stored separately.
Outcomes
Introducing AI personas fundamentally changed how we approached product development.
We were able to:
- Validate ideas earlier in the design process.
- Reduce dependency on SME availability for everyday questions.
- Increase confidence before customer research.
- Make specialist knowledge accessible across disciplines.
- Continuously improve the personas through real user feedback.
The result wasn't fewer conversations with users.
It was better conversations, stronger prototypes and faster iteration between research sessions.
This project reshaped how I think about AI-native ways of working.
The greatest opportunity wasn't generating designs faster or writing better prompts. It was creating living representations of our users that evolved alongside our understanding of them.
As organisations continue adopting AI development tools, I believe we'll see more teams move beyond static documentation and towards shared AI teammates that capture expertise, preserve institutional knowledge and help product teams make better decisions every day.