Working with AI Agents

Audra Flow includes four specialized AI agents, each designed for a specific part of the product delivery process. These are not generic chatbots — each agent understands its domain and draws on your project's context, uploaded documents, and existing artifacts to give you relevant, actionable output.

You can interact with agents through the chat panel on the right side of your workspace, or through the Generate with AI buttons that appear on individual artifacts. The agent you see depends on which phase and artifact you are working in.

UX Researcher

The UX Researcher agent is your partner during the Discovery phase. It helps you make sense of user research, spot patterns in interview data, and identify what you still need to learn before moving forward.

When to Use It

You have completed stakeholder or user interviews and need to pull out the key themes.
You want to check whether your research covers the right areas or has blind spots.
You are defining project goals and want suggestions based on what you have learned so far.
You need a competitive landscape overview or want to compare your approach to similar products.

What It Can Access

The UX Researcher reads your project's Discovery artifacts — interview transcripts, goals, research notes — along with any documents you have uploaded to the project's knowledge base. It does not access artifacts from other projects.

Example Prompts

“Summarize the top three themes from the interviews I have uploaded.”
“What research gaps should I address before moving into Definition?”
“Suggest three project goals based on the interview data and competitive analysis.”
“Compare the onboarding approaches of the three competitors mentioned in the research.”

Product Owner

The Product Owner agent helps you turn research and ideas into clearly defined product artifacts. It specializes in the Definition and Delivery phases, where precision matters — writing specifications, crafting user stories, and setting priorities.

When to Use It

You need to write a product specification and want a solid starting point.
You are breaking down a feature into user stories with clear acceptance criteria.
You want help prioritizing scope for an upcoming release or MVP.
You need to refine an existing story or spec based on new information.

What It Can Access

The Product Owner draws on your project's Definition and Delivery artifacts — personas, user journeys, service maps, specifications, and story maps — as well as your knowledge base documents.

Example Prompts

“Write a product specification for the user onboarding flow based on the personas and journeys I have defined.”
“Break this specification into user stories with acceptance criteria.”
“Which stories should be in the MVP and which can wait for a later release?”
“Refine this user story to cover the error handling scenarios mentioned in the service map.”

Architect

The Architect agent focuses on the Design and Delivery phases, helping you move from product requirements to technical plans. It generates technical specifications, designs system architecture, and creates API contracts that your engineering team can work from.

When to Use It

You need a technical specification that translates a product spec into an engineering plan.
You want to design the system architecture for a new feature or service.
You need to define API contracts, data models, or integration points.
You want to evaluate trade-offs between different technical approaches.

What It Can Access

The Architect uses your project's specifications, story maps, and existing technical artifacts, along with any architecture diagrams or technical documents in your knowledge base.

Example Prompts

“Generate a technical specification for the payment processing feature based on the product spec.”
“Design the system architecture for the notification service, including the message queue and delivery channels.”
“Create an API contract for the user profile endpoints with request and response schemas.”
“What are the trade-offs between a synchronous REST approach and an event-driven approach for this integration?”

Product Guru

The Product Guru is your project's knowledge-base agent. Unlike the other three agents, which specialize in specific delivery phases, the Product Guru is available throughout the entire workspace. Its job is to answer questions using the documents you have uploaded to your project.

When to Use It

You have a question about something covered in your uploaded documents — a policy, a technical constraint, a business rule.
You want to find relevant context from your documents without reading through everything manually.
You need to check whether your project documents address a specific topic or if there is a gap.
You want a quick summary of a long document or set of documents.

How It Uses Your Documents

When you upload documents to a project, they are analyzed and indexed so the Product Guru can search through them efficiently. When you ask a question, the agent finds the most relevant passages from your documents and uses them to form its answer. It will cite which documents it drew from, so you can verify the information.

Example Prompts

“What does the compliance document say about data retention requirements?”
“Summarize the key points from the stakeholder presentation I uploaded.”
“Do any of our documents mention accessibility standards we need to follow?”
“What gaps exist in our documentation around security requirements?”

AI Generation on Artifacts

Throughout the workspace, you will see Generate with AI buttons on individual artifacts — goals, personas, user stories, specifications, and more. These buttons let you use the appropriate AI agent to create or improve content directly within the artifact you are working on.

How It Works

Open the artifact you want to work on (for example, a user story or a product specification).
Click the Generate with AI button. The system will use the relevant agent for that artifact type.
The agent reads your project context — related artifacts, uploaded documents, and the current state of the artifact — and generates content.
The generated content appears in the artifact editor. You can review it, edit it, and keep what works.

Refining the Output

AI-generated content is a starting point, not a final product. After the agent produces output, you should:

Read through the content and check that it aligns with your project's goals and constraints.
Edit any sections that need adjustment — add details the agent missed, remove parts that are not relevant, or rephrase for clarity.
Run the generation again if you want a different take. The agent may produce different results each time, especially if you have added new context or documents since the last generation.

You can also use the chat panel to ask the agent to refine specific parts of an artifact. For example: “Rewrite the acceptance criteria for story #3 to include error handling.”

Tips for Better Results

The AI agents work best when they have good context to draw from. Here are some practical ways to get more useful output.

Upload Documents Early

Before you start generating artifacts, upload relevant documents to your project's knowledge base. Research reports, business requirements, technical constraints, brand guidelines, compliance documents — the more context the agents have, the more relevant their output will be.

Be Specific in Your Prompts

Vague prompts produce vague results. Instead of asking “Write a user story,” try “Write a user story for the password reset flow, including acceptance criteria for email verification and error states.” The more detail you provide, the closer the output will be to what you need.

Build Artifacts in Order

The agents use your existing artifacts as context. If you generate personas before user journeys, the journey generation can reference those personas. If you write product specifications before technical specifications, the Architect can work from what the Product Owner has already defined. Working through the phases in order gives each agent more to build on.

Review and Edit Every Output

AI-generated content is designed to save you time, not to replace your judgment. Always review the output, correct anything that is inaccurate, and add the nuances that only you and your team know. The best results come from treating AI output as a first draft that you refine.

Iterate When Needed

If the first generation is not quite right, you have several options: edit the output directly, add more context (upload a document or fill in related artifacts), and generate again. You can also use the chat to ask the agent to adjust specific sections. Each iteration tends to get closer to what you need.

Data Privacy

Your project data stays within your project. Here is what that means in practice:

Project-scoped access: Each AI agent only sees the artifacts and documents that belong to the project you are currently working in. It cannot access data from other projects, even if they belong to the same organization.
No cross-project training: Your project data is never used to train or improve the AI models. The agents use your data to generate responses in real time, but nothing is stored or shared beyond your project.
Knowledge base isolation: Documents uploaded to one project are indexed and stored separately. They are not visible or searchable from any other project.
Team permissions apply: The AI agents respect your project's role-based access controls. If a team member does not have permission to view certain artifacts, the agents will not include that content in responses for that user.

For more details on how Audra Flow protects your data, see the Security & RBAC page.

Next Steps

Learn how to build your project's knowledge base in the Knowledge Base & Documents guide.
See the Discovery guide to understand the first delivery phase where the UX Researcher shines.
Read about Design & Delivery to see how the Architect and Product Owner agents work together in later phases.