Meeting
Intelligence

Agent

LISTENING BECOMES UNDERSTANDING
LISTENING BECOMES UNDERSTANDING
LISTENING BECOMES UNDERSTANDING
LISTENING BECOMES UNDERSTANDING

An AI agent that turns workshop conversations into live process maps while teams are still talking.

01

What if documentation could keep up with the conversation?

Workshops are great at generating clarity, but turning that clarity into documentation often takes hours of follow-up work. This project explored whether an AI agent could transform live discussions into process maps and documentation as the conversation unfolded.

As UX Strategy and Interaction Design Lead, I defined the agent's behavior, interaction model, and approach to human-AI collaboration, ensuring participants could stay focused on the workshop while the agent handled documentation in the background.

Team1 Designer · 5 AI Developers · Architect · SMEs  

  • ROLEUX Lead

    CLIENTGlobal Fortune 50 Technology Company

    DATE2025

02

Workshops generated clarity. Capturing it required hours of work.

Context

  • Workshops were highly effective at bringing people together to align on processes, decisions, and next steps. The challenge was preserving ideas.

    Facilitators often left sessions with a collection of notes, whiteboards, screenshots, and transcripts that still needed to be organized, documented, and shared. Turning a productive workshop into usable documentation could take hours of additional effort.

    We saw an opportunity to explore whether AI could eliminate that gap, transforming conversation into structured documentation as the workshop unfolded.

Abstract sound-like waveforms representing a mountain.
Success looked like:

  • Accurate transcription

Useful process maps

Positive facilitator feedback

Reduced documentation effort

03

We started with the people running the workshops.

What we learned

  • Through interviews with facilitators and operations leads, a consistent pattern emerged: the workshop itself wasn't the problem.

    The real work began after the meeting ended.

    Facilitators spent hours organizing notes, reconstructing process maps, and turning fragmented discussions into documentation others could use.

Constraints observed

  • /01Workshops are collaborative and often involve multiple people speaking, correcting, and refining ideas simultaneously.

    /02Facilitators wanted less documentation work, not more software to manage.

    /03Any solution needed to fit naturally into existing workshop tools and behaviors.

Simplified User Flow
Simplified flow demonstrating the pipeline of speech to diagram creation.
Mapping the journey

  • To understand where AI could provide value, I mapped the journey from discussion → understanding → documentation → action.

    The goal was to preserve the decisions, relationships, and outcomes that emerged during the workshop.

04

More controls didn't create more confidence.

Evolving the agent

  • Our initial assumptionWe believed more controls would create more confidence.

    What we learnedParticipants wanted to focus on the workshop, not the tool.

    How the solution changedThe agent worked quietly in the background, continuously turning conversation into documentation as the discussion unfolded.

The best experience was the one users barely noticed

  • As discussions evolved, the agent continuously refined its understanding of the workflow, translating conversation into an increasingly complete process map.

    The examples below show how the documentation emerged.

Start where teams already workThe agent was embedded directly into Microsoft Teams, allowing participants to begin documenting discussions without changing their existing workflow.

Listening before actingThe agent began by understanding the conversation, identifying the people, decisions, and relationships that would eventually shape the process map.

Initial example of agent diagram rendering, with few nodes.

The map emergesRequirements are continuously translated into structure.

Second example of agent diagram rendering, with additional nodes and connections.

Ideas buildElements present as aspects of the workflow rather than isolated discussion points.

Final example of agent diagram rendering, a full end-to-end flowchart with logical nodes connected and key shapes from legend properly used.

Connections appearProcess map continues evolving alongside the conversation.

Third example of agent diagram generation, with structure mapped out and core nodes expressed.

Documentation createdThe entire map is created, eliminating much manual effort.

Ending the meeting, not the workParticipants could conclude the session while the agent finalized documentation, prepared exports, and organized supporting artifacts in the background.

Transparency in actionsThe agent communicates progress with the actions it takes.

Turning conversation into a shared assetProcess maps, transcripts, and supporting documentation were automatically stored and shared, transforming a one-time discussion into knowledge the broader organization could reuse.

05

As teams talked, the documentation built itself.

The meeting became the documentation process.

  • Instead of spending hours organizing notes, rebuilding process maps, and sharing outputs after a workshop, participants left with documentation already in progress.

    The prototype demonstrated that conversation could become structure in real time, generating process maps, documentation, and editable exports as discussions unfolded. What was traditionally a follow-up activity became part of the meeting itself.

  • 97-98% accuracySpeech-to-transcript accuracy  

94-96% accuracyTranscript-to-diagram accuracy

2-4 secondDiagram updates during discussion

Under 2 minuteEditable export generation

06

Designing AI means designing for human intent.

Agents are truly 90% UX and 10% UI
  • “The real value of AI isn’t in how much it automates, it’s in how faithfully it preserves human structure and meaning.”

    This project reinforced a lesson I've seen repeatedly when designing AI experiences: people don't want more systems to manage.

    The most successful version of this product was the one that got out of the way while gently supporting the users.


Next Steps
/ Help participants better understand how the agent reaches its conclusions. / Expand the workflow from documentation into planning and execution.  
/ Improve support for larger and more complex process maps.  
/ Introduce feedback loops that allow the agent to continuously improve over time.

Sound like a fit?

Let’s talk!

yuri@yurian.io
I care more. About the details, the systems, and the people who rely on them.