Agnes AI is a collaborative workspace that combines document editing, presentation creation, and design tools into a unified platform. The platform leverages generative AI to help teams work smarter, automate repetitive tasks, and create professional content faster.
Agnes is a powerful AI workspace that instantly generates polished slides and documents.
the UI discouraged commenting, feedback was scattered, and AI bulk-resolve lacked clarity.
Which risked confusion, misalignment, and reduced trust in both collaboration and AI-generated work.
The hardest part of this project wasn't the UI itself — it was defining the right ask and deciding what the end-to-end user flow should be.
There was a clear gap between users leaving comments and users actually making edits, and the existing flow wasn't helping bridge it.
Users could leave feedback, but resolving that feedback still required writing prompts to ask AI Agent to resolve each comment individually. The tool supported collaboration — but the workflow didn't.
Introducing bulk-resolve sounded powerful — but also risky. If AI edited everything at once, users might lose control over their content or feel unsure what had changed.
Signal vs. Noise: The AI identifies comments containing specific edit instructions (e.g., "Change this to blue") and pre-selects them for resolution.
Human-in-the-Loop: Subjective comments are left unselected, requiring manual user review. This built trust by proving the AI understands context.
Does AI pre-selection actually save time, or does it get in the way? If the system guesses wrong too often, users will bypass it entirely.
Add version history, conflict resolution, and permission controls so the tool works for larger, more complex team structures.
Progressive disclosure isn't about hiding complexity - it's about building confidence. I used to think progressive disclosure meant tucking advanced features away to keep interfaces clean. What I learned is that it's actually about sequencing information so users build mastery gradually. Each layer they uncover should feel like a natural next step, not a hidden Easter egg. The interface should grow with the user's confidence.
I learned that jumping into solutions without deeply understanding how current features function and how users have adapted to them leads to designs that look good but break real workflows. Spending time mapping every interaction, every edge case, and every workaround users developed wasn't just research - it revealed the constraints and opportunities that shaped every design decision afterward. The existing system tells you what users actually need, not what they say they need.
See my visual systems design work
Lina Wang
