- Role
- Product design lead
- Engagement
- Platform design
- Focus
- AI product design · Scientific software · Agentic UX · Complex workflows
01 / Context
An agentic scientific discovery platform that turns natural-language questions into traceable computational chemistry workflows.
02 / Challenge
AI had to remain powerful without becoming unpredictable.
El-Agente is a scientific AI product for computational chemistry workflows. The challenge was turning complex agent behavior, generated outputs, and scientific data into an experience expert users could understand, inspect, edit, and trust.
03 / Product direction
Understand
I structured the experience around the relationship between user intent, agent reasoning, generated work, and explicit human approval.
Structure
The interaction model separated what the agent was doing from what required user review or action, with clear states throughout the workflow.
Deliver
The resulting product direction combined natural-language interaction with structured outputs, transparent feedback, and editable scientific results.
04 / Key decisions
Agent states made visible
Idle, reasoning, proposing, awaiting approval, complete, and failed states were surfaced explicitly so users always understood what the system was doing.
Approval before action
Model proposals were separated from user-approved actions, preserving expert control rather than allowing the experience to behave like an opaque autonomous chatbot.
Structured outputs over chat alone
Scientific results were presented through editable outputs and data visualization rather than being buried entirely inside conversation history.
05 / System
A system built around transparency and control.
Reusable patterns for agent states, approvals, feedback, outputs, and failure handling created a foundation that could support increasingly complex AI workflows without losing transparency.
06 / Product screens
07 / Outcome
The product direction received stakeholder approval and moved into development with a clearer interaction model for agent behavior, scientific outputs, and human control.
08 / Reflection
“For expert AI products, trust comes from visibility and control, not from making the intelligence disappear.”