Docebo / 2026
AI Agents & Assistant
Led the product experience for Docebo's AI Assistant and Agent Hub — a shared conversational layer for discovering, configuring, and using specialized agents across learning and admin workflows.
Design Highlights
Overview
I led the product experience for Docebo's AI Assistant and Agent Hub: a connected system that brings general assistance, custom agents, and specialized System Agents into the learning platform. The work spans the persistent AI entry point, panel and fullscreen conversations, chat history, Agent Hub discovery, and the controls administrators use to configure identity, access, knowledge, and actions.
Challenge
Docebo's growing AI capabilities needed to feel like one coherent product rather than a collection of disconnected assistants. Learners and administrators had to understand which agent was active, what context it could use, and what it was allowed to do. At the same time, administrators needed a manageable way to govern custom and System Agents without exposing the complexity of the underlying runtime, integrations, and permission model.
Approach
I designed the experience as a shared AI layer with clear identity, consistent conversation patterns, and context-aware specialization. I reconciled the live runtime and Figma system across the complete state surface — entry points, suggestions, composer behavior, history, panel and fullscreen modes, activation gates, attachments, deep links, and agent-specific controls. In Agent Hub, I organized configuration around the questions administrators actually need to answer: who can use this agent, what knowledge can it search, and which actions can it take?
Solution
AI Assistant is available from the core LMS shell and preserves the conversation as people move between a compact side panel and focused fullscreen view. Specialized System Agents adapt the same foundation to platform guidance, authoring, content discovery, reinforcement, in-course tutoring, and analytics. Agent Hub gives administrators one place to create, find, enable, and configure agents with explicit instructions, triggers, audiences, role requirements, approved knowledge collections, and external actions. Grounded answers expose sources and preserve existing access controls so assistance stays useful without becoming opaque.
Impact
The result is one extensible experience model instead of a separate interface for every AI capability. A shared entry point now supports six specialized System Agent experiences while Agent Hub makes both custom and system-provided agents understandable and governable. The live environment demonstrates end-to-end continuity from discovering an agent, to configuring its capabilities, to using it inside the product.
Key Learnings
Agent platforms need more than a capable model. People need a clear sense of identity, scope, context, and control at every moment. The strongest design decisions made the system legible: persistent agent identity, permission-safe knowledge, visible sources, explicit audiences and roles, and deliberate boundaries around actions. Reusing one conversation system also made specialization feel coherent instead of fragmented.