Building Extreme's AI from Zero

Extreme Networks spent decades as a hardware company before moving toward cloud and AI. Until 2024, AI work stayed inside research while market pressure built: AI funding surged, expectations skyrocketed, and customers demanded answers.

I was the sole design lead for Extreme's company-wide AI initiative, owning the architecture, interaction model, and design language across four products, from foundational research through alpha launch. I co-invented the agent-orchestration framework the platform runs on.

A cross-functional tiger team formed in Q4 2024: the CTO, VP of Product, an AI researcher, and design. The team reframed the work around one question: what is the right relationship between humans and intelligent systems? That framing, more than any integration task, shaped every decision that followed. Three challenges defined the work:

• ·Adding intelligence: introducing explicit AI capability into an existing, near-launch platform with minimal disruption.
• ·GenAI primitives: evolving traditional UX building blocks once AI becomes a co-collaborator rather than a tool.
• ·Agent orchestration: managing agency, discovery, expectations, and error recovery across multi-agent systems with little precedent to borrow from.

With no precedent for applied AI in networking, the team weighed competing approaches with nothing to borrow from. AI Exchange gave the work tangible structure: a user-facing system for discovering, managing, and understanding active intelligence. Its mental model runs on three phases. Explore asks what the system can do and whether it's secure. Configure handles control, cost, and activation. Monitor asks whether it's helping. Together they form a continuous loop that drives adoption.

As the work scaled across the organization in Q2 2025, we codified a new language of intelligence inside the design system: a set of GenAI primitives that extend traditional UX affordances.

• ·Chat with Everything: any object becomes conversational, from data to docs to devices.
• ·Format Translation: fluid conversion across modalities, text to table to diagram to video to code.
• ·Multi-Player: shared environments for concurrent human and agent action.
• ·Semantic Resize: adjusting meaning on demand, shorter, deeper, simpler, or more formal.
• ·Human Shift: the user moves from maker to editor, curator, and supervisor.
• ·Asynchronous Agents: systems work independently while humans orchestrate priorities.

In production, the guided agent flows cut task-resolution time sharply: up to roughly 98% on the most bounded tasks, with the broader AIOps field sustaining 40 to 50% blended.

• ·Operationalized a cross-functional AI framework spanning research, engineering, and strategy.
• ·Architected Extreme's first AI product strategy and go-to-market motion.
• ·Co-invented an AI Agency Architecture for Networks (patent filed 2025).
• ·Influenced the architecture of the next-generation networking platform.

The alpha launch in Q3 2025 brought architecture, primitives, and language together, with design connecting research, engineering, and strategy into one shipping product.

• ·You can't design for AI the way you design for deterministic systems. Predictability has to be explicitly designed in.
• ·Leadership through design: system architecture drove the experience at first, then the relationship inverted as patterns matured.
• ·Exploration becomes a pattern language, moving from principles to primitives to patterns to components.
• ·AI experiences are dialogic and co-adaptive: the system and the person shape each other as they work, and the design has to hold that loop.