Physical AI agent
XAtom
A 2.9 cm embeddable AI interaction core that helps everyday objects sense context, understand needs, and act proactively.
The challenge
When AI enters physical objects, how can it balance limited space, complex contexts, and proactive interaction?
On-screen AI can wait for a prompt. In the physical world, it must also navigate limited space, continuous sensing, and the boundaries of proactive interaction.
Product system
One core, many physical forms.
XAtom does not prescribe a single device. It separates intelligence into core, sensing, and action layers.
2.9 cm embeddable core
Separate intelligence from a single hardware form so it can fit charms, desk objects, toys, and car accessories.
Multimodal sensing
Combine voice, radar, and environmental signals to understand user state and context.
Connect services and act
Move from understanding a need to invoking services, completing a task, and returning feedback.
Key design decisions
Must intelligence be tied to one device?
Separate the intelligence core from the object
Fit the same core into multiple shells and objects to test whether its capabilities hold across forms.
How can proactive interaction avoid interruption?
Let sensing precede expression
Prototypes compare when to initiate, how to be interrupted, and when the product should stay quiet.
Does every context require a different product?
Let context shape the role
Safety, learning, guidance, and travel scenarios compare role expression and service boundaries for the same core.
Scenario prototypes
Roles are shaped by context, not preset features.
The same core becomes a different everyday role through its object, environmental signals, and connected services.
Safety guardian
Live in a personal object to offer reminders and assistance during travel or time alone.
Learning companion
Live in a toy or desk object so questions, explanations, and companionship feel natural.
Guide and travel companion
Live in a charm or car accessory to provide information and task support on the move.
From AI core to everyday context
Give every object intelligence.
These films are not feature reels. They are scenario prototypes used to examine form, trigger timing, and role expression.
Prototype validation
Validation asks whether the interaction works—not how many features it has.
Does the form work?
Can the core enter different objects while keeping the interaction clear and usable?
Is proactivity appropriate?
Are triggers understandable, and can responses be declined, interrupted, or paused?
Does the role feel credible?
When the context changes, do tone, permissions, and behavioral boundaries remain coherent?
My role
I led product definition, experience strategy, and prototype validation, turning abstract AI capabilities into understandable, testable physical interactions and everyday scenarios.