Putting AI to work in everyday life.

Brands are under pressure to have an AI strategy. LLMs, generative AI, copilots, agentic systems, multimodal applications and AI search have opened up possibilities that were difficult or expensive to pursue before. Software can be built faster. More work can be automated. Small teams can attempt things that previously required substantial resources.

But putting those capabilities to work introduces its own problems.

Whether a business serves other businesses or consumers, its offering eventually has to become useful under those circumstances. Calling it “AI powered” tells us very little about whether it will.

Kitsune AI システム is a Physical Intelligence Lab working on that problem. Brands can work with us to develop applications around the moments in which their products and services have a meaningful role.

Robotics and autonomous systems are familiar expressions of physical AI. We believe the category also needs to develop around people carrying out their own activities: preparing a meal, making something, caring for someone or adapting when circumstances change.

For a brand, that creates a practical opportunity for engagement. Its knowledge, products and services can contribute to what someone is trying to accomplish. The value has to justify the participation being asked of that person.

Our Physical-First Intelligence position begins with presence and purpose. A person is participating somewhere, at a particular time, among things and relationships that matter to what they are doing. Presence helps establish which assistance belongs in that situation and who is entitled to participate.

Where something is, what happened earlier and what needs to happen next all affect its relevance. Those relationships ground the context an application uses, the choices it should consider and the consequences it needs to account for.

Digitized effort means selectively capturing something useful from what a person is already doing: a photograph, a spoken observation, a decision, a correction or a result. That gives intelligence evidence to work with. The effort involved in contributing that evidence needs to return something useful to the individual.

Agency has to show up in what the person can do: inspect an interpretation, correct it, choose a different course and determine how their information is used. Collecting data by itself does not provide that agency.

Consequence belongs in the process from the beginning. A recommendation can affect what someone buys, what they use, what they waste and what becomes possible later. We need to consider those effects before acting and learn from what actually happens.

We organize the work across four Moment Types: Private, Personal, Professional and Public. They allow for inquiry for ourselves, life within trusted relationships, work undertaken in a professional role and contributions intended for wider use. Each has its own purpose, responsibilities and terms of participation.

The same meal could matter differently within each of those contexts. A person's reflection, a household's shared experience and a professional's research are distinct accounts. Contributing something publicly requires its own decision and terms.

Our data-well methodology addresses the information an application needs to make its output useful in those moments. We connect existing digital records with selected evidence from physical life and organize their relationships into spatiotemporal data streams: relevant observations, actions and interpretations connected across place and time.

These streams develop within the purposes and permissions of Private, Personal, Professional and Public Moments. The same event can contribute to different streams where authorized, while each account retains its own purpose and terms of use.

Place, Space and Thing establish where the activity happens and what it involves. Goal, Method and Task describe what someone is trying to accomplish and how they approach it. Each Task connects an Action with the human (HI) or AI Agent responsible for carrying it out. Time connects what happened earlier, what is happening now and what needs to happen next. We can begin with one intention and add context as it becomes useful.

A receipt records a purchase. Product information describes an item. A photograph offers evidence of what appears to be available in a particular place. The stream connects those records to a purpose and to the actions that follow. A purchase last week does not establish what is available today; subsequent use and correction change the understanding.

The methodology keeps source, timing and uncertainty attached to that understanding. What someone stated and what a model inferred remain distinguishable. Purpose and authority determine what can be collected, processed, retained or contributed elsewhere. Transforming an experience into structured context must preserve the individual's authority over its use.

Before, During and After connect intention, effort and consequence across the stream. Processing can help prepare for an activity, support it as relevant evidence becomes available, and interpret what happened afterward. The result can inform the next moment.

Our ephemeral approach gives raw captures a defined lifetime. Processing must respect that boundary. As the source material expires, permitted findings can remain available with their source references, timing and uncertainty, and can be revised through later experience. Retention follows a purpose and its permissions.

The aim is to build relevant Perspective: an understanding that helps someone decide what to do next and can change when new evidence arrives. A useful stream can develop over time while much of its raw source material expires. Its value depends on what it helps people understand and do.

We chose food as a starting point for this work because it connects everyday experience with consequential decisions. Eating involves individual needs, household relationships, practical skills, products and production. It also gives people a familiar way to participate in examining these ideas.

Our Kitchen work brings them together. Someone might begin with groceries they have just bought, a meal they want tomorrow or a change of plans that leaves twenty minutes to cook. Food, storage, equipment and previous decisions become relevant in different ways. Information arrives at different times, and its meaning changes.

In this System of Things, a home provides the Place, the kitchen a Space and the pantry or fridge a Thing involved in the activity. Preparing tomorrow's dinner supplies a Goal. A chosen recipe provides a Method, with Tasks connecting Actions to the people or AI Agents responsible for them. A person might check an ingredient or prepare it; an AI Agent might interpret a photograph or help revise a preparation plan.

Spatial technology helps anchor a utility where it belongs. Its behavior has to fit the person's intention and circumstances. Someone starting with what they bought is working toward a possible meal. Someone starting with tomorrow's dinner is working backward toward the resources and preparation it requires.

Our Before, During and After method keeps those different starting points connected to their consequences. A photograph taken before preparation can support a decision; a correction during cooking can change the plan; a reflection afterward can revise the Perspective used for a future meal. The person can enter with a present resource, a future intention or a previous experience worth understanding.

Intelligence processes alongside that experience. The person continues preparing, deciding and adapting. Relevant evidence becomes available for processing, and the system returns support or a revisable interpretation. The next action can change the situation again.

Human participation therefore extends beyond approving a machine's output. People give the activity its purpose, carry out the effort, experience the consequences and decide what to do next. Particular automated actions can still require explicit approval.

Our Kitchen prototype includes experiments with an open-weight model running on a laptop to interpret ingredient photographs. These are working experiments. Their value depends on the accuracy and timeliness of the resulting support, the effort required to use it and the demands of operating the whole system. Local inference is one part of that work; storage and coordination currently use a backend.

For a food brand, this opens a relationship extending from purchase into use. Useful support can help someone work with a product in their own circumstances. Under explicitly agreed terms, what people choose to contribute can also inform product development, formulation or future support. Individual and shared benefit need to be designed into that relationship.

A commercial engagement with the lab begins with a specific product or service and a situation its customers encounter. We establish the useful intervention, design the experience and its data relationships, and test the assumption most likely to determine whether the application works.

The output is an application design, evidence from that focused test and a scoped pilot plan covering technical requirements, operating responsibilities, estimated costs and measures of benefit. The data design identifies what the business already has, what participation is needed, how the relevant streams develop and what people receive in return. It establishes how the resulting understanding can be used, including what expires and what may be retained.

A business can use that work to decide what to build, what it will take to deliver and whether the value justifies the investment.

Our website brings together the research position, method and working references. To discuss an application, use the contact desk to describe the product or service, the situation people encounter and the change you want to make possible.