Below is our recent interview with Kosuke Shirako, Founder and President at SHIRO & Co.:
Q: Could you provide our readers with an introduction to SHIRO & Co.’s broader research?
A: SHIRO & Co. is a Tokyo-based research and design company working across AI, technology, society, industry, and human experience.
Much of our work begins with observation. We look for weak signals, small behavioral changes, structural gaps, or situations that do not fit established categories. From there, we try to separate what is directly observed from what is inferred, identify the boundary conditions, and then turn the question into something operational: an observatory, a decision framework, an experimental system, or a working product.
Our broader research asks questions such as:
- When does AI capability become real-world authority?
- What do ranking and recommendation systems make visible, and what do they leave out?
- When should an AI action proceed, pause, or be denied?
- How can continuity, identity, memory, and provenance be preserved across changing systems?
- What remains human when technology works correctly?
The goal is not prediction for its own sake. It is to make emerging structures visible early enough that people and organizations can still decide how they want to respond.
Q: Any highlights about your product work?
A: A key part of SHIRO & Co. is that research does not remain only as essays or concepts. We try to turn ideas into working interfaces and systems.
One example is Customer to Proposal, which structures customer context, observations, and public information into a proposal workflow rather than jumping directly from raw input to recommendation.
Another is Visible World Audit, which records and compares what ranked or personalized systems actually make visible to different synthetic personas over time.
We also recently built HOME, a private family-only digital space. It began from a very small observation: messages in a long-running group were often read but not answered. Instead of treating that as an engagement problem, we asked whether some forms of communication might need less pressure, fewer people, and clearer boundaries.
The product removes many features that conventional social platforms normally add: public identity, discovery, recommendation feeds, likes, follower counts, and pressure to reply.
For us, these products are different in subject matter, but they are generated by the same underlying process: observation first, structure second, implementation later.
Q: Can you tell us more about Threat Observatory, Authority Surface, and Visible World Audit?
A: The Threat Observatory examines structural changes in technological risk rather than functioning as a conventional incident-alert service.
One of its central observations is the Authority Surface.
AI safety discussions often focus on capability: how intelligent, autonomous, or effective a model is. Authority Surface asks a different question:
What has the surrounding system allowed the AI to reach, influence, or control?
That may include credentials, APIs, financial systems, cloud infrastructure, industrial systems, or physical machines.
We examine that authority across dimensions such as access, permission, persistence, propagation, reversibility, and observability.
This connects directly to our Supervisory Layer work.
An AI agent may be known, authenticated, and technically authorized, but that does not necessarily mean every action should proceed. A system may need a runtime decision such as ALLOW, HOLD, or DENY depending on context, timing, scale, reversibility, and consequence.
Visible World Audit examines a different but related question: not what a system is allowed to do, but what world it makes visible.
People often choose from what systems choose to show them first.
Visible World Audit records ranking, exposure, framing, and subsequent change without pretending to know proprietary hidden algorithms. It preserves observable evidence: what appeared, in what order, to whom, and after what sequence.
Together, these systems look at different parts of the same broader problem:
Q: What is the best thing about the Kosuke Protocol?
A: The best thing about the Kosuke Protocol is that it does not begin by asking, “What product should we build?”
It begins with:
What did we actually observe?
That sounds simple, but it changes the process.
The protocol separates observation from interpretation, identifies what is known, what is inferred, what remains uncertain, and where the important boundaries are.
A simplified version is:
Observation → Structure → Interpretation → Form
The final form might be software, an observatory, a decision tool, a film, a proposal, or sometimes nothing at all.
That last part matters.
The protocol is not designed to force every signal into a product. It is designed to keep the original observation intact long enough to discover what kind of response is actually appropriate.
For me, that is the core value: it allows very small observations to become useful systems without making them artificially large before they become real.
