Separate agent state from work state
An available process is not the same as a completed task. Making that distinction visible improves the quality of oversight and the next decision.
A fleet operations product for people leading named agents across projects, teams and locations.
Visit project ↗Once several agents work together, coordination becomes a product problem. Who owns the task? Which run is active? What is blocked? Where does a human decision belong? SkiaFleet turns that operational context into something people can see and manage.
I build the product and the harnesses around it: defining the operating model, designing the coordination experience and developing the platform foundations. The work combines product decisions with execution safety, recovery and a practical review loop.
SkiaFleet
Explore the responsibilities and the boundaries between them.
Claims · Generation fencing · Approval · Durable recovery
Ownership and decisions
Tasks, context, tools
Results, review, next action
A fleet operations product for people leading named agents across projects, teams and locations.
An available process is not the same as a completed task. Making that distinction visible improves the quality of oversight and the next decision.
Conceptual architecture drawn from the project’s responsibilities and design decisions. Arrows show relationships, not an execution trace.
An available process is not the same as a completed task. Making that distinction visible improves the quality of oversight and the next decision.
Approval needs the proposed action, relevant context and a consequence a person can understand. It cannot be just a decorative confirmation button.
Claims and generation fencing address competing or stale execution. Recovery and retained context help the system resume useful work after interruption.
Agent operations is an exercise in making authority, ownership and progress legible. The interface and the harness have to agree.