Skip to products
assaynotes

AI products, built around people.

Start with what a person is trying to do. Design the context, actions and decisions into something they can keep using. Personal agents are a particular focus.

Personal agents, built on your context.

A personal agent starts with one person’s knowledge, goals and ways of working. The design challenge is to carry useful context between tasks, connect it to action, and leave room for that person’s judgment.

Design scenario: the next research task

The earlier research still matters. New material has arrived. The aim is to continue from what is already known, without collecting and explaining everything again.

Earlier context sits beside the current question.

Start where the last task left off.

Select the earlier questions, sources and corrections that matter to this task from longer-term memory.

Brought in
Past scope, sources and corrections
How it works
Retrieve for the current goal
Carried forward
Context for this task

Interactive design study. No live AI or saved data.

Design you can step into.

From a question to something you can explore. Wonder Island brings content, interface and interaction into one learning experience.

  1. An open book with a waterfall, a geometric tower and a plant-filled island growing from its pages.

    Wonder Island

    Open a book. Step onto an island. Follow questions through math and science, try hands-on experiments, and keep what you discover.

    • Math & science
    • English & Chinese
    Enter Wonder Island

    Private trial · Sign-in required

Connected capabilities. Clear boundaries.

A three-language publishing architecture: how components work together, where data lives, and who can make it public. Models propose content, a person approves an exact revision, and a separate component handles publication.

Select a component to inspect its responsibility and connections.

100%
Agentic publishing · Conceptual architecture An architecture diagram generated by Archify. Workspace · Input, revision and human review · Private work and publication authority Workspace Input, revision and human review Agent orchestrator · Coordinates; cannot publish · Private work and publication authority Agent orchestrator Coordinates; cannot publish Context store · Sources and personal judgments · Private work and publication authority Context store Sources and personal judgments Versioned drafts · Revision + language set · Private work and publication authority Versioned drafts Revision + language set Approval / policy gate · Edits invalidate approval · Private work and publication authority Approval /policy gate Edits invalidate approval Run log / checkpoints · Readback before retry · Private work and publication authority Run log /checkpoints Readback before retry Publication adapter · Approved version only · Private work and publication authority Publicationadapter Approved version only Model / tool adapter · Selected context only · External model / tools Model / tooladapter Selected context only Public notes · Approved public content · Public website Public notes Approved public content Task Read context Selected context Draft / evidence Write revision Review bundle Approve revision Verify version set Approved export Publish / read back Receipt Save checkpoint Resume Private work and publication authority External model / tools Public website

Pinch or use the controls to zoom. Drag to explore; select a component for details.

Use plus and minus to zoom, arrow keys to pan, and zero or Home to fit the diagram. Tab moves between components.
Data and dependencyVersion-bound approval

Human control

A person supplies the intent and decides what can become public.

Private workspace

Sources, drafts and execution records stay within a controlled workspace.

External services

Each call receives only the context its task needs, with no publishing credentials.

Public site

Only an explicitly approved public export crosses this boundary.

A proposed architecture for publishing notes in three languages. This diagram is not connected to a live agent, approval system or publishing service.

AI products should help people move their work forward—and understand, change and continue what happens along the way.

Explore the design studies