From Data to Agent

Tudou Data — AI Agent OS Product System

Industry

AI Agent OS / Government Big Data

Role

Lead UX Designer

Period

2023–2025

Collaboration

Product · Engineering · Government Client

AI Agent OS Product System

Agents turn data into completed work

From data to agent was not about adding a chat box to a GIS product. It was about reorganizing how people call on data, knowledge, rules, and tools to complete work. I designed those distributed capabilities as an Agent OS that can be built, discovered, executed, and checked.

The case follows five connected actions: enter a task, build an agent, discover and reuse capability, follow task execution, and deliver specialist results that people can act on next.

01 / ENTER A TASK

Let users enter through a task

The home experience does not ask people to understand the system before they can work. Task entry points, agent switching, and quick commands let users begin with the problem at hand, select the right collaborator, and start immediately.

AI Factory agent task home — Agent home: begin with the work at hand and a clear input point

Agent OS agent switcher — Agent switching: keep context continuous across specialist collaborators

Agent OS quick commands — Quick commands: make frequent tasks visible and immediately callable

02 / BUILD CAPABILITY

Make agents buildable and configurable

An agent's capability comes from skills, business logic, knowledge, and permissions working together. The canvas makes those dependencies explicit, while creation preserves the reasoning and configuration path. For complex processes, a compute graph, node parameters, and runtime states give specialists a debugging surface.

Agent OS agent-creation reasoning — Creation: reveal goals, reasoning, and configuration progressively

Agent OS skill, business, and knowledge canvas — Capability canvas: compose skills, business logic, and knowledge into explicit dependencies

Agent OS private knowledge-base permissions — Knowledge permissions: make access scope part of capability configuration

Kube workflow-debug compute graph — Workflow debugging: inspect complex execution with node state, parameters, and logs

03 / DISCOVER AND REUSE

Make capability discoverable and reusable

Once the number of capabilities grows, distribution becomes a product problem of its own. The agent marketplace creates a browsable entry point, while an agent profile clarifies its applicable work, capability makeup, and boundaries so users can decide what to use and why.

Agent OS marketplace — Agent marketplace: organize distributed capabilities into a browsable work entry point

Agent profile: clarify applicable work, capability makeup, and resource dependencies

04 / EXECUTE A TASK

Bring natural language into a real workflow

From searching for resources to starting a conversation, users do not need to move back and forth between tools. Task views hold input, context, process information, and result cards together, connecting a natural-language request to actual execution while it remains visible.

AI Factory search empty state — Search entry: begin matching resources and capability from a stated need

AI Factory search results — Search results: turn matched resources into a selectable next step

AI Factory in-progress agent task details — Task details: keep conversation, process cards, state, and outputs in one work surface

05 / DELIVER RESULTS

Make specialist results inspectable and actionable

Map diagnosis does not return a sentence alone. It structures issue locations, inspection categories, map annotations, and adjustment advice into one result. Document validation proves that the same delivery pattern supports another specialist task and can continue into revision and review.

Map-diagnosis agent website-diagnosis entry — Diagnosis entry: start a specialist check from a source or link

Map-diagnosis agent result overview — Result overview: review the issue summary with its spatial locations

Map-diagnosis agent detailed inspection — Detailed inspection: trace the basis for each issue category

Map-diagnosis agent adjustment advice — Adjustment advice: translate findings into executable changes

Document-validation agent result list — Document validation: apply the same structured delivery pattern to another specialist check and review

06 / DESIGN METHOD

Product principles

The Agent OS established a reusable approach to specialist work:

Organize the entry point around a task instead of a feature module
Combine knowledge, skills, business rules, and permissions into legible capability
Show conversation, structured results, and runtime state together so execution is not a black box
Let results continue into revision, review, or the next collaboration instead of ending as a single response

Next Case

From Map to Data→