# StateLab

Online alpha · 2026-10-26

An interaction lab that turns AI content-generation rules into working prototypes of waiting, approval, failure, and recovery—so teams can experience behavior before building it.

Online alpha: https://statelab.linadong.design

## Design question

How can a team experience the difficult moments of an AI workflow before those moments reach users?

Make behavior rules explicit, generate a three-step simulation, and let a designer experience six scenarios while recording their own observations and decisions.

## My contribution

Product adaptation, behavior-rule design, six scenario simulations, human review and export flows, and the controls around optional AI editing.

## Workflow

1. **Describe** — Define the content-generation feature or choose a template.

2. **Decide** — Confirm cancellation, partial-result preservation, and retry approval rules.

3. **Rehearse** — Generate a template-based prototype and trigger six scenarios.

4. **Review** — Record human checks, then export HTML, a handoff note, or a JSON backup.

## Mechanisms

**Agent · optional** — The local application includes structured AI planning and a code-editing agent limited to 12 tool steps. Model calls are disabled in this public alpha; its templates and scenarios work without AI.

**Harness · implemented** — Confirmed rules, six scenario inputs, a restricted prototype file, and an isolated iframe with network restrictions make execution boundaries explicit. Changes invalidate earlier human review results.

**Verification · human-led** — A static checker verifies required prototype interfaces. Reviewers operate each scenario and record their conclusions; these checks do not constitute automated browser evaluation.

## One feature. Four moments to design.

Illustrative walkthrough · the interactive alpha is available above

**Waiting** — A video is still generating.

Show meaningful progress and a clear next action, without promising an unverified completion time.

**Approval** — A retry would consume more credits.

Explain the extra cost before execution and let the user choose whether to continue.

**Partial failure** — Two scenes succeeded; one failed.

Keep completed work visible and separate the failed part from the rest of the task.

**Recovery** — The user retries the failed scene.

Make the retry scope explicit and preserve the already accepted scenes.

These examples explain the interaction rules. The online alpha simulates outcomes and credits; it does not generate real content or perform an automated UX evaluation.

## Current state and next steps

The online alpha supports templates, six simulated scenarios, manual review, browser-local saving, and HTML/Markdown/JSON export. It performs no real generation or payment.

Evaluate AI-assisted editing against the scenario rules, then expand beyond the fixed three-step content-generation template.

Adapted from a downloaded UIGen starter. The StateLab product layer adds scenario rules, simulation, review, persistence, and handoff. The public release is a browser-only build with AI features disabled.

https://linadong.design/lab/statelab