AI/UX PLAYGROUND
Agents

Sandbox Preview

Share

Sandbox preview is an AI interface design pattern that runs agent plans in a no-write mode first, listing every concrete side effect (files touched, rows updated, messages drafted, API calls) in human-readable form before the user promotes the same plan to real execution. This UX pattern replaces vague confirm modals with an inspectable manifest of what would happen, so users can catch wrong scopes, wrong accounts, or missing guardrails while rollback is still trivial. Promotion is a separate, deliberate step after review, not an accidental continuation of chat.

Use Case

Essential for agentic automation, bulk admin tools, and cross-app workflows where users need to trust a manifest of effects before granting execution rights.

Examples in Wild

CI dry runsTerraform plansEmail merge previewsAgent “plan then act” modes

Use this pattern in your project

Copy this prompt to generate a production-ready implementation in Cursor, Claude Code, Lovable, or any AI coding agent.

Generate a production-ready implementation of the "Sandbox Preview" AI interface design pattern.

Pattern Description:
Interactive Demo
Restart demo
Coral Ops

Preview before execute

Nothing below is live yet. Calendar, CRM, and mail stay read-only until you choose Run.

2 changes in this preview.

What would change

Would create 1 event · “Vendor check-in” · 30 min

Would update 1 CRM note on Acme Corp (no field deletes)

Adjust plan

Include follow-up email draft

Adds another step to the list above. Still preview-only.

Related Patterns

Agents

Plan & Execute

Breaking goals into steps

Agents

Tool Use

Visualizing AI using external tools

Agents

Workflow Builder

Visual drag-and-drop workflows

Agents

Memory Management

Viewing what AI remembers

Agents

Task Queue

Visual queue of agent tasks

Agents

Human Handoff

Escalating to humans

Get new patterns by email

Weekly AI interface UX notes and resources on Substack, no spam, unsubscribe anytime.

Subscribe on Substack