# How do I ask a coding agent for a demo I can inspect?

> A founder-friendly request and acceptance checklist that turns an agent's 'done' message into a running page, observable behavior, and a reviewable diff.

Canonical HTML: https://canopyide.dev/guides/ask-coding-agent-for-inspectable-demo
Article date: 2026-09-28

An agent can write a convincing completion summary while the product still fails at the first click. Ask for evidence you can inspect: a specific running URL, a short sequence of actions, the checks performed, and the changed files. You do not need to read every line of code to reject a broken demo.

## Define the behavior before the code

Start with one user action and one visible result. In the included practice project, the request is to add All, Open, and Done filters to a task board. A useful acceptance check says that after one task is open and another is complete, each filter shows the right tasks. That is more testable than “improve task management.” Name the current behavior, desired behavior, and what must stay out of scope, such as adding an account system to a tiny local page.

## Ask for a running result and a path to it

Tell the agent to run the existing project command and give you the local URL or exact file to open. In Canopy, saved service commands and detected URLs can put the result in Preview next to the relevant workspace. A process marked running only proves that a command started; open the page yourself. If the task touches an API or worker, ask which service and endpoint demonstrate the change and where its logs appear. For a static page, a local file or simple dev server may be enough.

- Record the URL, port, branch, and component so you know what you are viewing.
- Load the page once with your normal settings and once in a fresh browser context if persistent local state could hide a bug.
- Keep the original task and acceptance checks visible while you inspect.

## Watch the agent demonstrate edge cases

For the practice board, create an open task, complete another, switch filters, change a task's state while filtered, refresh, and inspect an empty view. Then use only the keyboard to reach and activate the controls. Ask the agent to show these states in the running product, not merely list them in its final message. Canopy's current-main README describes preview screenshots and element annotations; check your installed release before depending on a particular capture control. A screenshot proves one moment, so repeat the interaction yourself when behavior matters.

## Connect the demo to the changed files

Ask for the branch, files changed, checks run, and remaining limitations. In Canopy, the Agent Workspace and Git views connect session work to diffs, commits, and PRs according to the published README. Open the final diff even if the demo looks right: accidental extra changes, generated files, or removed checks are easy to miss in a preview. For a team PR, give a reviewer the original acceptance checks and the final diff, not only the agent's summary.

- Inspect the final diff after the last demo or fix, because a later edit can invalidate an earlier screenshot.
- Separate “command exited successfully” from “the behavior passed my acceptance check.”
- Record any missing test or platform that was not exercised.

## Give feedback as a reproduction

If the demo fails, describe where you were, what you clicked or typed, what you expected, and what appeared instead. Attach a scoped screenshot if the problem is visual, and include console or service logs when relevant. Ask for a small correction, then repeat the same acceptance checks on the new result. A founder can own this decision without guessing how the code should be written. The final acceptance still belongs to the person responsible for shipping it.

## Practice resource

[Try the first agent practice project](https://canopyide.dev/first-agent-project.zip): The small task board includes an example request to add All, Open, and Done filters, plus observable acceptance checks.

## Copyable resources

### Request an inspectable demo

Replace the bracketed fields and give this to a coding agent.

````text
Goal: [one user-visible change]
Current behavior: [what happens now]
Expected behavior: [specific action and result]
Acceptance checks: [three to six observable steps, including an edge case]
Constraints: [what must not change]
Before editing: inspect the project and explain the approach.
After editing: run the app and give me the URL or exact opening steps, branch, files changed, checks run with results, and remaining issues. Demonstrate each acceptance check in the running product. Show the final diff. Do not call the task complete until I can repeat the behavior.
````

## Frequently asked questions

### What if I cannot read code?

You can still define and repeat user-visible acceptance checks. Have a developer or reviewer inspect the diff and technical risk before a production merge.

### Is a screenshot enough evidence?

A screenshot proves one visual state. Repeat interactions, refresh, keyboard use, and relevant edge cases in the running product.

### Does a running dev server mean the feature is done?

No. It means a command started. Verify the requested behavior, logs, checks, and final diff.

## Sources and further reading

- [Canopy README: services, Preview, screenshots, Agent Workspace, and Git](https://github.com/FluidWorksApp/canopy-ide/blob/main/README.md)
- [Canopy latest releases for version-specific behavior](https://github.com/FluidWorksApp/canopy-ide/releases)

## Related Canopy pages

- [How to set up Canopy and run your first AI coding agent](https://canopyide.dev/guides/getting-started-with-ai-coding-agents.md)
- [How do I find the command that runs an AI-built app?](https://canopyide.dev/guides/find-command-to-run-ai-built-app.md)
- [How do I test an AI-built UI before merging it?](https://canopyide.dev/guides/test-agent-built-ui-before-merge.md)
- [The coding agent says done, but the app is still broken](https://canopyide.dev/guides/agent-says-done-but-app-is-broken.md)
- [AI-built app works locally but fails on Vercel Preview](https://canopyide.dev/guides/ai-built-app-works-locally-fails-vercel-preview.md)

Canopy runs installed coding CLIs; CLI accounts, model selection, and provider billing remain separate. Check the installed release before relying on version-specific behavior.