THE MEDIA FOR AI THAT ACTUALLY HELPS.English edition · Understand & take action
Spotted on the web

Require a code agent to demonstrate what he built

Use described publicly by simonwillison.net (February 2026), summarised by the desk. Not tested by iadujour.

The result to aim for

To verify the work of his agents, Simon Willison created two tools that make them produce a demonstration document, orders and captures in support, and he developed them from his phone.

What was done

In his 10 February 2026 post, Simon Willison presents Showboat, which has the agent build a Markdown document showing step by step what the code does (executed commands, outputs, images), and Rodney, a browser automation tool designed to be learned by an agent via his online help. Both projects were launched with Claude Code for the web from the iPhone application; He believes that the majority of the code he publishes is now written by agents piloted from his phone.

Before you start

Tool cited by the source: Claude Code. Fill in the bracketed items: [fonctionnalité]. Strip confidential data that is not needed and check your tool’s access terms. Features and plans can change.

The method, in three steps.

  1. Gather the context

    Spot the bracketed items in the prompt ([fonctionnalité]) and prepare what they ask for.

  2. Adapt the prompt

    Copy the prompt below, replace the bracketed items, then paste it into Claude Code.

  3. Reread before using

    Reread the result, correct it if needed and keep the final decision. Caveats are detailed below.

Your starting point.

Replace the bracketed items, then adapt the proposal to your situation. This prompt is written by the iadujour desk It is not a quotation from the source.

The prompt to customise
You just finished [fonctionnalité]. Before I reread the code, produce a demo.md file that proves it works:
- the exact commands you execute, followed by their actual output (no invented output)
- at least one case of error and the behaviour obtained
- a screenshot if it is an interface
- what you couldn't test, and why
If a command fails, leave the failure in the document and then correct.

What deserves your attention

A demo generated by the agent remains to replay yourself on a real case before deploying.
Origin of the case. Use described publicly by simonwillison.net (February 2026).
Proof level: Testimony published by an identified source. Source checked on 10 November 2026. Use described by the source, not tested by the desk. Original summary written by iadujour.fr: figures and claims are the source’s; the prompt is a desk proposal, not a quotation.
TO GO FURTHER

Other ideas to put to work.

The whole Dev theme
Dev

Make a pull request by an agent before the human review

A first automated passage to identify probable bugs, missing tests and safety risks, so that the human review focuses on the essentials.

Tool: Claude CodeType: Practical case
DISCUSSION

Comments.

A field note, a question, a precision on this use case? Comments are reviewed before publication.

Published comments

Loading comments…

    Leave a comment

    Moderation charter. Every comment is reviewed before publication. We refuse hate speech, insulting or discriminatory language, advertising, spam and promotional links, off-topic posts and other people’s personal data.