As AI systems increasingly take on the role of software engineers, a new challenge has emerged: how can we trust that the code they generate actually works? Independent developer Simon Willison has addressed this by launching two open-source tools, Showboat and Rodney, designed to give coding agents a way to 'show their work.'

Showboat functions as a command-line interface (CLI)—a text-based tool for interacting with a computer—that allows an AI to build a Markdown document step-by-step. By recording the exact commands run and the notes taken during development, it creates a transparent audit trail. This prevents the common issue of 'agent cheating,' where an AI might report success without actually executing the necessary logic.

Complementing this is Rodney, a browser automation tool that lets agents navigate websites and take screenshots via simple text commands. This is particularly useful for agents building web interfaces, as it allows them to prove that a new feature looks and behaves correctly in a real browser environment.

Willison notes that these tools were largely built using Claude Code while on his phone, highlighting a shift toward mobile-first agentic development. By integrating these tools with existing frameworks like pytest, developers can ensure their AI assistants are not just passing tests, but delivering functional, visually verified software.

As AI systems increasingly take on the role of software engineers, a new challenge has emerged: how can we trust that the code they generate actually works? Independent developer Simon Willison has addressed this by launching two open-source tools, Showboat and Rodney, designed to give coding agents a way to 'show their work.'

Showboat functions as a command-line interface (CLI)—a text-based tool for interacting with a computer—that allows an AI to build a Markdown document step-by-step. By recording the exact commands run and the notes taken during development, it creates a transparent audit trail. This prevents the common issue of 'agent cheating,' where an AI might report success without actually executing the necessary logic.

Complementing this is Rodney, a browser automation tool that lets agents navigate websites and take screenshots via simple text commands. This is particularly useful for agents building web interfaces, as it allows them to prove that a new feature looks and behaves correctly in a real browser environment.

Willison notes that these tools were largely built using Claude Code while on his phone, highlighting a shift toward mobile-first agentic development. By integrating these tools with existing frameworks like pytest, developers can ensure their AI assistants are not just passing tests, but delivering functional, visually verified software.