Devtools Must Be Open Source: The AI Personalization Era

In a blog post dated August 2, 2026, exe.dev co-founder David Crawshaw argues that open source is now a prerequisite for developer tools, given how AI agents have transformed software personalization. Five years ago, Crawshaw notes, most engineers he spoke with had written little or no software for themselves — the maintenance burden and opportunity cost made it impractical. That calculus has changed dramatically. Crawshaw identifies two key agent prompt patterns that underpin this shift: first, downloading and building source code locally while recording the motivation for changes in version control; second, running a nightly cron job that fetches upstream changes and automatically rebases local modifications on top of them. Together, these patterns lower both the entry cost and the ongoing maintenance cost of customizing software. Crucially, Crawshaw notes that these workflows can be embedded directly into an open source agent itself as a 'skill' — a block of text instructions the agent can discover and execute. His company has implemented this in their agent 'Shelley,' allowing users to personalize it with plain-language prompts such as 'make Shelley's UI high-contrast.' He also describes a personal side project, meat.dev, focused on AI-assisted code review, noting that model improvements have reduced the need for human reviewers to manually check edge cases. The core thesis: if devtools are closed source, agents cannot build, modify, or maintain them — making open source a structural requirement in the AI-agent era.