Tailscale's Shelley shows why devtools must be open source

Five years ago, few engineers wrote software for themselves—config files and plugins were the norm because the maintenance cost of custom code outweighed its benefits. Tailscale's post argues that AI agents have changed that equation. By issuing two simple prompts to an agent—"download the source and build it locally" and "set up a nightly cron job to fetch upstream and rebase local changes"—users can now personalize open-source tools and keep them synchronized with upstream releases automatically. The agent handles the grunt work of merging, testing, and replacing the installed version, slashing both the startup cost and the ongoing maintenance burden.

Tailscale has embedded this capability into Shelley, its open-source AI agent. Shelley can load personalization instructions as skills, so a user can simply say "make Shelley's UI high-contrast" and the agent edits its own source code, rebuilds, and installs the change. The post also walks through concrete example: meat.dev, a tool that uses LLMs to strip boilerplate from code diffs (imports, nil-checks, error handling) so a human reviewer can focus on architecture and edge cases. The author wanted to run this tool inside Shelley's UI and avoid waiting minutes for the LLM to process diffs. With agent-driven personalization, that kind of bespoke workflow becomes trivial to build and maintain—arguing strongly that devtools must remain open source for agents to operate on them.