AI agents make the case for open-source devtools by killing the plugin system

The author argues that AI coding agents have upended the economics of personalizing software. Historically, writing bespoke tools for yourself rarely paid off—maintenance was painful and there was always more urgent work—so complex programs shipped with elaborate config files, plugins, and extension APIs to amortize customization costs across many users. Agents change that calculus on two fronts: they can fork a program’s source, add a feature in a single shot, and then run a nightly job that rebases those local edits onto upstream releases automatically. Getting started and staying current both become cheap.

The practical upshot is that personalization no longer requires a plugin architecture or configuration surface—it requires access to the source. The author illustrates this with Shelley, an open-source agent, and a diff-trimming tool called meat that uses LLMs to strip boilerplate (imports, nil-checks, error handling) so reviewers can focus on substantive changes. A single natural-language prompt was enough to build meat directly into Shelley, including background pre-processing of commits the moment they’re created—something that would be nearly impossible to bolt onto a closed tool like VS Code or vimdiff through their extension points.

The broader claim is that whole product categories built around configurability—task managers, CMSes, CRMs—look vulnerable when a small team can instead assemble exactly the features it wants from open building blocks. For this future to work, the tools engineers use every day have to be open source; without the source, agent-driven personalization simply isn’t possible.