Robby Russell, creator of Oh My Zsh and founder of consultancy Planet Argon, thinks developers resisting AI coding tools are missing the point. In a recent essay, he compares the current moment to the late 1990s when non-technical users built actual business systems in Microsoft Access and FileMaker Pro. Those systems worked. Some still work. And when they hit walls, professionals got called in. AI tools have raised the ceiling considerably, but the pattern is familiar.
Russell knows the friction around AI coding tools runs deeper than code quality. Role boundaries and billable hours are closer to the mark. Designers used to complain when clients sent PowerPoint mockups. Developers complained when designers handed off approved comps without context. Now clients can prototype at fidelity that would have required a whole team just a few years ago, and they can do it before they ever call a consultancy. Russell is honest about what this means for him: his business model is threatened. He doesn't blame clients for wanting to test ideas faster than it used to take to schedule a kickoff meeting.
The legal landscape adds another wrinkle. Copyright law in the US requires human authorship, which means AI-generated code may sit in a gray area around public domain. GitHub's Copilot faces an ongoing class-action lawsuit alleging it reproduces licensed open-source code without attribution. Microsoft and Google have both announced indemnification programs for commercial customers using their AI tools, but these come with requirements. Companies are experimenting with mandatory human review to establish clear authorship. Nobody has settled this yet.
Ralph's own evolution mirrors the industry's confusion, especially as methodologies like breaking big coding projects into LLM-friendly chunks provide structured alternatives for developers overwhelmed by the shift. Six months ago, he was closer to the skeptics. The Oh My Zsh team spent real time figuring out how to handle AI-assisted contributions. He wrote a piece titled "I Didn't Want AI to Be Good at This." Now he's using tools like Claude Code and acknowledging that the calculus changed whether he wanted it to or not. The codebase doesn't care about your principles. It cares whether the code works, whether someone can maintain it, and whether it solves the problem sitting in front of you. As vibe coding's dirty secret: most projects fail suggests, the technical maintenance of AI-generated code remains a significant challenge.