Codeplain champions spec-driven development as the antidote to AI-era code maintenance

AI is producing code faster than teams can review it. But instead of accelerating review cycles, a growing contingent of developers is proposing something more radical: replace code review with spec review entirely.

Codeplain, founded in early 2025 by Dušan Omerčević and Predrag Radenkovic, is betting hard on this premise. The company has built tooling around Plain, an open-source specification language designed to serve as a single source of truth for software behavior. The idea is deceptively simple: when code breaks or needs updating, you edit the spec, not the code, and regenerate the implementation from scratch.

Specs over code#

Omerčević makes a counterintuitive claim: developers actually prefer reading specs to writing them. Through user testing, the team found that while specs felt burdensome to author, they were easier to reason about and review than the generated code they produced. This tension led Codeplain to develop plain-forge, an open-source agentic framework that lets AI agents draft and maintain specs conversationally, automating what was previously the most labor-intensive part of the process.

"Spec-driven development was really hard to communicate to regular developers," Omerčević told The New Stack. "What Chad did really well is ground the story in code, saying that the only difference is that this code is no longer permanent—it's ephemeral, and it can always be regenerated from other artifacts."

He's referencing Chad Fowler, a BlueYard Capital partner who's been articulating a broader philosophical framework called Phoenix Architecture—named for the mythological bird that cyclically burns and rebuilds itself. Fowler's argument centers on provenance debt: when developers manually patch AI-generated code, they lose the reasoning and decisions behind that change, accumulating hidden technical complexity over time. Specs, by contrast, preserve intent.

Economic and practical wins#

Codeplain already has early adopters. Incode, an identity verification platform, uses it to maintain integrations with external data providers—work that typically involves constant API research and frequent breakages. Because specs encode intent rather than implementation, when an external API shifts and breaks integration code, Codeplain can often regenerate a fix from the unchanged spec without manual tweaking.