Planning layer
Specifications made intent operational
OpenSpec converted proposed changes into explicit requirements, design decisions, and acceptance criteria before implementation began.
Developer case study
Cairn is a spec-driven build for a modern interactive-fiction player
Cairn is a native macOS player for classic Z-machine interactive fiction. This page focuses on the development practice behind the app: using explicit specifications and agentic AI to ship a real product while keeping product direction, architecture, review, and testing accountable to a human developer.
The project
A focused desktop experience that combines classic parser-driven stories with a native workspace for play, maps, and reference materials.
Product context
A real product made a useful proving ground
Cairn had to preserve the deliberate pace of classic interactive fiction while making long play sessions practical on a modern Mac. That gave the work concrete product and technical constraints: a reliable native workspace, a compatible interpreter, and interfaces that support the story without competing with it.
Those constraints made the development method meaningful. The work was not a demonstration of AI output; it was an ongoing effort to turn a clear product point of view into reviewed, testable software.
Development approach
An Implementer experiment with agentic AI
Cairn was built as a deliberate experiment in using agentic AI within a real software-development lifecycle. OpenSpec provided a planning layer for moving work from intent through requirements, design notes, acceptance criteria, implementation, review, and hands-on testing.
Implementation
Agents worked from defined constraints
Clear specifications gave agents a practical target: what to build, why it mattered, and how the resulting change would be evaluated.
Accountability
Developer judgment remained central
Product direction, architecture, prompting, code review, and testing against the acceptance criteria stayed under direct human ownership.
Implementer framing
Tools supported craft rather than replaced it
Infocom called its creators Implementers. That framing fits this project: the job is to guide tools toward software that is coherent, useful, and genuinely finished.
What this demonstrates
AI-assisted development needs a disciplined delivery system
Cairn is evidence that agentic workflows can contribute to a production application when they operate inside clear product intent, explicit specifications, human review, and real testing. The value comes from the system around the agents as much as from the agents themselves.