One Developer Built Five App Ports with AI and a Shared Specification

What they studied

The authors report a case study in AI-assisted software development: a single developer built five working ports of a vector illustration application in approximately 120 evening hours. The ports span Rust, Swift, OCaml, Python, and browser-based platforms.

The approach combines AI-assisted implementation with two safeguards. First, a precise, executable YAML specification acts as the single source of truth. Second, parallel implementations provide a built-in differential-testing layer. In plain English, that means comparing implementations against one another to look for differences in behavior.

The authors frame this as a revival of N-version programming: an approach from the 1980s based on multiple implementations, which they describe as having been abandoned on cost grounds.

What they found

The concrete result is five working ports sharing a 23,000-line specification. Native code per port ranges from zero to roughly 95,000 lines, reflecting what the abstract calls the specification’s “escape hatch.” The abstract does not explain that mechanism in detail.

The authors argue that AI-paired engineering can make work feasible that would conventionally require multiple developer-years, provided both safeguards are in place. That is an argument drawn from this case study, not a controlled comparison establishing a general productivity multiplier.

The paper reports concrete artifacts and acknowledges limitations of its single-developer design. The abstract does not report defect counts, measured testing effectiveness, or a comparison with development without AI.

Why it matters for builders

The useful idea is not simply to ask an AI tool to write more code. It is to give that work an explicit behavioral reference and a way to cross-check implementations. For beginners, an executable specification can be understood as a description of expected behavior that software can use directly, rather than prose alone.

If you are considering multiple versions of an app, this case offers a workflow worth examining. It does not establish that every project needs five implementations, or that the reported development time will transfer to your app.

Source: "Jas: AI-Paired Engineering as a Revival of N-Version Programming" (2026-06-05, arXiv (Cornell University)). This is our summary of the paper's abstract — read the original for full details.