Using Claude to re-create extinct software – Matlab from the 1980's

A Vintage Computer Federation forum post describes using **Claude** to re-create extinct software: a **Matlab** from the **1980s**. The thread is also listed on Hacker News under a **Claude/Codex/Fable** tag, with the comments page at item id **49008353**. At scrape time the story had **1** point and **0** comments, so the write-up is early on the public discussion curve rather than a heavily scored HN debate.

The technical claim is reconstruction by LLM, not recovery of a preserved binary or a line-for-line port from a surviving source tree. **Matlab** in that era was a matrix-oriented interactive environment: command-driven linear algebra, array-first data, and a REPL-style workflow that later commercial Matlab generalized. Recreating “extinct” software with **Claude** means the model is asked to restore behavior and surface that no longer ship as maintained products—syntax, defaults, and numerical habits of an old release—rather than to implement a modern Matlab clone with current toolboxes and licensing.

For engineers and builders, the useful signal is that generative models can act as a **software archaeology** tool when primary artifacts are gone. Lost internals, old demo code, and discontinued scientific packages are common in labs and archives; a model that can propose a working facsimile shortens the path from “we only have screenshots and manuals” to “we have something to run and compare.” That does not replace verification against known numerical results, papers, or remaining manuals, but it changes the starting cost of revival work.

In market terms this sits next to other **Claude**, **Codex**, and **Fable**-tagged experiments that treat large models as generators of legacy-compatible systems rather than only as chat assistants. Commercial Matlab still defines the mainstream matrix-computing product line; a 1980s-style recreation is not a competitor to MathWorks’ current stack. The competitive angle is among AI coding stacks: who can sustain long, constrained rebuilds of dead products with enough fidelity that vintage-computing and numerical-methods communities take the result seriously.

Practical takeaway: treat this thread as a case study in **prompted reconstruction** of extinct scientific software, not as a drop-in Matlab substitute. Watch for whether the author publishes runnable code, test vectors against historical behavior, and clear limits of what **Claude** got wrong. If those appear, the same method—model-assisted rebuild plus hard checks—applies to other lost tools where only partial documentation survives.