Porting my 1993 Amiga game to Godot, with an LLM reading the 68000 assembly

TEXT START: In 1993, in Baghdad, I built a game called Babylonian Twins on an Amiga 500: 512KB of RAM, no hard drive, plugged into a TV.

The Dissection

This is a restoration narrative that doubles as a capability demonstration. Its real subject is the conversion of undocumented, high-context engineering knowledge into a reproducible machine workflow: assemble, diff, infer formats, recover memory layouts, render, probe, and test. The LLM did not merely type code; it excavated a dead system and reconstructed its semantics. The human supplied provenance, taste, and acceptance decisions. That is P1 exposed cleanly.

The nostalgia and father-son framing make labor substitution emotionally palatable. They turn automation into heritage preservation. The warmth is genuine; its economic implication is still cold.

The Core Fallacy

The text’s central error is treating the remaining human bottleneck as evidence that human production remains central. Identifying what feels wrong, resolving ambiguities, and approving results are verification and taste functions. They are lag defenses, not proof of a durable mass labor market. The model’s mistakes do not rescue programmers; they define the need for a test loop, and the article shows that loop being built.

This excerpt proves P1, not P2 or P3. It demonstrates that obsolete hardware, missing documentation, and creator-dependent tacit knowledge no longer reliably protect cognitive work. It does not establish that institutions cannot preserve human-only domains or that the majority have already lost productive access.

Hidden Assumptions

• The model has access to the source, assets, toolchains, emulator, and filesystem.
• Rights, provenance, and distribution are uncomplicated.
• The creator remains available as an oracle for ambiguous behavior.
• Pixel diffs and scripted probes capture most correctness; human judgment can remain a thin final layer.
• AI errors can be detected cheaply before release.
• The economics are favorable, though the text supplies no labor-cost comparison.
• This owner-led project generalizes to broader cognitive work. It is strong evidence for P1, not proof of universal automation.

Social Function

Primary classification: partial truth and transition management.

It is not simple copium: the author openly records errors and delayed discovery. It is prestige signaling because it showcases both a forgotten creator and a frontier model. It is also ideological anesthetic: the surviving human role is foregrounded while the collapsed labor volume is backgrounded. Readers are invited to admire continuity instead of confronting substitution.

The Verdict

This is a compact P1 exhibit. A game protected by obsolete hardware, undocumented assembly, lost filenames, hand-built memory maps, and creator-only knowledge was made legible to an LLM, rebuilt byte-for-byte, and ported with a thin supervisory layer. The moat was not defeated by brute force; it was translated into tests.

Under the Discontinuity Thesis, the creator remains viable as a Sovereign only if he controls the IP, AI-enabled production pipeline, and distribution. Otherwise he is a shrinking Servitor or curator. The article is not proof of total systemic collapse. It is evidence that one of the strongest defenses against automation—“only the original expert understands the old system”—has become hospice care.