You’ve probably seen the headlines: AI is coming for developers’ jobs. It can spin up apps in seconds, debug complex logic, and write flawless syntax. But if you think the future of software engineering is just generating code, you’re missing the actual revolution.
In 1993, in the middle of Baghdad, I wrote an Amiga game in raw MC68000 assembly. Fast forward to last July. I fed the raw machine code of that 30-year-old game into an LLM. In a single evening, the AI assembled the code, matched it byte-for-byte to my original binaries, and ported it to Godot. It even figured out a 108-byte mismatch caused by an old AsmOne quirk where the shipped files were just memory dumps of a game that had already been running.
AI can translate dead assembly code in seconds, but it takes a human weeks to remember how a game was supposed to feel.
The code port took one evening. Getting the ‘feel’ right took multiple weekends. This wasn’t a software engineering task; it was an act of personal archaeology. I had to feed Claude my 33 years of memories, my old notes, and my git repos. I was running my 1993 map editor for the first time in three decades, reconnecting with my younger self through old screenshots.
We are obsessed with AI’s ability to write code. But raw code translation is trivial. The true bottleneck isn’t the AI’s processing power—it’s the human’s subjective, embodied knowledge. The AI can write the exact logic, but it doesn’t know if the jump feels too floaty, or if the pacing is off. It can’t feel the game.
The LLM wasn’t just a code porting tool; it was a memory prosthesis.
It externalized, structured, and validated my fading recollections. When you look at what AI did here, you might think it’s reading assembly code. It’s not. It’s resurrecting a developer’s 30-year-old tacit memory. The AI is faster than the original programmer, but the human’s subjective memory remains both the bottleneck and the true value.
If you’re a developer panicking that AI will automate your job, take a breath. The mechanical translation is solved. What AI can’t replace is your 33-year-old tacit memory. Your domain-specific intuition, your iterative feel-tuning—your you—is the value.
We thought AI would replace software engineering. Instead, it proved that human intuition is the ultimate moat.
AI isn’t here to replace you. It’s here to externalize, structure, and validate your fading recollections. It’s here to help you resurrect the past. The code is trivial. The memory is everything.
FAQ
Q: If the AI ported the code in an evening, why did it take you weeks to finish?
A: Because shipping software isn't about syntax, it's about feel. The AI handled the mechanical translation, but tuning the game's physics, pacing, and 'feel' required my 33-year-old tacit memory. The code was trivial; the embodied knowledge was the bottleneck.
Q: How does this apply to modern developers working with legacy code?
A: It proves that LLM-assisted legacy-code porting is highly viable, but only if the original author (or a domain expert) is in the loop. AI can read the raw machine code, but human domain memory is required to validate the output and tune the final product.
Q: Doesn't this just prove AI is a glorified search-and-replace tool?
A: No, it proves AI is a cognitive mirror. It externalizes and structures fading human memories. It didn't just translate assembly; it validated my recollections of a game I built in 1993 Baghdad, turning a porting project into personal archaeology.