I need your clothes, your boots, and your motorcycle
LLM-powered coding agents are now capable of rapidly building complex systems such as PowerPC emulation cores for MAME, given a human with sufficient domain knowledge to guide and correct them. Commenters debate whether this means “software is solved,” arguing over how much of programming is just code generation versus deeper work like design, requirements, and accountability, and what happens to the joy and meaning of the craft when much of the mechanical effort is automated. The conversation broadens into questions about whether AI can create genuinely new ideas, how its capabilities compare to human invention, and whether current systems resemble unconscious but hyper-effective tools more than sentient “Skynet”-style AIs.
Nature of Software Work vs. “Coding Is Solved”
- Some argue coding is increasingly like “shoving logs into CNC machines”: low-skill prompting where agents do the mechanical work, eroding the craft and joy of programming.
- Others push back: software engineering includes requirements, architecture, debugging, testing, integration, and accountability; only the literal code-writing is being automated.
- Several note that agents still go on “wild goose chases” without careful prompts and domain knowledge, so human thinking remains central.
- One view: the “terrible cost” isn’t job loss yet, but the sense that the intellectually satisfying part of development is being hollowed out, making it feel like a “fake job.”
Do LLMs “Think” or Invent?
- Skeptics claim LLMs don’t think or create new things; they just remix training data and “vibe” based on scale and existing text.
- Others counter that humans also mostly recombine prior ideas; the hard part historically has been implementing combinations, not imagining them.
- Debate over whether inventions like electricity, RF communication, lasers, and transistors are genuinely novel or just incremental on centuries of work.
- Some note that giving agents feedback loops (e.g., compiling and fixing their own code, interacting with the physical world) makes “invention-like” behavior plausible.
Agentic Coding in Emulation and Retro Projects
- The article is cited as a strong example of a coding agent building a PowerPC dynamic recompiler for MAME under human guidance.
- Another linked series describes an AI-assisted, largely unsupervised, cycle-accurate Intellivision emulator.
- A separate project is adding a JIT to classic Mac emulators, with humans still handling tricky hardware-register behavior.
Nostalgia and Classic Mac Tools
- Discussion contrasts today’s Grapher.app with the original Mac Graphing Calculator, highlighting its “legendary” guerrilla-development backstory.
- Some lament that such passion-driven, rule-bending efforts feel rarer in modern large tech companies.
Access and Meta Issues
- Multiple readers report 403 Forbidden errors, seemingly region/ISP-dependent; others confirm access via normal browsers, Tor, or web archives.
- A few complain that the title gives no hint the article is about emulation.
Terminator 2 and AI Futures
- The title line is traced to Terminator 2; many say the film’s effects and story have aged well, noting its heavy use of practical effects.
- Some argue our real trajectory resembles fiction about hyper-capable but unconscious systems more than self-aware Skynet; others insist current dangers still stem mainly from human actors.