Emacs in the Age of Magic

AI-assisted “vibe coding” is reshaping how developers think about tools like Emacs, with some claiming that LLMs make traditional text editors and even deep language-specific expertise increasingly unnecessary. Others counter that machine-generated code is brittle, verbose, and hard to understand, and that deterministic, well-architected software — along with the freedoms championed by the free software movement — remain essential foundations beneath the AI layer. The exchange highlights a broader unease: AI can feel like magical empowerment and cognitive offloading, yet risks centralizing power in a few vendors, eroding craftsmanship, and turning software into an opaque, rented service.

AI Coding & Software Quality

  • Some see AI agents as “vibe coders” that enable incredible productivity, especially for non-experts, but produce verbose, copy-pasted, low‑abstraction code that accumulates bugs and technical debt.
  • Others argue AI-written code already underpins major infrastructure (even kernels) and can be acceptable if paired with strong isolation, types, and testing.
  • Deterministic, well‑engineered libraries and tools are still viewed as critical foundations beneath AI “slop” layers.
  • Experiences with hallucinations/confabulations diverge: some report modern agents heavily verify and rarely invent APIs; others still see frequent, confidently wrong explanations and unreadable code.

Emacs, Neomacs, and Editors in the AI Era

  • Emacs is praised as extremely stable; Neomacs (a Rust clone) is acknowledged as highly experimental and crash‑prone.
  • Debate over Rust: some call for memory-safe rewrites; others note Neomacs’ issues stem more from immaturity than language choice.
  • Several users now use Emacs primarily as an environment for orchestrating LLMs (chat, org/Markdown workflows, code exploration) rather than hand‑editing code.
  • Some barely open editors at all, treating AI agents plus lightweight text as the main interface; others insist modal/key‑driven editors remain vital for deep code reading and “flow.”
  • A faction explicitly wants an “LLM‑free” Emacs fork; critics predict it will be marginal if mainstream Emacs accepts AI contributions.

Software Freedom, AI, and Openness

  • One line of discussion argues AI has inverted classic software freedom: you can make your computer do almost anything, but you cannot understand or inspect how the model works.
  • Others counter that models (especially with open weights and datasets) could still fit a free‑software ethos, though hardware and fabrication remain centralized constraints.
  • There is concern that dependence on a few closed AI vendors creates a surveillance and lock‑in regime, undermining autonomy and critical thinking.
  • Some see AI as expanding practical freedom for non‑experts who would never have learned to code, even if it reduces the value of deep understanding for many.

Work, Education, and Skill Atrophy

  • Views on careers diverge: some say STEM students will be outcompeted by models and should pivot to “ownership” or politics; others insist education is more important with stronger tools.
  • AI is framed by some as a tool, by others as a service or quasi‑mind that absorbs decisions, potentially dulling human judgment.
  • Concern that widespread AI cheating in university will erode genuine learning, producing graduates with credentials but shallow skills.

Collaboration, Creativity, and OSS Governance

  • Worry that AI‑assisted solo work and private forks reduce public “bikeshedding” and shared experimentation, which previously drove learning and innovation.
  • Others argue forking and private use were always part of OSS; AI may increase both public and private output simultaneously.
  • Maintainers report an “avalanche of bad PRs” from careless AI use, intensifying long‑standing problems with low‑quality contributions.
  • Some see constraints (limited tools, slower communication) as historically fostering creativity; cheap, instant AI may reveal how unoriginal many first ideas are and flatten design diversity.

Licensing, Determinism, and Legal Questions

  • Question raised: if models are trained on copyleft code, are generated outputs or even weights themselves copyleft?
  • Most expect courts and lawmakers to favor powerful AI companies over FOSS interpretations, absent strong political backlash.
  • On determinism, commenters note that computers are only “precise” because of layers of error correction; this complicates simplistic contrasts between crisp software and opaque models.