We need a clearer framework for AI-assisted contributions to open source

AI, staffing, and productivity

• Some report significant productivity gains: fewer engineers needed, faster feature delivery, happier remaining staff who can focus more on product and design than raw coding.
• Others push back: code-writing is only part of engineering. Architecture, systems thinking, protocols, rollout strategies, and clear specs still require experienced engineers.
• Skeptics question whether reduced headcount just shifts more workload and future tech debt onto a smaller team, with management incentivized to “get theirs” before long‑term issues hit.

Code vs specification

• Several argue that code has become cheaper while high‑quality specifications and tests have become more valuable.
• LLMs can generate plausible code but still require humans to define problems correctly, set constraints, and own the results.

Open source, “slop” PRs, and contribution norms

• Many maintainers see AI-generated drive‑by PRs as noisy, under‑tested, and lacking ownership.
• Format/convention issues are seen as the easy part; the real problem is complexity without thought, tests, or long‑term responsibility.
• Suggestions:
  • Stronger contribution guidelines plus automated linters/CI.
  • Treat LLM code as prototypes or demos, not merge‑ready PRs.
  • Limit large PRs from new contributors; encourage starting with small bugs.

Policies: bans, disclosure, and reputation

• Hard “no AI” rules are seen by some as useful filters but fundamentally unenforceable; good AI‑assisted code is indistinguishable from human code.
• Others prefer disclosure policies: reviewers spend minimal time on AI‑generated PRs and more on human‑written ones.
• Ideas floated: reputation/“credit scores,” staged trust levels, triage volunteers, monetary or other “proof of work” barriers; concern exists about raising entry barriers and eroding anonymous contribution.

LLM capabilities, hype curve, and “alien intelligence”

• Several feel the “replace all developers” hype is fading; tools are settling into roles as powerful assistants, especially for debugging and small, local changes.
• Others argue improvement is still on a trajectory toward broader automation, though returns may be diminishing.
• The “alien intelligence” framing resonates: LLMs can be simultaneously brilliant and disastrously wrong, and must not be anthropomorphized or trusted without verification.

Prototypes, hackathons, and slop culture

• Multiple commenters link AI‑slop PRs to a broader culture of low‑effort prototypes and “innovation weeks” that morph into production systems, creating brittle architectures and long‑term pain.
• More generally, the near‑zero cost of generating code, plans, and “research” amplifies asymmetry: it’s cheap to produce slop, expensive to review it—making norms, reputation, and aggressive triage increasingly critical.