Open Source and GenAI?

Tim Bray reflects on his personal experience using Claude (via a collaborator) to improve his open-source Quamina project, then weighs whether GenAI in software development deserves the same condemnation as GenAI broadly. He acknowledges the societal harms of GenAI — environmental cost, job displacement, inequality — but argues software development is a special case: the market is too small to meaningfully worsen carbon load or create extractive monopolies, and open-source LLM tooling will likely prevent lock-in. However, he raises serious concerns: code review remains a bottleneck that AI doesn't speed up, developer burnout is increasing, productivity gains are unproven at scale, and capitalism's incentive structure means workers won't benefit from efficiency gains. His conclusion is cautious: use LLMs for well-scoped, non-strategic tasks with rigorous review, but don't abandon engineering culture for vibe coding.