This reads less like a principled stand and more like a community drowning in low-effort AI spam and reaching for the bluntest possible filter. The rule is also practically unenforceable — AI assistance is already woven into everyday dev workflows, and checking code provenance is trivially easy to evade. The 'non-English speakers' clause betrays the real problem: moderation capacity, not ideology. In our view, this is an early symptom of a coming split between 'human-made' and 'AI-assisted' open source, and we expect more communities to experiment with provenance labels rather than blanket bans. Outright exclusion may feel satisfying, but it risks shrinking the contributor pool exactly when open source projects need hands the most.

Sharka runs an enterprise AI company, so the framing naturally flatters his own market — but the core thesis holds. Procurement logic (liability, auditability, data residency) genuinely rewards a different vendor than benchmark leaderboards do, and the Swedish decision reads as a clean signal that sovereignty concerns now outweigh raw model quality in institutional buying. The weak spot: 'most deployable stack' is a moving target, since OpenAI, Anthropic and Google are bolting enterprise controls onto frontier models fast. The moat for specialists is compliance depth, not distribution. Expect the two races to partially re-merge at the top of the market — and diverge hardest in defense and regulated infrastructure, exactly where he points.