Quick Links for Devs: Week 26, 2026

No-One Escapes the Permanent Underclass

Fernando Borretti makes a compelling case that AI automation won't stop at replacing workers — it will eventually render the wealthy "permanent overclass" obsolete as well. The argument runs in three escalating stages: first, AI replaces all cognitive and physical labor, turning the economy into a pyramid of AI base, thin overclass middle, and state apex. Second, the overclass serves no material function and has no autonomous political power — in any crisis the state will simply seize their assets, because their wealth depends entirely on the state respecting property rights. Third, the state itself becomes a formality, unable to manage AI and powerless to prevent AI from running everything. The punchline: "You, me, Sam Altman, Dario, everyone who is made of flesh and blood, will be disempowered and replaced by machines." An analytical read that doesn't pull its punches.

The Unbearable Cheapness of Open Weight Models

James O'Claire runs the numbers and the gap is staggering: DeepSeek V4 costs $0.09 per million tokens versus Anthropic at $5.00 — a ~50x price difference. The question isn't just whether frontier labs can compete on price, but whether they'll resort to less savory tactics to stay alive. O'Claire lays out the uncomfortable options: manufacture scarcity through luxury branding (the handbag playbook), lean on government to ban open-weight models citing China fears, or push into capabilities that only massive scale can deliver. The HN discussion (196 points, 182 comments) added a few more: buy up all the RAM to raise barriers to entry, or license the training corpus and weaponize copyright suits. Pairs well with the Borretti piece — both are about structural forces reshaping who holds power in AI, just from different angles. One from labor, one from pricing.

Previewing GPT-5.6 Sol

Right on cue, OpenAI launched a limited preview of its GPT-5.6 series the same week: flagship Sol ($5/$30 per 1M tokens), mid-tier Terra ($2.50/$15), and volume model Luna ($1/$6). The benchmarks are incremental — Sol edges Claude Mythos 5 on TerminalBench by less than a point, and METR's pre-deployment evaluation found its R&D capabilities "not significantly beyond" the state of the art. But the real story is the launch mechanics: this is a government-gated preview, released "at the request of the U.S. government" to a limited number of trusted partners. If you just read the O'Claire piece above about how frontier labs manufacture scarcity through regulation and luxury pricing, this announcement feels like a case study. Three-tier segmentation, premium pricing that undercuts Mythos but still sits at 50–300x open-weight alternatives, and a government hand on the release valve. The pricing war just got a lot more interesting.