Json House

RTX GPU unlock doubles some tensor math on RTX 20, 30 and 40 cards, yet token generation stays flat. Which local LLM work speeds up, and when a Mac still wins.

Solo developers have a window: skills cut build costs now, but platforms absorb features fast. Who survived absorption, and what to own before AGI arrives.

Agentic AI 2026: in four weeks Google, OpenAI and Anthropic shipped agent models at $2/$10, cut cache prices, moved into AWS, and OpenAI cut Pro allowances.

Anthropic named seven Chinese labs running industrial-scale Claude distillation in 2026. What they took, what they could not, and where China already leads.

GPT-6 Astra costs 2.5x GPT-5.6 Sol per token yet 43% less per BenchCAD task. We invert every published cost claim into the token ratio it implies.

A plugin packages capability; MCP connects to it. What separates the two, what each keeps resident in your context window, and why most useful plugins go unfound.

6G and AGI converge because capital forces it: telecom capex falls 2% in 2026 while five AI firms spend over $600B. One company already owns both layers.

CoreWeave raised compute prices 25% in July, citing component costs. Token prices fell anyway. Five snapshots, 48 models, and the gap capital is paying for.

Cloudflare gave AI agents wallets. We priced our own measured crawler traffic against it: 5,227 crawls in 20 days, and what a per-request toll would actually pay.

Twenty days of first-party Cloudflare logs: AI crawlers made 5,227 requests to this site and sent back 7 human visitors. Full dataset, method, and limits.