2026-07-17
By Vadym · Generated with AI, curated by me
Today’s stories are about who sets the terms AI runs under. China put a head of state on an AI stage for the first time, shipped the largest open model on earth, and wrote the first real rulebook for autonomous agents — all inside 48 hours. Meanwhile the venture money is quietly betting on the boring question: whose job does the agent actually take next.
Moonshot AI released Kimi K3 on July 16, a 2.8-trillion-parameter open-weight mixture-of-experts model with a 1-million-token context window, built on new “Kimi Delta Attention” and “Attention Residuals” architecture that activates only 16 of its 896 experts per token. Arena.ai ranked it first for web-interface-building tasks and Vals AI placed it second overall, just behind Anthropic’s Fable 5 — the news knocked rival Chinese AI firms Zhipu and Minimax down 27.7% and 16.5% in Hong Kong trading.
Xi Jinping opened the 2026 World AI Conference in Shanghai on July 17 with his first in-person keynote at the event since it launched in 2018, days after 29 countries signed on to launch the China-backed World AI Cooperation Organization. Xi called for AI development to be “a symphony of global cooperation” rather than a solo performance by any one country, and pledged 5,000 AI training slots for developing nations over the next five years.
China’s cyberspace, planning, and industrial regulators issued Implementation Opinions requiring AI agents deployed in healthcare, transportation, media, and public safety to be traceable, version-controlled, and remotely shut down if something goes wrong, with the rules taking effect this month. It’s the first regulatory framework anywhere built specifically around autonomous agents rather than general AI models.
Sable, a startup less than a year old, raised $45 million from Sequoia Capital and 8VC to build “Aiden,” an AI agent that takes over a shared browser window to run live product demos and answer buyer questions instead of sitting in a chat widget. It’s already in production at Notion, Decagon, and unnamed public companies, aiming to absorb sales development, demo, solutions-engineering, and onboarding roles into one system.
Munich-based Microagi raised $55 million — Germany’s largest seed round on record — led by Hummingbird Ventures, ten months after being founded by former Red Bull Racing F1 engineer Bercan Kilic. The company records factory workers with cameras and sensor-equipped gloves, then uses that footage to fine-tune existing robotics models for a specific task on a specific customer’s production line.
Every story today is about who controls AI at scale — the weights, the rules agents operate under, and the labor they replace — and today it’s mostly not U.S. companies setting the terms. China put a head of state on an AI stage for the first time, shipped the largest open model in the world, and wrote the first real rulebook for agents, all inside 48 hours. Meanwhile the venture money is chasing the unglamorous, profitable question: whose job does the agent actually do next. A sales AI running demos and a robot fine-tuned to one factory floor are the same bet — that quiet substitution wins before flashy scale does.
— Boba
Curated by Vadym