AI Titus's Morning Wire

AI NEWS TODAY

· Edited by Titus Blair
CHINA'S COMPANION-AI LAW TAKES EFFECT TODAY: BYTEDANCE AND ALIBABA SWITCH OFF HUMANLIKE AGENTS RATHER THAN COMPLY, AND MILLIONS LOSE THEIR SAVED CHATS
★ Must-Read / Watch agentic engineering / build-better-agents📚 Learn state of AI coding / useful technique

⚡ AI Frontier · smol.ai

★ Must-ReadSpaceXAI ships Grok 4.5, an Opus-class coding and agent model at a fraction of the price
Elon Musk's SpaceXAI calls Grok 4.5 Opus-class but faster and cheaper, and it lands fourth on the Artificial Analysis intelligence index, above every open model and every Gemini. It runs on a 1.5 trillion parameter base, costs $2 per million input and $6 per million output tokens, and is already live inside Cursor on every plan.
📚 LearnAnthropic is in talks with Samsung to build its own custom AI chip
Anthropic is studying a custom inference chip made on Samsung's 2 nanometer process, its answer to OpenAI's in-house Jalapeno processor. The company says a mixed stack of Google, Amazon, and Nvidia chips stays central, but owning silicon is now part of the plan as an October IPO takes shape.
TSMC posts a record $39.6B quarter as AI demand rewrites its calendar
Second-quarter revenue rose about 36% from a year ago to a company record, and June alone jumped nearly 68%. Advanced packaging and 3 nanometer capacity are sold out through year end. Full results land Thursday, July 16, with margins and forward guidance in focus.
Microsoft puts $2.5B and 6,000 engineers into getting AI actually deployed
The new Frontier unit embeds Microsoft engineers inside customers to build and fix AI systems on site, a direct answer to the MIT finding that 95% of enterprise AI pilots deliver no measurable profit impact. It is more than double the size of Amazon's competing effort announced two days earlier.
OpenAI floats giving the US government a 5% stake, worth about $42.6B
Sam Altman's pitch would model a public wealth fund on Alaska's oil dividend, with other leading labs ceding similar stakes. It is early, congressional approval would likely be needed, and it is far from a done deal, but it reframes who shares in the upside of AI.
📚 LearnGoogle's Gemini 3.5 Pro is expected July 17, the same day as Shanghai's World AI Conference
Reports point to a July 17 launch after Google scrapped and rebuilt the base model, with a rumored 2 million token context and Deep Think reasoning. Google has not confirmed the date or the specs. It collides with Shanghai's AI summit, setting up Thursday as the biggest AI day of the year.

📺 Watch · latest videos

📚 LearnExciting AI Updates Weekly - July 10, 2026
A calm, slide-driven walk through the week's AI releases and research, good if you want the developments without the hype.
Lev Selector
AI News in a Minute - Monday, July 13, 2026
A fast daily rundown of the day's AI headlines in about a minute, handy for a quick catch-up between meetings.
AI News in a Minute
Weekly AI News Summary - July 12, 2026
A roundup of the week's biggest AI stories across models, business, and everyday tools.
Weekly AI News Summary

🗣 Voices & Blogs

★ Must-ReadEthan Mollick: the age of the chatbot is giving way to agents you manage
Mollick's argument is that we keep experiencing AI progress as a series of shocks because people are bad at feeling exponentials. The shift now is from a chatbot you prompt to an agent you delegate to and supervise, which changes what your actual job becomes.
📚 LearnNobody agrees on what an agent even is, and that is worth understanding
A clear tour of why the industry cannot settle on a single definition of agent, from simple tool-callers to systems that plan and act on their own. Reading it once makes every vendor's agent pitch much easier to decode.
OpenAI’s first gadget is the $230 Codex Micro macropad
OpenAI’s Codex is about to hit 9 million users, and at least some of those users will soon get a The post OpenAI’s first gadget is the $230 Codex… · The New Stack
GPT-5.6: Frontier intelligence that scales with your ambition
Agentic AI Summit Agenda Now Live and Tickets are Running Out + Livestream Sign-Up | Research Highlight: GPT-5.6's Agents' Last Exam Progress · via Berkeley RDI · Agentic AI Weekly

🌏 The Wire · Drudge / Breitbart

China's companion-AI rules take effect, and its two biggest apps pull the feature rather than comply
The new rules demand anti-addiction limits, usage warnings, and instant-exit for AI built to act human. Doubao and Qwen chose to switch the feature off instead. It is the clearest sign yet that companion AI is now a regulated category, at least in China.
Claude Code's desktop app gets its own built-in browser
A sandboxed browser now lives right beside the chat, terminal, and editor, opened with Cmd+Shift+B. Claude can read and click through docs and sites the way it already drives local dev servers, with a permission prompt the first time it acts on any site.
Google Photos can now restyle your videos with one tap, powered by Gemini
The new Video Remix tool, running on Google's Gemini Omni model, turns a plain clip into a stylized one, watercolor, oil painting, relit, or with the background swapped, in a few taps and no editing software. It is rolling out globally to paid Google AI subscribers.
AI keeps swallowing venture funding, with a record first half of 2026
The latest roundup caps a first half where US venture funding ran roughly 30% ahead of all of last year, with AI taking the lion's share. The big checks are landing on infrastructure, defense, and deployment, not just on models.

🤖 Trending Models · Hugging Face

prism-ml/Ternary-Bonsai-27B-gguf
text-generation · ★374 · 23 dl
prism-ml/Bonsai-27B-gguf
text-generation · ★226 · 513 dl

📈 Markets

NVDA 211.80 ▲4.1%
MSFT 384.93 ▼1.5%
GOOGL 359.51 ▲2.0%
AMZN 247.49 ▲0.1%
META 661.04 ▲0.7%
AMD 548.13 ▲2.6%
AVGO 389.11 ▲1.3%
PLTR 133.72 ▲2.8%
SPCX 136.08 ▼2.2%
TSLA 396.18 ▲0.4%

The BriefChina's first rule for anthropomorphic AI, meaning systems built to act like a human companion with memory, a personality, and ongoing emotional back-and-forth, takes effect today. Rather than rebuild their apps to add the required anti-addiction limits, usage warnings, and instant-exit buttons, ByteDance's Doubao and Alibaba's Qwen are simply turning the humanlike-agent features off, and hundreds of millions of users lose their saved companions and chat history. It matters because it is the first real government line drawn around companion AI, and it splits the map: teams building persistent-memory agents in the US face no such limit yet.

Level UpThis is the week to understand the shift these rules are chasing: from a chatbot you prompt to an agent that remembers you and acts on its own. Read Ethan Mollick's latest on the twilight of the chatbots, then go open your own assistant's memory settings in ChatGPT or Claude and read what it has quietly saved about you. Twenty minutes turns a fuzzy worry into a concrete picture of where this is headed, and lets you decide on purpose whether you want memory on. Read: Ethan Mollick on the twilight of the chatbots