Titus's Morning Wire

AI Titus News

THURSDAY, JULY 9, 2026
OPENAI LAUNCHES GPT-LIVE, A VOICE AI THAT LISTENS WHILE IT TALKS
★ Must-Read / Watch agentic engineering / build-better-agents📚 Learn state of AI coding / useful technique

⚡ AI Frontier · smol.ai

★ Must-ReadClaude Cowork expands to mobile and web
Anthropic's task-agent tool now works from a phone or browser tab, not just the desktop app. Usage data shows most people use it for office admin work like reports and onboarding checklists, not code.
★ Must-ReadAnthropic finds a hidden 'workspace' inside Claude
Researchers found a small set of internal patterns, nicknamed J-Space, where Claude holds an idea in mind before putting it into words. Suppressing it barely hurts simple fact lookups but wrecks multi-step reasoning.
★ Must-ReadxAI opens a no-code builder for phone-answering voice agents
The beta tool bundles phone lines, tools, and call review so anyone can spin up a voice agent in minutes, launched alongside 21 new multilingual Grok voices.
📚 LearnMicrosoft starts swapping OpenAI and Anthropic for its own AI in Excel and Outlook
Tens of thousands of weekly prompts in Excel and Outlook now run on Microsoft's in-house MAI models instead of GPT or Claude, part of a plan to cut what Microsoft pays Anthropic.
Cursor launches a native iOS app for coding agents
Developers can start a cloud agent from their phone, get a notification when it's done, review the diff, and merge the pull request without opening a laptop.
Claude Code and Cowork go into public beta for US government desktops
The FedRAMP High authorized version gives federal teams the same coding and task agents as commercial users, plus audit logs and spend controls for agencies.

📺 Watch · latest videos

★ Must-WatchClaude Cowork: coming to mobile and web
Anthropic's own clip showing a task started on desktop keep running and show up on your phone.
Claude
★ Must-WatchThe next generation of ChatGPT Voice
OpenAI's official demo of GPT-Live, showing the model listening and responding without waiting for you to finish talking.
OpenAI · 79.4K views
OpenAI gave ME early access to the new ChatGPT voice model (GPT-Live-1)
A hands-on first look at GPT-Live-1 outside OpenAI's own demo, testing how it handles interruptions in a real conversation.
FatherPhi
This New OpenAI Voice Model is INSANE!
An independent reaction and test run of the new voice model, checking how natural the turn-taking actually feels.
Julian Goldie SEO
📚 LearnNew Claude Dispatch — Use Cowork From Your Phone!
A walkthrough of Dispatch, the feature that lets you send a task to your desktop Claude session straight from your phone.
Pocketful of AI
★ Must-WatchWe just figured out how AI actually works (J-Space)
Matthew Berman breaks down Anthropic's J-Space finding and what it means for understanding what a model is 'thinking' before it answers.
Matthew Berman · 94.8K views · 4.1K likes
Gemini 3.5 Pro's Entire Base Model SCRAPPED!
A rundown of reporting that Google scrapped its Gemini 3.5 Pro base model and restarted training, pushing the launch to July 17.
Universe of AI
Chinese AI Models Gaining US Tech Customers - USA Losing to China
A plain-spoken take on why cheaper Chinese open models are pulling enterprise token usage away from OpenAI and Anthropic.
Eli the Computer Guy

🗣 Voices & Blogs

★ Must-ReadBetter Models: Worse Tools
Armin Ronacher (Sentry, ex-Flask) shows Anthropic's newest models fight harder against tool schemas that don't match Claude Code's exact shape, and calls it a real regression worth testing for.
📚 LearnUnderstand to participate
Simon Willison reflects on a talk arguing that as coding agents produce bigger changes, you need to understand the code deeply enough to keep steering it, not just approve diffs.
Using Local Coding Agents
Sebastian Raschka walks through running a fully local coding agent with open-weight models, arguing the harness around a model matters as much as the model itself.
Agentic AI in observability: accelerating root cause analysis
The growing development, promise, and use of generative AI and agentic AI continue to drive dramatic change in the IT The post Agentic AI in… · The New Stack
Introducing Claude Sonnet 5
Researching Highlight: SageCTF | Agentic AI Summit Agenda Now Live and Tickets are Running Out! · via Berkeley RDI · Agentic AI Weekly

📦 What's Being Built · GitHub

affaan-m/ECC
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first dev · ★227.7K · JavaScript
NousResearch/hermes-agent
The agent that grows with you · ★212K · Python
Significant-Gravitas/AutoGPT
AutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the too · ★185.4K · Python
ollama/ollama
Get up and running with Kimi-K2.6, GLM-5.1, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models. · ★175.8K · Go
f/prompts.chat
f.k.a. Awesome ChatGPT Prompts. Share, discover, and collect prompts from the community. Free and open source · ★165.1K · HTML
huggingface/transformers
🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, a · ★162.4K · Python

🌏 The Wire · Drudge / Breitbart

Together AI raises $800M to expand open-source AI infrastructure
The Series C values the company at $8.3 billion and will fund a roughly 50x expansion of its compute capacity over five years.
Prime Intellect raises $130M to help companies build their own AI agents
The $1 billion valuation round, led by Radical Ventures, backs tools that let enterprises train agents without depending on OpenAI or Anthropic directly.
ElevenLabs in talks for a deal that would double its value to $22 billion
The voice-cloning startup passed $500 million in yearly revenue in May. This is a share sale letting employees cash out, not new money into the company.
Chinese AI models now handle up to 46% of enterprise AI usage at US companies
Cost is the driver: open Chinese models run 60 to 90 percent cheaper than top US models, and one startup saved millions by switching its traffic to DeepSeek.
US lawmakers open a probe into Chinese AI model use at American companies
The inquiry follows the CNBC token-usage findings and raises questions about data handling at firms that switched providers to save money.
UN's new AI governance commission holds its first meeting, with tech CEOs at the table
The 44-member commission, co-chaired by Salesforce's Marc Benioff and Rwanda's president, seats the heads of Nvidia, Amazon, Microsoft, and Anthropic as formal members, not just observers.
Nearly 9 in 10 companies had an AI-agent security incident in the past year
AvePoint's survey of 750 IT leaders found data leaks and manipulation by bad inputs as the top causes, even as almost half of employees now use agents weekly.
FBI: AI voice-cloning and deepfake scams have cost Americans $893 million
Fewer than 5% of victims are believed to report losses, so researchers think the real damage from cloned voices imitating relatives in distress is much higher.

🤖 Trending Models · Hugging Face

empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF
image-text-to-text · ★1.9K · 1.9M dl
zai-org/GLM-5.2
text-generation · ★3.7K · 362.3K dl
baidu/Unlimited-OCR
image-text-to-text · ★1.9K · 1.2M dl
InternScience/Agents-A1
text-generation · ★415 · 23.1K dl

📈 Markets

NVDA 204.12 ▲3.7%
MSFT 383.34 ▼1.4%
GOOGL 361.92 ▼1.4%
AMZN 243.62 ▼1.0%
META 603.12 ▼2.0%
AMD 517.40 ▲0.3%
AVGO 388.69 ▲4.8%
PLTR 132.22 ▼1.6%
SPCX 148.30 ▼0.8%
TSLA 394.06 ▼2.2%

The BriefOpenAI released GPT-Live, a new voice model built to be 'full duplex,' meaning it can listen and speak at the same time instead of taking turns like a walkie-talkie. That lets you interrupt it mid-sentence and have it react naturally, closer to how a real phone call works. The smaller GPT-Live-1 mini is now the default voice mode for everyone in ChatGPT, while the bigger GPT-Live-1 goes to paid subscribers, and it matters because voice is quietly becoming a primary way people use AI, not just typing.

Level UpThis week's headlines are full of AI-agent security incidents (see The Wire below), so the skill to build now is testing whether your own agents can be tricked. Eugene Yan lays out practical patterns for cybersecurity evals: how to probe an agent for prompt injection, data leaks, and misuse before it ships. If you run any agent that touches customer data or calls tools, spend an hour writing a few of these adversarial tests this week. Patterns for Building Cybersecurity Evals (Eugene Yan)