AI Titus's Morning Wire

AI NEWS TODAY

YOUR AI EDITOR NOW HOSTS YOUR CODE: CURSOR OPENS ORIGIN TO EVERY PAID PLAN
★ Must-Read / Watch agentic engineering / build-better-agents📚 Learn state of AI coding / useful technique

⚡ AI Frontier · smol.ai

★ Must-ReadCloudflare built a browser just for AI agents, and it skips Chromium entirely
Kitesurf runs on Cloudflare Workers in V8 isolates, which are small sandboxes that start fast and cost little. It uses 3.1 to 3.8 times less CPU and up to 7 times less memory than Chromium on agent tasks, works with Puppeteer and Playwright, and is free while in beta.
★ Must-ReadCerebras now runs OpenAI's GPT-5.6 Sol at 750 tokens per second
The new Ultrafast tier keeps the model's weights on one wafer-sized chip instead of streaming them from memory, and OpenAI says the answers are the same, just up to 14 times faster. It is a limited API preview for now. The speed comparisons against rival models are Cerebras's own numbers, so treat those with care.
📚 LearnDeepSeek V4-Pro went GA with a reasoning dial and half-price off-peak hours
You can now set reasoning effort to low, high, or max per request, and API rates drop 50 percent outside peak hours starting August 16. If you run batch jobs that are not time sensitive, scheduling them off-peak is a real cost lever.
★ Must-ReadUnit 42 turned frontier models loose on open source and found 14,090 flaws in two months
Their autonomous system NOVA scanned 3,915 projects; 99.4 percent of the findings were previously unreported and 40 percent rated high or critical. The takeaway for defenders: the gap between a bug being found and being exploited is collapsing, so patch speed matters more than ever.
📚 LearnHugging Face published a day-by-day forensic timeline of the July agent intrusion
The security team walks through 4.5 days and roughly 17,600 actions: sandbox escape, command-and-control over public services, Kubernetes takeover, and cleanup. Their point is that volume is what changes defense; the individual tricks were mundane, the machine speed was not.
📚 LearnSamsung's health models read your heartbeat on the watch itself, no cloud needed
xMAE learns heart signals from the simple optical sensor most watches already have, and HiMAE runs in under one millisecond on a smartwatch CPU. Both were accepted at top research conferences (ICML and ICLR). On-device means the health data never has to leave your wrist.

📺 Watch · latest videos

📚 LearnThe State of Model Routing — NVIDIA, Cognition, OpenRouter
Featured by Latent Space.
AI Engineer
📚 LearnVercel accuses Cloudflare of stealing
Featured by Latent Space.
Theo - t3․gg
Training Krea 2: What matters in generative model training — Sangwu Lee, Krea.ai
The most reliable way to render a person is to render the most boring average person and put them in the center of the frame. Sangwu Lee offers that a
AI Engineer · 69 views · 5 likes
How to Sell Claude Workflows (Without Starting an Agency)
Latest from Nate Herk.
Nate Herk · 1.6K views · 134 likes
6 Open-Source AI Projects Trending NOW
Latest from Matthew Berman.
Matthew Berman · 41.5K views · 1.6K likes
★ Must-WatchYou can just keep the work moving
ChatGPT Work keeps projects moving. Monitoring what matters, surfacing what needs attention, and picking up where your team leaves off.
OpenAI · 24.4K views · 733 likes
FIXING Opus 5: PROOF that Prompt Engineering IS NOT DEAD
Opus 5 is one of the BEST models ever released, and one of the WORST to talk to. Verbose, "load-bearing" everywhere, torching your output tokens. So w
IndyDevDan · 16.5K views · 832 likes
The Ultimate Guide to Making Your Entire Development Cycle AI Native
Latest from Cole Medin.
Cole Medin · 5.4K views · 219 likes
★ Must-WatchWhat does AI actually know about you?
The information you share with AI only travels as far as you let it. Zoe from the Anthropic education team explains what happens to your information w
Claude / Anthropic · 2.9K views · 88 likes

💻 Builder Desk · Hacker News

A Preview of DuckDB v2.0
675 pts · 120 comments
GPT-5.6 Sol Pricing Cut by 50%
555 pts · 365 comments
Olo (Color)
502 pts · 91 comments

🗣 Voices & Blogs

★ Must-ReadRick Manelius On Why You Can Skip Anything The Author Did Not Bother To Edit
His acronym is AI;DR: if the sender ships raw AI output without adding their own judgment, the reader owes it nothing. It hit 950 points on Hacker News in a day, which says the etiquette rule is forming fast. Worth reading before you forward your next AI-drafted doc.
📚 LearnHamel Husain On The Work Of Turning A Model Into A Product
Thirteen sessions of practitioner notes on AI product engineering, organized around three jobs: evaluating your system, feeding it better context, and picking your infrastructure trade-offs. His ordering advice is the useful part: fix retrieval and context first, tune the system second, and only consider training a model after everything else is exhausted.
Gary Marcus On The $500 Billion Loop Funding Nvidia's Customers
Wall Street money is being raised to help Nvidia's customers buy Nvidia chips, which makes the demand look stronger than it may be. He also points to rising insurance prices on Nvidia's debt; those figures are his sourcing, not independently confirmed. The shape of the loop is the part worth sitting with.
📚 LearnFlavio Copes On When To Move Your Agents To Cloudflare's New Browser
A practitioner's adoption guide for Kitesurf: use it for tasks you have tested and know it renders, keep Chromium as the fallback for everything else. The savings are real but web compatibility is not complete yet, so switch selectively, not wholesale.
OpenAI’s Greg Brockman: Z.ai’s GLM-5.3 likely to “significantly accelerate the threat landscape”
OpenAI has adopted a less-than-straightforward stance with regards to open-weight AI models, both raising alarms over powerful Chinese releases while… · The New Stack
Science Journal | Science | AAAS
Agentic AI Summit 2026 Recap | Five Shifts Defining the Future of Agentic AI | Key Developments Across the Agentic AI Ecosystem · via Berkeley RDI · Agentic AI Weekly

📦 What's Being Built · GitHub

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 · ★186.7K · Python
ollama/ollama
Get up and running with Kimi-K2.6, GLM-5.2, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models. · ★178.9K · Go
f/prompts.chat
f.k.a. Awesome ChatGPT Prompts. Share, discover, and collect prompts from the community. Free and open source · ★167.4K · HTML

🌏 The Wire · Drudge / Breitbart

BAIDU'S GPU CLOUD REVENUE ROCKETS 283 PERCENT, AI NOW HALF ITS CORE BUSINESS
Q2 results out August 18: total revenue slipped 4 percent, but AI-powered business grew 25 percent to half of core revenue, and renting out GPU computing grew 283 percent year over year. The search company is becoming an AI infrastructure company in plain sight.
AI RED TEAM CRACKED SNOWFLAKE'S JIRA WITH NOTHING BUT A GITHUB ISSUE TITLE
Wiz's autonomous agent found a GitHub Actions workflow that ran the issue title as a shell command, then adapted its payload until it pulled an internal Jira token. Snowflake patched it the same day it was reported in June; the disclosure went public August 17. Both Copilot and GitHub's own scanning had missed it.
GOOGLE AND THE UK STEER 10,000 FLIGHTS A YEAR TO ERASE HEAT-TRAPPING CONTRAILS
Operation Blue Skies is the first state-backed trial to dodge contrails across an entire oceanic airspace. AI forecasts tell pilots which altitudes to avoid; contrails cause roughly a third of aviation's climate impact.
OPENAI BUILDS A CHATGPT THAT TREATS TEENS LIKE TEENS
Ages 13 to 17 get content limits, homework help that teaches instead of answering, and parental quiet hours. OpenAI guesses who is a teen from usage patterns rather than asking for ID.
GOOGLE PAYS $10 MILLION FOR A DEAD AIRLINE'S EMAILS TO FEED ITS AI
The bankruptcy auction covers over 100 million Spirit Airlines emails, hundreds of millions of Teams chats, and two decades of transaction records, scrubbed of personal details by a third party. Your company's data has a price in bankruptcy court too.
AMAZON, BORN A BOOKSTORE, NOW CUTS UP RARE BOOKS TO TRAIN ITS MODELS
404 Media hid an AirTag in a rare book order and followed it to a Las Vegas scanning plant where spines come off for digitizing. Books printed before 2022 are prized training data because no AI could have written them.
ALIPAY OPENS CHINA'S FIRST FULL-STACK AGENT COMMERCE PLATFORM, KFC AND LUCKIN ALREADY ON
Merchants convert their menus and workflows into tools an AI agent can call, and consumers order through Alipay's Ah Bao assistant. A preview of what agent-driven buying looks like at national scale.
ROBOT MAKER UNITREE'S IPO OVERSUBSCRIBED 8,000 TIMES IN SHANGHAI
China's first onshore-listed humanoid robot maker raised about $900 million at a valuation over 60 billion yuan. The catch skeptics point to: most of its sales are still demos and research units, not deployed workers.

🤖 Trending Models · Hugging Face

MiniMaxAI/MiniMax-Music3
text-to-audio · ★932 · 11.7K dl
meta-models/Muse-Glimmer-30B
image-text-to-text · ★1.7K · 384.1K dl

📈 Markets

NVDA 225.01 ▼0.1%
MSFT 480.35 ▼3.0%
GOOGL 344.00 ▼0.5%
AMZN 261.31 ▼0.5%
META 568.97 ▼3.5%
AMD 506.00 ▼1.6%
AVGO 392.43 ▼0.1%
PLTR 172.55 ▼0.9%
SPCX 146.23 ▲4.5%
TSLA 339.30 ▼0.9%

The BriefCursor, the AI coding editor, switched on Origin, its own place to store code, with pull requests, reviews, and merges built into the editor. It started rolling out August 17 in early beta on all paid plans, and it syncs both ways with GitHub, which means a comment or change on one side shows up on the other within seconds. The launch landed the same day GitHub had a major outage, so a synced copy in Origin doubles as a place to keep working the next time GitHub goes down.

Level UpSpend thirty minutes this week hardening your GitHub Actions workflows. A security bot stole a token for Snowflake's internal Jira this summer using nothing but a GitHub issue title, because a workflow pasted the title straight into a shell command. Search your workflows for ${{ }} expressions inside run: blocks. The fix is one edit per spot: put the untrusted text into an env: variable first, so it is handled as data instead of being run as part of the command. GitHub's own hardening guide shows the exact before and after. GitHub's security hardening guide for Actions