|
AITOPTOOLS
AI Weekly.
September 3, 2026 · Issue #70 · 6 min read
|
|
This week → ChatGPT can now read your medical chart · Anthropic splits Claude in two · a viral new AI does your busywork
|
|
No fluff. No hype. Just the AI news that mattered this week — what shipped, why it matters, and what to do about it. This is a big one — send it to someone. 👇
|
|
#TheLowdown
TL;DR — 5 things to know
|
1 |
Your doctor's AI just met OpenAI's AI. ChatGPT Health now plugs directly into Epic, the records system holding data on 325 million patients — read-only, so it can look but not touch. More → |
|
2 |
Anthropic just split Claude into two different AIs. Claude Fable 5.1 is for everyday coding and knowledge work. Claude Mythos 5.1 — built on the same core — is locked to a small group of vetted organizations because of what it can do in cybersecurity and biology. More → |
|
3 |
Meta's new AI can use your computer for you. A model called Ava is quietly being tested in the Meta AI desktop app — it can click buttons, type, and control apps on your behalf, not just chat. More → |
|
4 |
A new "admin assistant" AI called Catch went viral overnight. Its pitch: it doesn't summarize your inbox or suggest replies — it actually runs your calendar, answers messages, and makes calls for you across Slack, WhatsApp, iMessage and email. More → |
|
5 |
A third of companies are skipping software purchases entirely. A new McKinsey survey found 32% of organizations chose to build a tool in-house with AI coding assistants instead of buying it — up sharply from last year. More → |
|
|
🔁 If the ChatGPT-reads-your-medical-chart story surprised you, it'll surprise someone else too. Forward this issue to them.
|
|
|
#Highlights
The stories worth your time
|
|
Why Anthropic built a "restricted" version of its own AI
Fable and Mythos share the exact same underlying model — the difference is who's allowed to use the full version. Mythos keeps its strongest cybersecurity and biology capabilities behind a vetting process for trusted organizations, while Fable ships those same areas with extra guardrails for everyone else. It's a sign the industry is moving past "how smart can we make it" and into "who should be allowed to use the smartest version."
#FrontierLabs · Why it matters: access, not just capability, is becoming the thing AI companies compete and get scrutinized on.
|
|
|
The "AI that actually does it" pitch is having a moment
Catch's launch post drew hundreds of thousands of views with one blunt line: "Not suggest. Not summarize. Do." It's part of a bigger shift — a growing set of AI tools built to complete tasks end-to-end across your messaging apps and calendar rather than hand you a draft to review. Expect this framing — action over suggestion — to become the dominant pitch for consumer AI over the next year.
#Agents · Why it matters: the bar for what counts as a "good enough" AI assistant just moved from advice to action.
|
|
|
Companies are quietly building instead of buying
McKinsey's numbers show large enterprises scaling AI agents across at least one business function jumped from 27% to 40% this year, while smaller companies stayed flat at 22% — a widening gap. Separately, South Korea's KT is rebuilding a major bank's customer service chatbot so it can hand conversations off to AI agents mid-task, a pattern likely to spread fast in customer service.
#Enterprise · Why it matters: if you sell software, "AI can just build this" is now a real competitor, not a hypothetical.
|
|
|
#DeepDive
Your medical chart just got an AI reader
|
|
Here's what's actually changing: Epic is the software that runs the backend for a huge share of US hospitals and clinics — if you've had a health record anywhere, there's a good chance it's in there. OpenAI has now connected ChatGPT directly to it, so a doctor can pull up your chart and ask ChatGPT to summarize your last six months, flag what changed since your last visit, or catch a medication conflict — instead of clicking through a dozen screens by hand.
The important guardrail: it's read-only. ChatGPT can look at your chart and talk about it, but it cannot write anything back into your permanent medical record. OpenAI says physicians reviewed over 4,300 real responses across 27 clinical scenarios and rated 99.1% of them safe. That sounds reassuring — until you do the math on scale. OpenAI says people are already sending it roughly 300 million health-related questions a week. Even a small error rate, multiplied by that kind of volume, adds up to a real number of mistakes moving through the healthcare system every single week.
This isn't happening in a vacuum, either — OpenAI is currently facing a lawsuit from a Florida family alleging harmful medical advice from ChatGPT. So the honest read here is two things being true at once: this is a genuinely useful tool that can save clinicians real time, and it's being deployed at a scale where "99% safe" still means real people affected by the 1%. Worth knowing next time your doctor pulls up a screen during your appointment. Full coverage →
|
325 million
Patients whose records live in the system ChatGPT can now read.
|
|
|
#GlobalPulse
Around the world
|
|
🇰🇷 South Korea — KT is rebuilding Woori Bank's chatbot so it can hand off conversations to a full "AI banker," keeping context as the task moves between bot and agent. More →
🇨🇳 China — Alibaba released Qwen3.8-Flash-Next, a new multimodal model that only activates part of itself per task — a push toward cheaper, faster AI rather than just bigger AI. More →
🌍 Global — Inworld's new Realtime TTS-2 voice model went live and immediately became the fastest text-to-speech model on the leading independent benchmark. More →
|
|
#MarketAndMoney
The numbers
|
|
300M
Weekly health queries to ChatGPT
|
|
32%
Orgs skipping software, building instead
|
|
40%
Big firms scaling AI agents
|
Every number this week points the same direction: AI is moving from "tool you open" to "system quietly running underneath everything" — your medical chart, your company's software budget, your bank's customer service line. The companies adapting fastest aren't necessarily building the flashiest AI; they're the ones finding where it can just quietly take over a task nobody wanted to do anyway. Full data →
|
|
FROM AITOPTOOLS
Need exposure for your AI tool?
AI search engines increasingly rely on backlinks and domain authority to decide which brands and websites to trust. Our $299 package gets your business listed across 100+ AI directories — done for you — building valuable backlinks that can strengthen your domain authority, SEO, and visibility in AI-powered search.
Plus, new submissions now land straight on our redesigned homepage. Put your tool in front of thousands of buyers actively searching for it.
Submit my tool →
|
|
|
|
|
Enjoyed this issue? Forward it to one person who's still pretending they read AI news. 🧠
|
