🧑‍💻 1,000 Claudes Built an OS (Fable 5.2)

Anthropic’s new model takes on week-long projects alone, a robot building in orbit, an ER question that flags PTSD, and more...  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌

November 09, 2026   |   Read online   Hey, Jhon! Welcome back to the #1 AI newsletter in the world! Here’s what we have today: 1,000 Claudes built an OS in eight days. A robot builds in orbit with no astronaut on board. One ER question helps AI flag PTSD risk. A drone that tracks people without GPS.       Claude Fable and Mythos 5.2 Drop Claude’s newest model can take a project too big for one AI and split it across more than 1,000 copies of itself. The copies run for days and check each other’s work as they go. It’s live for everyone now, and Anthropic showed it off with three tests. The operating system. One engineer gave Claude an empty folder and one task, to design the software that runs a computer the way Windows or macOS does. The copies wrote and reviewed code around the clock for eight days. The result started up on a real laptop, with its own core, drivers for the disk and the network, file storage, a login screen and an installer. The gadgets. From one paragraph describing three devices, Claude wrote the specs, picked the parts and laid out the circuit boards. A board shop built them, and all three worked with no electrical fixes. The proof. Another thousand copies wrote out a full formal proof of Fermat’s Last Theorem in 30 hours. The earlier attempt took 11 days, with a person guiding it along the way. Here’s what improves for you: Answers come back up to 30% faster, with an optional fast mode in Claude Code that goes faster still. It wrongly refuses 55% fewer requests than the last version, so fewer “can’t help with that” replies on things it should help with. It uses fewer tokens to finish hard tasks, which means more work per message on your plan. Paid plans get extra Fable usage for a limited time, as resets that refill your usage. How many you get depends on how long you’ve been subscribed: Less than six months: one reset Six to 12 months: two resets More than 12 months: three resets Mythos got the same upgrade. It’s the same model without the extra safeguards Fable carries for security and biology work, and it only goes to vetted organizations through Anthropic’s security and life sciences programs, plus businesses using Claude Security on the Enterprise plan. In testing, it worked out a way to predict the shape of the tiny protein shells that carry gene therapies into the body, and got 28 of the 34 it tried right. Why it matters: Breaking a big project into pieces and feeding them in one session at a time is the part this takes off your plate. You give it the goal, and Claude works out the split. Anthropic hasn’t said how anyone checked eight days of work by a thousand agents beyond the agents testing each other. Our take: Try it on the project you’ve been feeding Claude a session at a time. Give it the full goal, ask it to plan the split before it starts, and read the plan before you let it run. A bad split shows up in the plan long before it shows up in the code. Source: Anthropic       Together with Attio AI Agents That Only Access What They Should Before an AI agent can work through your customer list, you have to give it access to something. An agent set up for one rep shouldn’t be editing accounts that rep can’t touch, or changing a deal they were never on. Attio’s Custom Agents each run as a person you name, with exactly that person’s permissions and nothing more. You describe the job once, like spotting every deal that’s gone quiet and working out why, and the agent runs it on every record around the clock. The agent log shows what it read and what it decided, and you can pause or replay any run. When you ask Attio in chat to update a stalled deal or send a follow-up, it shows you exactly what it’s about to do and waits for your go-ahead.       Robotics Robot Builds the First Structure in Orbit Anything sent to space has to fold up small enough to fit inside a rocket’s nose. That’s what limits how big a space telescope or antenna can get. A robot from Caltech is working on a way around it. It’s been in orbit since spring, building a round antenna 4.6 feet across with no astronaut anywhere near it. It’s the first time a robot has built part of a structure in space on its own. The robot went up in a box packed with all its parts. Little motorized dispensers hand it thin fiber tubes and metal joints that it fits into one section of frame. Then it stretches a thin, shiny film across the frame to bounce back radio signals before pushing the section out to start the next one. The finished antenna needs 11 sections. Two went up already built, with the robot adding a third in orbit. Engineers can only reach it seven or eight times a day, for 15 minutes at a time. When something goes wrong, they recreate the problem on an identical robot back at Caltech and send up the fix once it works there. Our take: The robot won’t finish the full antenna before its mission ends. It did show that a robot can handle the assembly step in space. Next comes a simpler version built on what this one learned, then structures too big for any rocket to carry. Source: Caltech       Pitch In Focus AI Decks That Look Like Yours AI can draft a deck in seconds. What comes back uses some other company’s fonts and colors with your logo nowhere on it. You end up fixing every slide by hand. Perplexity, Legora, Synthesia, and Poolside make their decks at scale with Pitch. Pitch Agent builds fully editable decks from your own templates, with your fonts and colors on every slide. Hand it the brief and the slides come back on brand, ready to edit like any deck you’d build yourself. Connect Pitch to Claude and it can pull your call notes straight into the slides.       Mental Health One ER Question Helps AI Flag PTSD “What brought you to the ED (Emergency Department)?” It’s one of the first questions a nurse asks someone when they walk into the ER after an injury. More than a quarter of those patients develop PTSD within a month. Fewer than a third of ERs screen for it. That leaves the rest walking out without anyone asking. NYU Langone built an AI that reads the answer to that one question for signs of PTSD risk. They recorded 145 patients at two New York trauma centers within a day of their injury. Some answers ran a few sentences. Others went on for pages. Three layers of open AI models go through each story. The first looks for warning signs. It also weighs what protects someone, like people to lean on at home. The last layer adds basic details like age and the kind of injury before giving a risk score with a short explanation of why. A month later, a clinician tested each patient for PTSD with the standard interview. The AI had flagged 95 of every 100 people who developed it. On 30 patients picked at random, it caught every single case. A clinician with 24 years in emergency medicine and PTSD research caught 89% of them. The code is public for other hospitals to test on their own patients. It only works in English for now and still needs testing in larger groups before any ER uses it. Our take: Fewer than half of the people it flagged went on to develop PTSD, which can sound like a lot of false alarms. In an ER that’s the right trade. A flag sends someone to a short follow-up, while a missed case can turn into years of symptoms that could’ve been caught early. Source: Nature Communications       Coming in Hot AI Tools of the Day 🧭 Yahini works inside Claude or ChatGPT, where you ask for keyword research and briefs that pull from your Search Console. 🧩 NodeTerm lays out your Claude and Codex sessions on one canvas, flags the agent waiting on you and brings them back after a reboot. 🖥️ Hostinger VPS now links anything you run to every major AI model through one key, no separate provider accounts needed. 🍳 fridgesnap.AI turns a photo of your fridge into recipes using the food you’d otherwise toss, matched to your diet. 🛋️ RoomLab.app swaps only the sofa or the floor in a snapshot of your room and keeps the rest as it is.       Notable AIs Notable AI Tools 📊 SE Visible tells you where your brand stands across AI engines and markets, what shapes the answers, and where to get ahead. Try it for free.       Interesting AI This Drone Tracks People Without GPS Purdue has a drone that can lock onto a person running through thick forest and follow them on its own, with no GPS signal or map of the trees. The runner weaves between trunks. The drone stays with them, dodging branches at speed while keeping them in the middle of its camera’s view. When a tree blocks its view, the drone predicts where the runner will come out on the other side and holds its course. It picks them back up the moment they reappear. A laser scanner on board builds a 3D map of the trees as it flies. All of the tracking and steering runs on the drone’s own chip, with nothing waiting on a server in the split second it takes to swerve around a thin branch. The lab designed it themselves, down to a 3D-printed carbon fiber frame. It’s a research demo for now, with the paper still under review. Source: Purdue IDEAS Lab       Free Tool of the Day Get a Bird's-Eye View of Any Photo A shot from straight overhead makes a dog or a classic car look like a magazine cover. Getting that angle for real calls for a high balcony or a tall ladder. Aerial View turns a regular photo into that shot. Upload a picture of your subject and it places them on a wide green lawn, seen from straight above in soft daylight with natural shadows. Your dog still looks like your dog. It’s 100% free to use, commercial use included.       Open Source Finds From the Source 🧑‍💼 Orkas is a free desktop app for Mac and Windows where you hand one goal to a lead agent, which splits it into tasks for 16 specialist agents and gets back a report, a deck or a video. 🐦 Fugleramme listens for birds outside your window, names each one by its song and shows it as an old illustrated print on a small e-ink frame, a TV or any browser.       Prompt of the Day The Project Splitter Important Note: You’ll need to click the button to get the complete prompt. View and copy today’s full prompt This prompt takes a goal too big for one chat and breaks it into separate work packages, each written as a brief you paste into its own session. It maps what has to happen first, what runs side by side, and where two pieces will collide. The finished plan includes a dependency map with ready-to-paste briefs, plus a final check for stitching the results together.       That’s all for Monday! Before you go, set up one Attio agent that works with a single teammate’s access. Hand it the deal check you never get to and let it run while you’re off. How did we do today? Rate this issue, then tell us what you thought. We value your input and will use it to make TAAFT better for you. ⭐⭐⭐⭐⭐ Absolutely Loved It ⭐⭐⭐⭐ Better Than Usual ⭐⭐⭐ Meh, It Was Okay ⭐⭐ Could Be Better Thanks for reading, — There’s An AI For That     Reach out to over 1.8 million AI enthusiasts today. Become a sponsor Update your email preferences or unsubscribe here © 2026 There's An AI For That® SRL 228 Park Ave S, #29976, New York, New York 10003, United States