Microsoft Rebuilds Copilot As An OS For Work With Agents That Keep Working When You Log Off

Microsoft Rebuilds Copilot As An OS For Work With Agents That Keep Working When You Log Off

0:00 / 3:15

Chapters

DAILY ROUNDUP

Microsoft Rebuilds Copilot As An OS For Work With Agents That Keep Working When You Log Off

calendar_today Date:
schedule Duration: 3:15
visibility 2 Views

Microsoft rebuilds Copilot as an OS for work with Home, Code and Autopilot, Claude computes a record nine-loop physics amplitude, AI maxes out the Mensa Norway IQ test, and Tesla Semi enters volume production.

  • 01. Copilot Home combines Chat and Cowork with Word, Excel and PowerPoint built in; Code builds apps from plain English on GitHub Copilot tech inside the tenant; Autopilot is a persistent, proactive agent
  • 02. Fable 5.1 ran largely unsupervised for days in Claude Science to compute the nine-loop hexagon amplitude in planar N=4 SYM, beating the eight-loop record; Lance Dixon verified it; cost $1,000-2,000
  • 03. TrackingAI's Mensa Norway chart (average of last 7 tests) puts Claude Fable 5.1 and GPT-6 Astra Ultra at 151, the test's ceiling; human scores are reported 85-145
  • 04. Tesla Semi high-volume production in Sparks, Nevada: 1.8M sq ft, 50,000 trucks a year, Standard Range 325 miles ~$260K, Long Range 500 miles ~$290K
Today's AI news: Microsoft rebuilt Copilot as an "OS for work" with Home (Chat plus Cowork, with Word, Excel and PowerPoint inside), Code for building apps in plain English, and Autopilot, a persistent agent that keeps working when you don't. Anthropic's Fable 5.1 computed the nine-loop six-particle amplitude in planar N=4 super-Yang-Mills for $1,000 to $2,000, verified by SLAC's Lance Dixon. Claude Fable 5.1 and GPT-6 Astra Ultra hit 151 on TrackingAI's Mensa Norway IQ test, the maximum possible score. And Tesla Semi entered high-volume production at a 50,000-truck-a-year factory in Nevada. Chapters: 0:00 Today's AI News 0:26 Copilot OS For Work 1:11 Claude Nine Loops 1:55 AI Maxes Mensa IQ Test 2:34 Tesla Semi Production Blend Roundup 2026-09-25 https://x.com/satyanadella/status/2103455884366188544 https://x.com/wallstengine/status/2103457685022867494 https://x.com/AnthropicAI/status/2103541577083719888 https://x.com/AISafetyMemes/status/2103369359481852116 https://x.com/tesla_semi/status/2103289460767203424 https://x.com/tesla_semi/status/2103288140991746253 Microsoft has torn Copilot down and rebuilt it as what Satya Nadella calls an operating system for work, with agents that keep going after you log off. Claude has pushed a famous particle physics calculation to nine loops for a couple of thousand dollars, AI has hit the ceiling of the Mensa Norway IQ test, and Tesla's Semi is finally in volume production. Here's today's AI news. Microsoft has announced its biggest Copilot update yet, and Satya Nadella is calling it a new operating system for work. It's built around four pieces. Home brings Chat and Cowork together: Chat for quick questions, Cowork for bigger jobs you hand over completely, like an RFP response or a month-end close. Word, Excel and PowerPoint are now right inside it. Code lets anyone describe an app, tracker or dashboard in plain English and have Copilot build it, running on GitHub Copilot's technology inside your company's own tenant. Autopilot is a persistent agent you give a role and objectives, and it keeps working across Teams, Outlook and your documents when you're not there. This one will all depend on how seamless the implementation is. Anthropic's Science Blog has a remarkable one. Physicists predict particle collisions using formulas called scattering amplitudes, calculated in layers called loops. Each loop makes the answer more precise and costs exponentially more computing, and the record in the standard test model was eight loops, set by SLAC's Lance Dixon and colleagues. Last month physicist Matt von Hippel challenged AI labs to go further. From a single prompt, Fable 5.1 ran largely unsupervised for days in Claude Science and computed the nine-loop result for one to two thousand dollars. Dixon independently verified it. It used known methods rather than new physics, but it finished a fragile calculation nobody had completed before. AI has now maxed out the Mensa Norway IQ test. The latest results on TrackingAI, averaged over each model's last seven attempts, put Claude Fable 5.1 and OpenAI's GPT-6 Astra Ultra at 151, the highest score the test can give. That's effectively a perfect paper. For context, the human version of the test only reports scores between 85 and 145. TrackingAI also runs an offline test written by a Mensa member and kept off the internet, to check models aren't just recalling answers from training data. Either way, this test has run out of headroom, and it's going to need a harder one. Tesla Semi is finally in high-volume production, seven years after the 2019 date Tesla originally promised. The dedicated factory in Sparks, Nevada covers 1.8 million square feet and is built for 50,000 trucks a year. The Standard Range truck does 325 miles fully loaded for around $260,000, and the Long Range does 500 miles for about $290,000. Megachargers put back 60 percent of range in 30 minutes. PepsiCo, DHL and US Foods are among the first customers, alongside a 2,500 truck order from the ZET alliance. The new electric steering lays the groundwork for autonomy, but there's no self-driving on these trucks yet.
Stories In This Briefing

5 stories covered

VIEW ALL NEWS
Related Stories

More on these topics