Get the latest AI news, understand why it matters, and learn how to apply it in your work — all in just 5 minutes a day. Join over 2,000,000+ subscribers.

Thank you! Your submission has been received!

Oops! Something went wrong while submitting the form.

Microsoft brings GPT-5 to Copilot

Rowan Cheung • 6 minutes

^{Read Online}^|^{Sign Up}^|^Advertise

Good morning, AI enthusiasts. GPT-5 has landed — and Microsoft has wired it straight into its entire Copilot ecosystem, from chat and coding to dev tools.

Whether you're debugging software or triaging your inbox, this upgrade shifts Copilot from helpful sidekick to the backbone of your workflows. Here’s our take on how it works, where it excels, and what it means for how you get things done.

In today’s AI rundown:

Microsoft’s Copilot gets GPT-5 overhaul
Microsoft 365 Copilot’s context-aware upgrade
GitHub and VS Codetdelivers even smarter dev experience
Azure AI Foundry powers seamless GPT-5
Copilot gets GPT-5 smarts too
Our top use cases, how to use, and more

MICROSOFT UPDATES

THE ANNOUNCEMENT

🧠 Microsoft’s Copilot gets GPT-5 overhaul

Image source: Kiki Wu / The Rundown

The Rundown: ICYMI, Microsoft has rolled out GPT-5 across its entire AI ecosystem, bringing the long-awaited OpenAI update to users of Microsoft 365, the standalone Microsoft Copilot app, GitHub Copilot, and Azure AI Foundry.

The details:

The new “Smart mode” in Copilot shifts between quick-fire suggestions and in-depth analysis without needing user toggles.
GPT-5 brings expanded context window to understand not just prompts but entire workflows, including email and document history and task intent.
It can even detect tone shifts, embedded meaning, and cross-reference related content across apps, enabling more relevant, human-like responses.
Unlike earlier AI tiers, GPT-5 is available across all Copilot tiers, including non-enterprise users, giving everyone access to its advanced AI functionality.

Why it matters: GPT-5 is now available to all Copilot users (including non-subscribers) and is easy to access via Windows, Mac, web, or mobile. Using the model within Copilot differentiates itself from ChatGPT by being able to understand context, take action, and fit into real workflows, moving from passive tool to active collaborator.

MICROSOFT 365 COPILOT

🔥 Microsoft 365 Copilot’s context-aware upgrade

Image source: Kiki Wu / The Rundown

The Rundown: Microsoft 365 Copilot with GPT-5 brings enterprise AI that combines web intelligence with your organization’s data — delivering precise summaries, analyzing emails, docs, spreadsheets, and meetings across Microsoft Graph for context-aware insights that ChatGPT can’t easily access.

The details:

Microsoft 365 Copilot reasons across Graph, including emails, SharePoint docs, Teams meetings, and OneDrive files — for responses grounded in actual work context.
The update also removes complexity for users, eliminating the need to understand model hierarchies or switch modes.
Security, privacy, and compliance controls remain unchanged, assuring safe integration into enterprise environments.
Note: Microsoft 365 Copilot is the business version, distinct from free consumer Copilot. Separately, GitHub Copilot offers a range of options for developers across free and premium plans.

Why it matters: Microsoft 365 Copilot’s ability to access your org’s entire knowledge base through Microsoft Graph now allows for invaluable contextual insights from your entire workflow. Plus, with an automatic model router adapting to task complexity on the fly, enterprise AI becomes accessible to everyone — not just advanced power users.

GITHUB COPILOT

⚡️ GitHub and Visual Studio deliver even smarter dev experience

Image source: Kiki Wu / The Rundown

The Rundown: GitHub Copilot, newly upgraded with GPT-5, is reinventing what developers expect from AI-powered coding with smarter suggestions, agentic workflows, and rewindable chat sessions.

The details:

GPT-5 superpowers GitHub Copilot’s code suggestion engine, delivering context-aware completions, instant refactoring, and ultra-precise fixes directly.
The new “chat checkpoints” feature in Visual Studio lets developers restore workspace states and Copilot chat history to any moment for smooth, safe, reliable coding.
Dedicated coding agent sessions allow users to manage long-running workflows, track multi-step plans, and interact with agents across projects.
GPT-5’s multimodal upgrades let Copilot absorb not only code but also documentation, design assets, and user stories.
GPT-5 is now available for all paid users of GitHub Copilot, which supports multiple AI models with different capabilities.

Why it matters: This upgrade turns Copilot into an agentic coding partner that learns from context and workflow, dramatically cutting friction in building, debugging, and collaborating. Developers now have an AI that understands project nuance, supports deep audits, and powers agent-based coding.

AZURE AI FOUNDRY

☄️ Azure AI Foundry powers seamless GPT-5

Image source: Kiki Wu / The Rundown

The Rundown: Azure AI Foundry just made every GPT-5 model instantly available through Azure AI Foundry Models, allowing developers to plug advanced AI into their apps without wrestling with infrastructure.

The details:

Devs get instant access to four distinct GPT-5 models (standard, mini, nano, and chat) in Foundry Models, each tailored for different use cases.
GPT-5 from Azure OpenAI handles massive context windows (up to 272K tokens), supports multimodal data inputs, and delivers context-rich results.
Microsoft’s real-time model router, released earlier this year, picks the best Azure OpenAI model for the job and for each prompt — faster, smarter, and cheaper — so teams can focus on building, not backend grunt work.
GPT-OSS models from Azure OpenAI are also available, enabling instant integration into Copilot-powered workflows and custom apps. OpenAI’s OSS models are also available on Microsoft Windows.

Why it matters: With Microsoft’s automatic model router, Azure AI Foundry can pick the right model for each prompt — allowing developers to focus on shipping code instead of optimizing costs. Combined with robust safety, instant scalability, and strong context windows, app-building workflows just got a powerful upgrade.

MICROSOFT COPILOT APP

🚀 Copilot gets GPT-5 smarts too

Image source: Kiki Wu / The Rundown

The Rundown: Microsoft Copilot now comes with Smart mode, powered by GPT-5, putting the most advanced next-gen AI in everyone’s pocket — letting users generate content, organize tasks, and ideate on the go using natural language or voice commands.

The details:

Access GPT-5 by selecting "Smart mode” from the dropdown menu — no Microsoft account required for basic use.
Copilot tightly links with Bing, Edge, and Microsoft’s voice AI, enabling voice commands, browser-based generative features, and seamless access.
Users can use Copilot Chat for personal brainstorming, travel planning, or everyday Q&A.
The app’s Copilot Pages feature allows users to generate, ground, and edit content such as stories, infographics, and code in real time.

Why it matters: Consumer users can access Copilot across Windows, Mac, web, and mobile, enjoying integrated AI features across familiar workflows in Word, PowerPoint, Outlook, and more. Free users also get 5x more GPT-5 reasoning queries daily than ChatGPT’s free tier — giving broader access to advanced AI anywhere it’s needed.

REAL-WORLD TESTS

REAL-WORLD TESTS

🧪 How is GPT-5 in Copilot 365 different from ChatGPT?

Image source: Kiki Wu / The Rundown

The Rundown: To see how GPT-5 inside Microsoft’s Copilot ecosystem truly differs from standalone ChatGPT 5, we tested its reasoning capabilities across the Microsoft 365 product suite (Outlook, Teams, OneDrive, and more).

The use cases that stood out:

Taming the email pile-up (Outlook): “Summarize all emails from the past week into a single update, with bullet points on what I need to reply to today.”
Meeting you missed (Teams): “Summarize yesterday’s marketing meeting I couldn’t attend, highlight decisions made, and share action items for me.”
Find key insights in Excel files (OneDrive): “Find the Excel file in OneDrive with The Rundown's Forecast for 2025. Tell me what's changed since last year.”
Prep before a big call or workshop (Teams + Outlook): “I have to present a workshop on GPT-5 in Microsoft Copilot. Prep me based on all recent emails, proposals, meeting notes, and public information.”

In summary: GPT-5 in Copilot isn’t about a shiny new model or AGI, it’s about enabling reasoning across tools people use in work every day. In true Microsoft fashion, the focus is on real-world results, echoing what Satya Nadella told The Rundown: “we celebrate tech companies far too much versus the impact of technology.”

Tech

Peloton stages AI comeback

Jennifer Mossalgue • 5 minutes

^{Read Online}^|^{Sign Up}^|^Advertise

Good morning, tech enthusiasts. Peloton, the pandemic’s breakout fitness darling, is betting big on an AI-fueled reboot. New CEO Peter Stern is pushing personalized wellness with smarter bikes and treadmills to reverse sagging sales.

But with rivals like Meta charging into immersive workouts, can Peloton muscle its way back to the top, or was its golden era a lockdown fluke?

In today’s tech rundown:

Peloton gets an AI overhaul
Apple revives blood oxygen tracking
HTC debuts Meta-like smart glasses
Amazon launches fresh same-day delivery
Quick hits on other major tech news

LATEST DEVELOPMENTS

PELOTON

🚴🏻‍♂️ Peloton gets an AI overhaul

Image source: Peloton

The Rundown: In a make-or-break move, Peloton is betting on AI, rolling out a new line of smart at-home bikes and treadmills to revive slumping sales and claw back its crown in connected fitness, Bloomberg reports.

The details:

Peloton is launching a next-gen stationary bike and treadmill in October, with a comprehensive hardware refresh aimed at reinvigorating its core lineup.
Insiders hint at on-device intelligence that adapts workouts on the fly and new ways to compete and measure performance beyond the classic output metrics.
Peloton also plans to release branded add-ons, such as smart weights, new sensors, and potentially other connected strength equipment.
In fiscal 2025, Peloton slashed expenses by 25% across sales, marketing, R&D, and administration, while closing 24 of its 37 retail showrooms.

Why it matters: Following years of sinking sales and staff cuts, Peloton is going toe-to-toe with Apple, Lululemon, and even Meta, all betting on an AI-powered fitness future. With machine learning baked into its bikes and platform, the company is chasing hyper-personalized workouts to win riders back before the clock runs out.

APPLE

🍏 Apple revives blood oxygen tracking

Image source: Apple

The Rundown: In a clever legal sidestep, Apple is reviving blood oxygen monitoring for select U.S.-based Apple Watch users after a months-long patent standoff with medical tech company Masimo that forced the feature’s removal last year.

The details:

Owners of newer Watch Series 9, Series 10, and Ultra 2 models can reclaim a workaround to the functionality via updates to iOS 18.6.1 and watchOS 11.6.1.
The watch’s sensors will gather oxygen data, but all measurements and calculations will happen on the paired iPhone, not on the Watch itself.
Blood oxygen results will also appear only in the Respiratory section of the Health app on iPhone; you can’t check readings directly on your Apple Watch.
Apple received a green light from U.S. Customs for this “redesigned” feature, signaling legal compliance and enabling immediate rollout.

Why it matters: This software reboot is Apple’s response to a U.S. International Trade Commission ban that knocked the blood oxygen app off the shelves for 18 months. By shifting the calculations off the wrist and onto the phone, Apple delivers a partial win to U.S. users. No more on-wrist readings, but it’s better than nothing for fitness fans.

HTC

👓 HTC debuts Meta-like smart glasses

Image source: HTC

The Rundown: Taiwanese tech giant HTC launched the Vive Eagle, its first AI-powered smart glasses, unveiled in Taipei with a sleek design that echoes Meta’s Ray-Ban Meta smart glasses.

The details:

The Vive Eagle features a lightweight, fashion-forward design (49 g) in a palette of translucent finishes and is engineered for everyday, discreet wear.
Its core hardware includes a 12MP ultra-wide camera, Qualcomm Snapdragon AR1 Gen 1 chip, 4GB RAM, and 32GB onboard storage.
Wi-Fi 6E and Bluetooth 5.3 enable easy phone pairing (Android 10+ or iOS 17.6+) using the Vive Connect app; open-ear stereo speakers are also built in.
The device includes HTC’s proprietary Vive AI assistant, compatible with both ChatGPT and Gemini.

Why it matters: The new AI glasses offer real-time translation in 13 languages, process data locally, and employ military-grade AES-256 encryption — so privacy is key here. Priced at NT$15,600 ($520), the specs look to take on Meta and Oakley smart glasses, but for now, they are hitting the shelves exclusively in Taiwan.

AMAZON

🛒 Amazon launches fresh same-day delivery

Image source: Amazon

The Rundown: Amazon just upped its quick-commerce game, launching Same-Day Delivery for fresh, perishable groceries — from milk to eggs to fresh salmon — in more than 1K U.S. cities as it takes direct aim at rivals like Instacart and Walmart+.

The details:

Customers can bundle fresh food with any other Amazon items — electronics, cleaning supplies, books — in a single order, all delivered within hours.
For Prime members, same-day grocery delivery is free on orders over $25; non-Prime customers pay a flat $12.99 fee per order.
Temperature-controlled fulfillment centers and insulated packaging ensure perishables arrive fresh, passing a six-point quality check before shipment.
Amazon also said it plans to more than double its coverage to over 2.3K locations in the coming months.

Why it matters: For Amazon, it’s about owning the anytime‑everything basket, collapsing the line between grocery store and e‑commerce checkout. For consumers, it’s all about convenience: a few taps, one cart, and within hours, dinner and everything else you need for the week show up in a single knock.

QUICK HITS

Apple accidentally leaked software code revealing that its next Vision Pro headset will use an M5 chip, the new iPad Mini will get an A19 Pro chip, among other chip upgrades.

Donald Trump signed an executive order that seeks to largely exempt commercial space launches from environmental review, a win for Elon Musk and Jeff Bezos.

Anthropic has acqui-hired the co-founders and core team of Humanloop, a London-based startup specializing in prompt management, LLM tooling, and AI observability.

Google pledged $9B over the next two years to expand its AI and cloud infrastructure in Oklahoma.

AI startup Cohere raised $500M in new funding at a $6.8B valuation, hiring former Meta and Uber executives.

Whoop is defying an FDA warning that its new blood pressure tracking tool reclassifies the Whoop 5.0 as a medical device requiring regulatory clearance.

Sam Altman is in early talks to co-found Merge Labs, a brain-computer interface startup that could rival Neuralink, aiming for an $850M valuation.

The U.S. Department of Energy chose 11 advanced reactor projects for Trump’s Nuclear Reactor Pilot Program, aiming to fast-track construction and operation.

Tony Robbins and Peter Diamandis’ Fountain Life raised $18M in new funding to expand its network of AI-driven preventive health centers across the U.S.

Lovable, the fast‑growing “vibe coding” startup, is targeting $1B in annualized sales by next year.

COMMUNITY

🎥 Watch our last workshop

Check out our last workshop with Brian Byrne, Customer Education Manager at Intercom. By the end of the workshop, you’ll confidently be able to deploy and manage AI agents that deliver faster, smarter, and more personalized customer service.

Watch it here. Not a member? Join The Rundown University on a 14-day free trial.

See you soon,

Rowan, Jennifer, and Joey—The Rundown’s editorial team

Superbugs meet their AI match

Rowan Cheung • 6 minutes

^{Read Online}^|^{Sign Up}^|^Advertise

Good morning, AI enthusiasts. Superbugs have been evolving faster than we can develop drugs to fight them, but MIT just put AI on the case to explore chemical territories humans never even knew existed.

With the system designing two brand-new antibiotics that attack in ways we've never seen before, AI’s new approach could take humanity into what researchers call a “second Golden Age” of drug discovery.

In today’s AI rundown:

AI’s new antibiotics to fight superbugs
Google’s new Gemma model is smaller than ever
Automate web monitoring with AI agents
HTC’s new AI glasses take aim at Meta
4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

AI & DRUG DISCOVERY

💊 AI’s new antibiotics to fight superbugs

Image source: MIT

The Rundown: MIT researchers just used AI to design two new antibiotics capable of killing drug-resistant gonorrhea and MRSA bacteria, potentially opening a new front against infections that cause millions of deaths annually.

The details:

Scientists trained AI models to generate 36M theoretical compounds, then screened them for bacteria-killing potential and human safety.
The algorithms produced two promising drugs (named NG1 and DN1) that attack bacterial cells through mechanisms never seen in existing antibiotics.
Both compounds cleared infections when tested in mice, with DN1 eliminating MRSA skin infections and NG1 combating drug-resistant gonorrhea.
The MIT research team said that AI advances in the drug sector could create a “second golden age” for the discovery of antibiotics.

Why it matters: Bacteria are evolving faster than our current drugs, but MIT's study shows that AI can navigate unexplored chemical territories that human researchers might never consider, potentially unlocking approaches that move antibiotic discovery from a game of catch-up to more proactive design.

TOGETHER WITH FREEPIK

💡 Creative AI tools for large teams

The Rundown: Freepik Enterprise gives your team access to AI-powered tools for creating and editing images, videos, and more—using multiple models tailored for different creative needs. Enjoy legal peace of mind, flexible admin controls, and unlimited users—trusted by 700K paying customers.

The complete enterprise package includes:

All AI tools in one place, with access to models like Runway, Veo, Kling, and more
Unlimited users with credit-based pricing that scales efficiently with usage
Legal indemnity included, protecting against IP claims on commercial AI content
SSO and admin controls with enterprise-grade security and compliance standards

Learn more about the platform trusted by 700,000 paying customers.

GOOGLE

🤏 Google’s new Gemma model is smaller than ever

Image source: Google

The Rundown: Google released Gemma 3 270M, an even smaller version of its open-source model family, which can run directly on smartphones, browsers, and other consumer devices while remaining efficient and capable at the same time.

The details:

Gemma 3 270M outperforms similarly small AI systems at following instructions, despite being a fraction of the size of most current models.
In internal tests, the model handled 25 conversations on a Pixel 9 Pro while consuming less than 1% of the battery, demonstrating extreme efficiency.
Developers can also fine-tune it in minutes for specific tasks, with Google demoing a Bedtime Story Generator as an example of an offline creative task.

Why it matters: As intelligence continues to scale, so do the capabilities of ultra-efficient, small models, making AI able to run on any consumer device. With Liquid AI’s LFM2 release also pushing the on-device model competition forward, some massive gains are being seen in the smallest corner of the AI world.

AI TRAINING

🔗 Automate web monitoring with AI agents

The Rundown: In this tutorial, you’ll learn how to use Yutori Scouts, an AI web monitoring agent that watches for specific updates online and alerts you via email. No more refreshing pages or manually checking for changes.

Step-by-step:

Enter your request in the Yutori homepage input box (e.g., “New releases from OpenAI or Anthropic or Gemini or xAI”) to create your Scout
Choose how often you want alerts — instant, daily, or weekly — then click Start scouting to activate it
View and manage all your active Scouts from the “My Scouts” dashboard, where you can edit, pause, or delete them anytime
Check reports sent to your email or in-app, each with clear findings and a direct link to the source for quick action

Pro tip: Use Scouts for time-sensitive opportunities like reservations, product restocks, or industry news, and pair with automations to get updates.

PRESENTED BY FIDDLER

🎧 The blueprint for production-ready enterprise AI agents

The Rundown: Fiddler AI's 13-minute podcast breaks down how to build agentic systems that deliver real value - sharing the observability patterns that separate demos from production-ready enterprise deployments.

In the episode, you’ll learn:

How to improve agent performance and diagnose the root causes of agent failures
Key patterns to monitor, from reflection loops to tool usage and planning accuracy
Examples of enterprise agentic systems and applications

Listen now.

HTC

😎 HTC’s new AI glasses take aim at Meta

Image source: HTC

The Rundown: Taiwanese giant HTC introduced Vive Eagle, a new line of AI glasses that let users choose between AI assistants and feature strong battery life, advanced translation capabilities, and other features to challenge Meta’s Ray-Ban dominance.

The details:

Users can switch between AI models from OpenAI and Google for the wearable’s assistant, activated via a “Hey Vive” voice command.
Built-in real-time photo-based translation works across 13 languages through an embedded camera, with all data processed locally for privacy.
Other features include a 12 MP ultra-wide camera, extended battery life, video recording capabilities, music playback, and more.
The wearable will currently only be available in Taiwan, with a starting price of $520 compared to Meta’s $300 Ray-Bans.

Why it matters: Zuck pointed to “personal devices like glasses” as the computing devices of the future, and competitors are emerging to compete with Meta's successful Ray-Ban (and now Oakley) lines. With styles gravitating towards normal, subtle integrations, it feels like a product close to breaking through to the mainstream.

QUICK HITS

🛠️ Trending AI Tools

⚡️ Gemma 3 270M - Google’s compact, hyper-efficient new open model
🤖 Anything - An AI agent for making products without coding
🌎 Hunyuan Gamecraft - Create interactive game worlds from a single image
✈️ Flight Deals - AI-powered search tool within Google Flights to find bargains

📰 Everything else in AI today

DeepSeek’s long-awaited R2 model is reportedly being delayed due to training issues with Huawei’s Ascend chips, after rumors of an August release circulated earlier.

Meta’s Superintelligence Lab added three more OpenAI researchers, with Alexandr Wang revealing Edward Sun, Jason Wei, and Hyung Won Chung have joined the team.

Cohere announced a new $500M funding round at a $6.8B valuation, also adding Meta’s VP of AI Research, Joelle Pineau, as its new Chief AI Officer.

T-Mobile parent company Deutsche Telecom officially launched its AI phone and tablet in European markets, which come integrated with Perplexity’s assistant.

Meta is facing backlash after a report revealed an internal document that outlined permitted AI outputs, which included romantic conversations with kids.

Google announced that its Imagen 4 image generation model is now GA in the company’s AI studio, with up to 2k resolution and a new fast model for quicker outputs.

Former Twitter CEO Parag Agrawal launched Parallel, a new startup creating a web API optimized for AI agents as users.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Michael C. in Phoenix, AZ:

"As a Creative Marketing Manager, AI has become part of my daily toolkit: ChatGPT Pro, and Grok for market research; 4o, Ideogram, Imagine, and Magnific for images; Higgsfield for video, and more. What’s been most surprising is the range of reactions from colleagues. Some are curious, a few are starting to experiment, but many remain hesitant or even resistant. I see the potential every day in how much faster and more polished my work has become, and I hope more of the team will discover those same benefits in their own roles."

How are you using AI? Tell us here.

🎓 Highlights: News, Guides & Events

Read our previous newsletter: Apple plots AI comeback with home robots
Today’s AI tool guide: Automate web monitoring with AI agents
Watch our last workshop: AI agents for Customer Service

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer—the humans behind The Rundown

Robotics

Figure 02 now folds laundry

Jennifer Mossalgue • 5 minutes

^{Read Online}^|^{Sign Up}^|^Advertise

Good morning, robotics enthusiasts. Figure AI just gave a glimpse of Figure 02 humanoid’s laundry folding skills — without human help or scripts, just its Helix neural network nailing towel duty.

The same model that does heavy logistics now casually handles one of our most mundane chores. If it can pivot from warehouses to laundry baskets, what’s next?

In today’s robotics rundown:

Figure 02 can now fold laundry
Nvidia’s new Cosmos world models
Sound-controlled swarms of microbots
New robot wrist smoothly opens doors, jars
Quick hits on other robotics news

LATEST DEVELOPMENTS

FIGURE

🧺 Figure 02 can now fold laundry

Image source: Figure

The Rundown: Figure just dropped a clip showing its Figure 02 robot folding laundry — autonomously, with no teleoperation or hardcoded programming — using nothing but its proprietary Helix neural architecture.

The details:

The video shows Figure 02 neatly grasping, folding, and stacking towels without human assistance, a world first for a robot with multi-fingered hands.
All folding was performed inside the home of Figure’s founder, without any special sensors or staged lab conditions.
Figure pulled it off with its Helix VLA model using just 500 hours of new human demonstration data to jump from warehouse work to household chores.
The robot’s hands feature multi-fingered dexterity and can trace edges, locate corners, smooth fabric, and adapt instantly to slips or wrinkling.

Why it matters: Figure achieved this using the same Helix model that recently worked a 20-hour shift at BMW’s Spartanburg plant, without any code changes or model retraining. All it took was 500 hours of fresh demo data. The result: a system that can generalize from warehouse tasks to homes just by being exposed to more examples.

TOGETHER WITH SANA

💡 The most powerful way to bring AI to your company

The Rundown: Most teams have tried AI tools, but connecting them to your business context changes everything. Sana Agents is an all-in-one platform that gives you no-code AI agents grounded in your company data. Deploy across teams in minutes, not months.

With Sana Agents, you can:

Draft reports, generate presentations, analyze data, and handle emails through agents
Connect to 100+ platforms, including Salesforce, Teams, and your CRM
Automate workflows that know your processes and permissions
Scale from isolated AI experiments to real work impact

Drive enterprise-wide innovation today with Sana Agents.

NVIDIA

🧠 Nvidia’s new Cosmos world models

Image source: Nvidia

The Rundown: Nvidia just dropped a robotics power-up at SIGGRAPH 2025: new Omniverse libraries and Cosmos world models to fast‑track physical AI, turning real‑world data into precision digital twins and smart autonomous agents.

The details:

Nvidia has released Cosmos Reason, a 7B parameter VLM built to help robots plan, adapt, and act in the physical world.
Also joining the current lineup is Cosmos Transfer‑2, a model designed to speed up synthetic data generation from 3D-simulation environments.
The new RTX Pro Blackwell Servers and DGX Cloud are designed to support robotics development for tasks like training, simulation, and deployment.
Nvidia also unveiled neural reconstruction libraries, plus a rendering method that uses sensor data to create 3D simulations of real-world environments.

Why it matters: Nvidia is making physical AI more accessible with faster tools for synthetic data and digital twin creation. The payoff could be shorter development cycles, safer robots, and a boost in industrial automation, opening high-end robotics to everyone from startups to global manufacturers.

SWARM BOTS

📢 Sound-controlled swarms of microbots

Image source: PennState

The Rundown: Penn State engineers have shown for the first time that sound waves can steer and coordinate swarms of micro‑robots, each just millimeters in size and equipped with a motor, tiny speaker, and microphone.

The details:

By “talking” and “listening” acoustically, these simple bots can synchronize movement, adapt to obstacles, and self‑organize into formations.
When confronted with obstacles, damage, or changing environments, the robots can reform, heal their collective, and adapt on the fly.
These self-organizing micro-machines could one day clean up pollution, deliver precision medicine, or probe dangerous, hard‑to‑reach places.

Why it matters: The system uses sound waves to let tiny robots communicate quickly, over long distances, and with little energy loss, an advantage over slower, less efficient chemical signals. Each robot is simple on its own, but together they can form flexible, resilient swarms for tasks in real-world settings based on current physics.

YALE UNIVERSITY

💡New robot wrist smoothly opens doors, jars

Image source: Yale University

The Rundown: Yale engineers just created the Sphinx Hand, a robotic gripper with a built‑in spherical mechanism that rotates objects in all directions without bulky wrist joints, letting robots work smoothly in tight spaces.

The details:

The Sphinx integrates a spherical mechanism right into the hand that can grasp and rotate everyday objects in all three axes (roll, pitch, and yaw).
That means twisting open jars, turning doorknobs, or screwing in lightbulbs, all without cameras, sensors, or elaborate control systems.
The design eliminates the need for a robot to move its entire arm for object rotation, enabling faster and more efficient manipulation in tight spaces.

Why it matters: This pared-down yet elegant design brings robots closer to human-like utility. By fusing the job of gripping and rotating directly at the point of contact, the Sphinx avoids awkward arm movements and adapts smoothly to complex scenarios. Possible use cases: untangling ropes, opening packaging, or handling fragile items.

QUICK HITS

📰 Everything else in robotics today

DHL is investing £550M ($746M) to deploy over 1K new robots across its UK operations, expanding automation in e-commerce and healthcare logistics.

EngineAI unveiled the T-800, its first heavy-duty, full-sized humanoid, standing 1.85m tall and weighing 85 kg with 41 DOF joints and a solid-state battery.

U.S. researchers developed muscle-like elasto‑electromagnetic actuators for insect‑sized soft robots, overcoming the limitations of miniaturized motors.

Chinese scientists developed GEAIR, an autonomous robot that performs genome editing and precision cross-pollination to breed new crop varieties faster.

Wang He, founder of Beijing Galbot, told the 2025 World Robot Conference that the market for humanoids could hit 100B yuan ($14B) within the next decade.

A lifelike AI-powered Tibetan antelope robot is now patrolling the high-altitude wilderness of Hoh Xil in Qinghai, China, collecting real-time data on wild antelopes.

HistoSonics, makers of robotic medical devices, announced that a management-led syndicate of private and public investors acquired its majority stake in a $2.25B deal.

Chinese robotics firm AgiBot signed a major agreement with an automotive parts maker to deploy nearly 100 A2‑W robots across the company’s factories.

Waldog, an AI-powered robotic dog, is roaming the streets of Monterrey, Mexico, to promote animal welfare and raise awareness about animal abuse, Reuters reports.

COMMUNITY

🎥 Watch our last workshop

Watch it here. Not a member? Join The Rundown University on a 14-day free trial.

See you soon,

Rowan, Jennifer, and Joey—The Rundown’s editorial team

Apple plots AI comeback with home robots

Rowan Cheung • 6 minutes

^{Read Online}^|^{Sign Up}^|^Advertise

Good morning, AI enthusiasts. Apple's answer to falling behind in AI? Build an army of home robots, complete with motorized arms, AI personalities, and a long-awaited Siri upgrade built from the ground up.

But with an Apple Intelligence emergency already on hand, can Apple's hardware talents overcome its internal AI struggles to crack the code on the next big home device?

In today’s AI rundown:

Apple plots AI comeback with home robots
OpenAI brings back 4o after GPT-5 anger
Create a self-updating slide deck with Manus
Microsoft goes on the offensive for Meta AI talent
4 new AI tools & 4 job opportunities

LATEST DEVELOPMENTS

APPLE

🏠 Apple plots AI comeback with home robots

Image source: Apple ML Research

The Rundown: Apple is pushing for a smart home push with four AI-powered devices, according to a new report from Bloomberg, including a desktop robot, smart display, and security cameras — targeting launches between 2026 and 2027.

The details:

The desktop robot will feature a motorized arm that controls a display to track users and lock onto speakers, infused with an AI-upgraded Siri.
Bloomberg said Apple is planning a personality-driven character for the new Siri called “Bubbles,” comparing it to Clippy from the original Microsoft Office.
A smart display is also set to arrive mid-2026, running a new ‘Charismatic’ OS for controlling home apps, music, web browsing, and more via voice or touch.
New AI-powered security cameras with months-long battery life will automate household tasks like dimming lights or playing personalized music.
Engineers are rebuilding Siri from scratch with AI models under the codename Linwood, while testing Anthropic's Claude as a backup codenamed Glenwood.

Why it matters: Apple’s much-hyped AI-powered Siri overhaul is still just talk, and the home AI robotics market has yet to see a product that truly sets the standard. With Google, Amazon, and OpenAI’s Jony Ive–designed device advancing quickly, Apple has little time to make its move.

TOGETHER WITH SANA

💡 The most powerful way to bring AI to your company

The Rundown: Most teams have tried AI tools, but connecting them to your business context changes everything. Sana Agents is an all-in-one platform that gives you no-code AI agents grounded in your company data. Deploy across teams in minutes, not months.

With Sana Agents, you can:

Draft reports, generate presentations, analyze data, and handle emails through agents
Connect to 100+ platforms, including Salesforce, Teams, and your CRM
Automate workflows that know your processes and permissions
Scale from isolated AI experiments to real work impact

Drive enterprise-wide innovation today with Sana Agents.

OPENAI

🔄 OpenAI brings back 4o after GPT-5 anger

Image source: ChatGPT

The Rundown: OpenAI CEO Sam Altman announced a series of changes to ChatGPT following backlash from the company’s GPT-5 launch, including the return of the popular 4o model, expanded rate limits, and new controls for model choice.

The details:

GPT-4o is returning to the model picker for all paid users, with Altman saying there will be “plenty of notice” if the model is ever deprecated.
Weekly rate limits for advanced reasoning in GPT-5 jumped from 200 to 3,000 queries, with Altman also clarifying the 196k context window for the new model.
Users also gain new "Auto," "Fast," or "Thinking" options for GPT-5, addressing anger from queries frequently being routed to the wrong model at launch.
Altman also revealed a personality update is coming for GPT-5, but said the real learning is the need for “per-user customization and model personality.”

Why it matters: GPT-5 is by nearly every measure a strong step forward, but a rocky rollout and forced user actions set a bad tone for what was a massively hyped launch. The 4o saga also shone the light on a (clearly larger than anticipated) corner of the user base that cares more about personality than coding or benchmarks.

AI TRAINING

🤖 Create a self-updating slide deck with Manus

The Rundown: In this tutorial, you’ll learn how to use Manus, an AI “super agent” that can handle complex, multi-step tasks, to build professional and data-rich slide decks automatically — without the manual work.

Step-by-step:

Give Manus a clear, detailed prompt in the dashboard (e.g., “10 AI opportunities for my hospitality business + 1 slide on near-term trends”)
Select Slides as your output, choose a theme, and click Confirm to start the process
Let Manus work in the cloud — it will research, gather data, and design your slides with visualizations like graphs while you do other things
When done, review the deck, export to Google Slides for edits, or go back to set up automatic updates for fresh content

Pro tip: Include audience, purpose, and format in your prompt to get a polished, ready-to-present slide deck.

PRESENTED BY IBM

🚀 Lockheed Martin partners with IBM for AI

The Rundown: Lockheed Martin’s AI takeoff began with data cleanup — cutting its tools by 50% and replacing them with a single unified system that runs on IBM’s watsonx.data to drive faster, smarter engineering.

Learn how Lockheed Martin leveraged watsonx to:

Help 10,00 engineers build scalable AI products in its AI Factory
Respond to employee questions using AI, boosting accuracy by 20%
Overcome data silos to create a single accessible, connected data environment

Read the full story.

AI TALENT WARS

🎣 Microsoft goes on the offensive for Meta AI talent

Image source: gpt-image / The Rundown

The Rundown: Microsoft is targeting Meta’s AI talent in a new recruiting offensive, according to a report from Business Insider, using multi-million dollar offers of its own to lure researchers from labs outside of the new Superintelligence Labs division.

The details:

Microsoft is reportedly aiming to match Meta’s compensation and using ‘special recruiting teams’, with a list of targets circulating via hiring managers.
Teams targeted include Reality Labs, GenAI Infra, and Meta AI Research, with recruiting led by Mustafa Suleyman and former Meta engineer Jay Parikh.
New processes for “critical AI talent” allow for streamlined offers and higher-up approvals within 24 hours.

Why it matters: Microsoft is taking a page out of Meta’s own playbook, though matching the money that Zuck has shown the willingness to give to top AI talent will be no small feat. That said, with reports of Meta’s AI unit being plagued by culture issues, it’s possible that some of the non-MSL employees are feeling ready for a move.

QUICK HITS