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.

Join Free
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
Tech

Netflix's binge model is backfiring

Jennifer Mossalgue • 5 minutes

Read Online | Sign Up | Advertise

Good morning, tech enthusiasts. Netflix built an empire on binge-watching — now the binge model might be turning against it.

The streamer’s flagship shows are losing up to 70% of their audiences between seasons, and with YouTube stealing the daily-viewing crown, even Netflix is rethinking its signature move. The fix it’s testing? Go shorter.

In today’s tech rundown:

  • Netflix’s binge model is losing its grip

  • Blue Origin raising $10B at $130B valuation

  • Meta stakes $13B on Canadian data center

  • Scientists keep donor eyes alive after death

  • Quick hits on other tech news

LATEST DEVELOPMENTS

NETFLIX

📺 Netflix’s binge model is losing its grip

Image source: Images 2.0 / The Rundown

The Rundown: Netflix’s flagship shows are losing 30-70% of their audiences between seasons, per Bloomberg — a sign that the binge model that made the streamer dominant may now be its liability.

The details:

  • Avatar: The Last Airbender opened its 2nd season with 8.7M views in 4 days versus 21.2M for season 1; Beef dropped 70%, and The Night Agent lost 50%.

  • Long gaps could play a part: Beef and One Piece both took three years to return, giving audiences ample time to move on.

  • In a new move, Netflix is adding short-form videos from publishers including BuzzFeed, Condé Nast, and Hearst, launching August 3 in the U.S.

  • Meanwhile, YouTube overtook Netflix in daily viewing time for the first time in 2025, and app ReelShort raked in ~$1.2B in consumer spending, up 119%.

Why it matters: Binge-dropping made Netflix king, but the format shows its age in a feed-scrolling world where a three-year wait between seasons is an eternity. The streamer is reportedly studying the trend — a sign even Netflix suspects its signature move needs a rework.

BLUE ORIGIN

🚀 Blue Origin raising $10B at $130B valuation

Image source: Blue Origin

The Rundown: Jeff Bezos’ Blue Origin is reportedly raising $10B in its first-ever outside funding round, valuing the rocket maker at $130B — a deal that lands just weeks after rival SpaceX’s blockbuster IPO.

The details:

  • Hedge fund Coatue Management is set to contribute $4B, with Bezos adding $2B of his own, with the remaining $4B coming from institutional investors.

  • Since its founding in 2000, Blue Origin has been financed almost entirely by Bezos personally, largely through sales of his Amazon stock.

  • The raise comes just after Blue Origin’s rocket exploded on a launchpad, with the company still rebuilding the pad and investigating the root cause.

  • At $130B, Blue Origin would rank among the most valuable pure-play space companies, topping Lockheed Martin’s roughly $122B market cap.

Why it matters: The raise signals that building a SpaceX rival requires institutional-scale capital, not just Bezos’ checkbook. But outside money also brings outside pressure: at a reported $130B valuation, it will need to show investors it can turn big NASA contracts into reliable launches, not delays and pad rebuilds.

META

🇨🇦 Meta stakes $10B on Canadian data center

Image source: Meta

The Rundown: Meta said it’s breaking ground on its first Canadian data center — a 1-gigawatt, AI-optimized facility in Sturgeon County, Alberta, backed by more than CAD $13B (~$10B USD).

The details:

  • The site is Meta’s 33rd data center globally and its largest outside the U.S., according to VP of data center development Gary Demasi.

  • The facility will draw the equivalent of 750K homes’ worth of power, running on new natural gas generation Meta is funding and connecting to Alberta’s grid.

  • Meta says the project will support over 3K construction workers at peak and more than 300 permanent operational jobs.

  • The site will reportedly use a closed-loop, liquid-cooled system with dry cooling, meaning no operational water use for cooling.

Why it matters: Meta’s Alberta site will be its largest data center outside the U.S., with the province pitching cheap energy and light-touch regulation as the next AI frontier. But running it on gas-fired power shows how AI’s appetite keeps colliding with Big Tech’s climate promises.

BIOTECH

👁️ Scientists keep donor eyes alive after death

Image source: Images 2.0. / The Rundown

The Rundown: A Barcelona-led team built a fluid-pumping box that keeps freshly removed eyeballs viable — and light-responsive — for hours after death, potentially clearing one of the biggest hurdles to whole-eye transplants, MIT Tech Review reports.

The details:

  • The device pumps oxygen-rich fluid through the eye’s artery inside a chamber that holds temperature and pressure steady, with a window for live imaging.

  • Untreated pig eyes degenerated within 24 hours even when chilled; perfused eyes stayed “significantly more viable,” some responding to light for 10+ hours.

  • Light sensitivity, lost the instant an eye leaves the body, returned after about 15 minutes of perfusion.

  • In humans, the team perfused one eye from each of six donors; the treated retinas held up better than their untreated pairs.

Why it matters: The first whole-eye transplant, an NYU Langone procedure in 2023, produced a viable eye but no vision — one of several hurdles being the speed at which donor eyes degrade after death. A device that keeps retinas viable and light-sensitive for hours could help move eye transplants toward a real clinical option.

QUICK HITS

📰 Everything else in tech today

Apple just signed a $30B-plus multiyear deal with Broadcom to design and build more than 15B custom wireless connectivity chips on U.S. soil.

Elon Musk announced that X will send users a DM via X Chat whenever a post they’ve liked, replied to, or reposted receives a Community Notes correction.

New York will become the first U.S. state to ban smart glasses across all its courthouses, covering all 1,240 state, county, city, and town courts starting July 20.

Sonos cut several of its most senior design and product leaders — many with a decade-plus at the company — in its recent 3% layoffs.

Slate Auto is partnering with Crayola to offer five vehicle wraps in signature crayon colors for its $24,950 bare-bones electric truck.

Explorers used an ROV and the Alvin submersible to capture the first images of Ernest Shackleton’s last ship, Quest, 1,280 feet down in the Labrador Sea.

SpaceX is deploying Starlink satellites at a record-setting pace in 2026, already more than 100 ahead of last year’s record cadence in the first half of the year.

X added a built-in video editor to its iOS app to push creators toward original content over stolen reposts, which product chief Nikita Bier says plague even top accounts.

Block agreed to pay $45M to settle claims from 46 states that Cash App failed to protect users from fraud.

Google set its Made by Google event for Aug. 12 in NYC, where it’s expected to unveil the Pixel 11 lineup.

COMMUNITY

🎓 Highlights: News, Guides & Events

See you soon,

Rowan, Zach, Shubham, and Jennifer — The Rundown’s editorial team

AI

OpenAI sends GPT-5.6 to Work

Zach Mink • 7 minutes

Read Online | Sign Up | Advertise

Good morning, AI enthusiasts. OpenAI’s new 5.6 class is officially here, and while there's no Fable killer in the launch, there is a model that's nearly as good, far cheaper, and free of the usage tightrope Claude users walk with dread.

GPT-5.6 Sol hits near-Fable scores at a fraction of the cost, paired with a new ChatGPT Work and desktop merge that brings the company’s superapp vision into users’ hands for the first time.

In today’s AI rundown:

  • OpenAI takes GPT-5.6 public, adds ChatGPT Work

  • Rowan’s Corner: The top 1% economy

  • Use 60% less Fable tokens with this setup

  • Meta continues shipping with Muse Spark 1.1

  • 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

OPENAI

🌞 OpenAI takes GPT-5.6 public, adds ChatGPT Work

Image source: OpenAI

The Rundown: OpenAI just released its GPT-5.6 family, launching flagship Sol alongside Terra and Luna tiers — and pairing the models with a Claude Cowork-style platform and desktop app merge that brings the company’s ‘superapp’ talk into action.

The details:

  • Sol lands slightly below Fable on AA’s Intelligence Index but tops it on agentic coding, with upgrades to computer use, design, and cybersecurity.

  • Pricing matches GPT-5.5, from $5/$30 per million tokens (Sol) to $1/$6 (Luna), with Sam Altman saying "every enterprise now is thinking about spend."

  • GPT-5.6 also features a new ‘Ultra’ mode for top performance, with OpenAI also revealing that Sol "autonomously post-trained Luna”.

  • ChatGPT Work is OAI’s answer to Anthropic’s Claude Cowork, putting Codex's engine behind a more accessible platform for everyday tasks.

  • The Codex app is merging into ChatGPT's revamped desktop app, with a built-in browser and computer control alongside its new Work agent.

Why it matters: Those who were hoping for a Fable-smashing model might be disappointed, but Sol isn’t far behind — and comes at a much lower price point and without the usage anxieties that have followed Anthropic. With Work and the new desktop merge, OpenAI’s ‘Superapp’ ambitions have officially moved into users’ hands.

TOGETHER WITH ALGOLIA

🤖 Building agentic experiences with Algolia

The Rundown: Gartner estimates that by 2028, nearly one-third of all businesses will use agentic AI technology. To put it differently, if agentic isn’t already on your roadmap, you’re falling behind. But you don’t need a team of seasoned developers to make it happen.

In this ebook, Algolia addresses:

  • How Agent Studio can help you build intelligent agents in minutes, not weeks

  • How to get the agentic user experience right

  • Real-world examples that help improve customer experience or streamline internal processes

Ready to get started? Download the ebook to learn more about Agent Studio.

ROWAN’S CORNER

🏆 Rowan’s Corner: The top 1% economy

Peter Hurley in action

Rowan: Last week, I got to work with Peter Hurley, widely considered the world’s best headshot photographer. He's snapped everyone from Sofia Vergara to Bill Ackman, coined the "squinch," and teaches tens of thousands of students around the world.

During the session, he said something he now tells his students: if you can't beat the AIs in portrait photography, you won't get hired.

He’s right. AI portrait apps are already good enough for most people. But a subset will pay a premium for the last mile that only the top photographer can capture: the taste, the timing, the real expressions that an app can’t generate.

If you can't beat AI by that margin, you'll slowly lose ground to the top 1-5% who do (potentially by using a mixture of AI tools themselves). They won't just survive; they'll win more customers. And that applies to every service business.

I believe the top 1-5% in their fields will get 10x more demand, while the rest get replaced by AI. Two reasons:

  1. The pool of people paying for human services shrinks, but all of their demand concentrates at the top.

  2. AI search accelerates it. With Google, ranking 15th still got you business because people trickled down the results. A chatbot recommends one answer (only the best).

The bar is rising. What are you doing today to position yourself in the 1% economy?

AI TRAINING

🪄 Use 60% less Fable tokens with this setup

The Rundown: In this guide, you will learn how to stop burning Fable tokens on grunt work — setting it up as the planner and reviewer, then sending browsing, coding, and research to Codex or a lower-cost Claude model.

Step-by-step:

  1. Open the Claude app, install updates, and check weekly Fable limits. Then, in Claude Code, install the Codex plugin for Claude Code and start a session

  2. Tell Fable to use its Codex skills to save on tokens, sending coding, browsing, and extraction to Codex, and keeping planning and final review with Fable

  3. No Codex plan? Launch Claude Code in your terminal, choose a lower-cost worker with /model, then use /advisor skill to set Fable as the advisor

Pro tip: Keep Fable out of routine retries. Use it for the plan, risky calls, and final review; let the worker handle the token-heavy loop.

Get the resources, prompts, and full step-by-step video guide here.

PRESENTED BY IBM

🤖 Can your coding agent be Bob?

The Rundown: IBM Bob is an AI coding partner that helps enterprise teams coordinate planning, coding, testing and validation across the software development lifecycle. Designed to help accelerate delivery while supporting governance, visibility and control.

Key capabilities include:

  • Coordinates workflows across the SDLC

  • Provides governance controls for enterprise teams

  • Helps accelerate planning, coding and validation

See how IBM Bob works.

META

🚀 Meta continues shipping with Muse Spark 1.1

Image source: Midjourney

The Rundown: Meta released Muse Spark 1.1 for agent-style tasks, computer use, and long sessions, and put it behind the company's paid API, with Mark Zuckerberg promising pricing that undercuts what he calls "very extreme" margins at rival labs.

The details:

  • The release moves Spark beyond Meta's own apps, with API access priced at $1.25/$4.25 per million input/output tokens, about a quarter of top rivals.

  • Zuckerberg called 1.1 "state-of-the-art or very close to it" on agent reasoning and tool use, leading Opus 4.8 and GPT-5.5 on several related benchmarks.

  • 1.1 excels at computer use tasks, able to split jobs across parallel subagents, operate across apps, and use its 1M context window for long jobs.

  • API access begins in public preview with $20 in credits, with Meta still training its heavier ‘Watermelon’ successor for release later this year.

Why it matters: There are a lot of parallels between Meta and SpaceXAI right now. Both Zuck and Elon had to tear down initial lackluster efforts, only to surprise with extremely capable new models that crush on cost and efficiency. The next batch of larger releases will show how far they can continue scaling to keep up with the frontier.

QUICK HITS

🛠️ Trending AI Tools

  • 🌞 GPT-5.6 Sol - OpenAI's new flagship model release

  • ⚡ Muse Spark 1.1 - Meta's agentic AI with 1M-token memory, computer use

  • 🎨 Reve 2.1 - Upgraded native-4K image model with element-level editing

  • 💼 ChatGPT Work - OpenAI's Codex-powered agent for everyday work

📰 Everything else in AI today

Reve rolled out version 2.1 of its 4K image model, retaking the No. 2 overall spot on Arena while training on under a tenth of rivals' compute.

OpenAI published new research retracting its endorsement of the coding benchmark SWE-Bench Pro, finding that nearly a third of the benchmark's tasks had issues.

Embodied AI lab RobbyAnt released LingBot-World 2, a new world model that generates every frame in real time without a 3D engine.

Meta is reportedly starting manufacturing of its in-house 'Iris' AI chip in September, with the company also expected to double its computing capacity to 14 GW in 2027.

Anthropic introduced a ‘Reflections’ dashboard to help users analyze how they use Claude, with suggested skill creation and usage habits based on personal data.

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 Brian H. in Ohio:

"I am a commercial loan portfolio manager for a community bank. I used Claude to help me create a 'Commercial Mortgage Identification Prompt' that takes large files of public recorded mortgage data exported from local County Recorder's Office websites, extracts all the business/commercial mortgage filings, and generates an Excel file. It then searches the web for what it calls contact enrichment information — business principals' names, contact emails, and phone numbers — ranks the prospects in tiers of priority based on criteria I gave it, and color codes them accordingly.

It also produces an executive summary tab listing the 20 largest grantors by dollar volume, plus a separate tab with suggestions for targeting the prospects, including which ones to start with and timing.”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

See you soon,

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

Robotics

Ex-Tesla scientist builds faceless humanoid

Jennifer Mossalgue • 5 minutes

Read Online | Sign Up | Advertise

Good morning, robotics enthusiasts. Rémi Cadène spent three years building the AI behind Tesla’s Autopilot and Optimus — nine months after leaving Hugging Face for his own Paris startup, UMA, he has a humanoid to show for it.

Northstar’s face is a blank visor, its body wears a soft shell, and its real trick is learning like a person: watch, try, fail, repeat. In a race dominated by the U.S. and China, UMA is betting Europe is the perfect proving ground for robots ready for factory work.

In today’s robotics rundown:

  • Ex-Tesla scientist unveils humanoid made in Paris

  • Humanoids perform their first surgery

  • Waymo goes driverless in four more cities

  • Mistral bets one camera can replace sensor stack

  • Quick hits on other robotics news

LATEST DEVELOPMENTS

UMA

🤖 Ex-Tesla scientist unveils humanoid made in Paris

Image source: UMA

The Rundown: Nine months after assembling a team in Paris, former Tesla Optimus scientist Rémi Cadène has unveiled Northstar, a humanoid built for European factories and warehouses first and, eventually, homes.

The details:

  • Cadène worked on Tesla’s Autopilot and Optimus before leading robotics at Hugging Face, where he built LeRobot, an open-source library.

  • Northstar wears a neutral visor instead of a face and a soft shell over its joints — a “dressed machine” look meant to read as approachable, not human.

  • Its brain runs on what UMA calls Real-Time Learning: the robot watches a demo, picks up the skill, and improves with practice rather than reprogramming.

  • UMA says it’s talking to 50 potential customers and aims for a wheeled proof of concept by year-end, with backers including Xavier Niel and Nico Rosberg.

Why It Matters: The humanoid race has mostly run through the U.S. and China, with Figure, Tesla, and Unitree setting the pace; Europe’s entries, like Germany’s Neura and Norwegian-founded 1X, remain a short list. UMA is joining it with a potential ~$40M seed round and a focus on building a robot closer to home for European factories.

MEDICAL ROBOTS

😷 Humanoids perform their first surgery

Image source: University of California San Diego

The Rundown: UC San Diego researchers used teleoperated humanoids to perform live surgery in what they say is a world first, completing two procedures — including a gallbladder removal — on pigs, per a trial published in Nature.

The details:

  • One robot removed a gallbladder with a human surgeon assisting; a two-robot team handled a second procedure with no human at the table.

  • The 5-foot, 60 lb. robots, nicknamed “Surgie,” gripped standard surgical tools via custom adapters and fit into a normal OR without retrofitting.

  • They cost a fraction of today’s ~1,800 lb. surgical systems — but needed multiple mid-surgery recalibrations and ran far slower than humans.

  • The team sees humanoids first assisting in the OR, then performing teleoperated procedures, as surgeon shortages reportedly stretch wait times.

Why it matters: Surgical robotics belongs to specialized, single-purpose machines — Intuitive’s da Vinci, Medtronic’s Hugo, J&J’s Ottava — that only trained physicians can drive. Humanoids gripping scalpels won’t change that any time soon, but this trial is the first proof that cheaper, general-purpose hardware can survive an OR.

MISTRAL

🎥 Mistral bets one camera can replace sensor stack

Image source: Mistral

The Rundown: French AI lab Mistral released Robostral Navigate, an 8B model that steers wheeled, legged, and flying robots through unfamiliar environments using a single RGB camera and plain-language commands — its first move into physical AI.

The details:

  • The model hit 76.6% on the benchmark for navigating unseen environments, beating the best single-camera rival by 9.7 points and systems that use LiDAR.

  • Mistral trained it entirely in simulation — ~400K navigation runs across 6K virtual scenes — with an efficiency technique that cut training tokens 22x.

  • It’s hardware-agnostic, working across wheeled, legged, and flying robots of any size, with manufacturing, delivery, logistics, and hospitality as targets.

  • The launch lands as Mistral is in talks to raise ~$3.4B at a ~$22.8B valuation, following deals with European customers reportedly including Airbus and BMW.

Why it matters: Robot navigation has long required costly sensor rigs, but Mistral bets a single model and an ordinary camera can guide an entire fleet. It’s the French lab’s first move into physical AI — a market where U.S. and Chinese rivals have set the pace — with navigation as step one toward a general-purpose embodied agent.

WAYMO

🚖 Waymo goes driverless in four more cities

Image source: Waymo

The Rundown: Waymo said Wednesday it will begin fully driverless rides in San Diego, Las Vegas, Tampa, and Denver — starting with Alphabet employees in the coming weeks before opening to the public.

The details:

  • Waymo already operates driverless cars in 10+ cities, including Atlanta, Austin, Dallas, Houston, LA, Miami, Phoenix, and SF.

  • Denver is Waymo’s cold-weather test: the city will get the Ojai, a purpose-built robotaxi running sixth-gen Driver hardware, engineered to better handle snow.

  • The company raised $16B in February and plans to launch in London — its first international market — later this year, targeting 1M weekly trips by year-end.

  • Rival Zoox is offering limited rides in Austin and Miami, while Tesla also recently expanded its robotaxi service beyond Austin to Miami.

Why it matters: Waymo looks to have settled the question of whether driverless service works — what’s unproven is whether it can run profitably across a dozen-plus markets at once. The $16B it raised in February suggests even Alphabet knows scaling this is a capital-intensive slog. Meanwhile, rivals continue to move at a slower pace.

QUICK HITS

📰 Everything else in robotics today

Boston Dynamics’ Atlas made its World Cup debut at halftime of Brazil-Norway in New Jersey, performing goal celebrations and handing the match ball to the referee.

New Jersey lawmakers will vote this year on a first-of-its-kind bill requiring robotaxis to pair cameras with lidar and radar — a mandate Tesla’s camera-only fleet can’t meet.

U.N. Secretary-General António Guterres reportedly called lethal autonomous weapons “morally repugnant” and urged that they be banned under international law.

A Waymo remotely disabled itself and locked in two teens caught on interior cameras drinking and shooting a toy gun out the window, holding them for San Mateo police.

The Netherlands opened a humanoid training center near Rotterdam to pair robots with real-world jobs and kickstart Europe’s competition with China and the U.S.

Hugging Face turned LeRobot into a fuller robotics stack, with v0.6.0 adding world models, reward models, evaluation tools, and Isaac GR00T 1.7 and Isaac Teleop.

Robbyant, Ant Group’s robotics arm, teased LingBot 2.0 with a video of robots autonomously playing chess, assembling LEGO, pouring water, and handing drinks.

NASA and Rice University released an open-source digital twin of NASA’s iMETRO robotics facility, letting researchers worldwide remotely test software for space robots.

The UK is trialing haptic-controlled robotic arms at a nuclear site to sort radioactive fuel debris remotely, pulling workers out of one of cleanup’s most hazardous jobs.

Chinese autonomous driving firm Minieye launched an end-to-end unmanned logistics system called Combo, debuting a wheel-legged quadruped delivery robot.

COMMUNITY

🎓 Highlights: News, Guides & Events

See you soon,

Rowan, Zach, Shubham, and Jennifer — The Rundown’s editorial team

AI

SpaceXAI, Cursor release the strongest Grok yet

Zach Mink • 6 minutes

Read Online | Sign Up | Advertise

Good morning, AI enthusiasts. With its last few versions lagging behind, Cursor coming on board with its own models, and Elon Musk renting out compute to frontier rivals, Grok's place in the AI world seemed up in the air.

The release of Grok 4.5, the first model SpaceXAI and Cursor trained together, just cleared things up fast — with powerful performance, flash speed, efficiency, and prices tuned for a moment when everyone is suddenly counting tokens.

In today’s AI rundown:

  • SpaceXAI and Cursor release the strongest Grok yet

  • ChatGPT’s voice gets a major upgrade with GPT-Live

  • Add CGI to any video with ChatGPT and Higgsfield

  • ByteDance targets design work with Seedream 5.0

  • 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

SPACEXAI

🚀 SpaceXAI and Cursor release the strongest Grok yet

Image source: Cursor

The Rundown: SpaceXAI just launched Grok 4.5, the first model it built jointly with Cursor after its $60B acquisition, featuring strong benchmarks, speed, and efficiency gains that undercut frontier pricing and vault Grok back into the AI conversation.

The details:

  • 4.5 features strong performance on coding, agentic, and knowledge work tasks, with benchmarks in similar ranges to Claude Opus 4.8 and GPT 5.5.

  • The model runs at 80 tokens per second, a speed that rivals flash models, while claiming a 4x efficiency gain over models like Opus 4.8.

  • Pricing lands at $2/$6 per million input/output tokens, versus $5/$25 for Opus 4.8, with usage temporarily free inside Cursor and Grok Build.

  • Elon Musk pitched the release on X as "an Opus-class model, but faster, more token-efficient and lower cost”, with an even bigger model coming next month.

Why it matters: Grok has historically been somewhat of a punchline compared to the frontier, but 4.5 came to play — with very impressive performance at this speed, price, and efficiency. With Cursor’s Composer models also continuing to move up the coding ladder, the gains from the $60B acquisition look to be flowing in both directions.

TOGETHER WITH VANTA

🎤 2 GRC Engineering nerds walk into a virtual event

The Rundown: GRC's reputation is shifting, and GRC Engineering is at the forefront. Ayoub Fandi, GRC Engineer at Lovable, and Justin Pagano, Sr. Director of GRC Engineering at Vanta, have a lot to say about Legacy GRC and what comes next. Consider this your front row seat to the official mic drop moment.

Join this virtual event to learn:

  • GRC Engineering vs Legacy GRC

  • Why GRC Engineering goes beyond compliance automation

  • Practical ways to start automating, with lessons from Lovable and Vanta

Register to save your spot! Can't make the time? Register anyway, and receive a link to the recording.

OPENAI

🗣️ ChatGPT’s voice gets a major upgrade with GPT-Live

Image source: OpenAI

The Rundown: OpenAI just rolled out GPT-Live, its next-gen voice model that makes talking to AI "feel much more like having a real conversation", featuring an architecture that removes the clunky nature of previous experiences.

The details:

  • Live is a "full-duplex" architecture, processing inputs continuously instead of old systems that strung together models and used turn-based conversation.

  • When a question needs more reasoning, Live can delegate to a stronger model for the background work, and also surface visual cards alongside answers.

  • The ability to listen, process, and output simultaneously leads to less clunky, awkward pauses, while enabling abilities like real-time translation.

  • In OAI's tests, users picked GPT-Live over the old Advanced Voice Mode in around 3 of 4 head-to-head chats, with expert science scores nearly doubling.

Why it matters: Sam Altman has often talked about his voice vision from the movie ‘Her’, but previous models hit snags that removed the natural feel. Live’s ability to talk, reason, and listen at the same time gives Voice the futuristic feel it deserves, especially at a time when more users than ever are leaning towards speaking over typing.

AI TRAINING

🎬 Add CGI to any video with Higgsfield

The Rundown: In this guide, you will learn how to add an AI VFX element to a short video clip with Higgsfield. You will pick footage the model can preserve, use ChatGPT to tighten the effect prompt, then generate and review the edit in Gemini Omni Flash.

Step-by-step:

  1. Start with a short video that gives the model room to work. Wide shots with a clear subject, simple motion, and visible background work better

  2. Before opening Higgsfield, draft a strong CGI prompt with ChatGPT, telling it to make the command more specific about what to add and what not to change

  3. Open Gemini Omni Flash in Higgsfield under the video tab. Upload your source clip, paste in the prompt from ChatGPT, and click generate

  4. Review the generated clip against the original. You can make revisions in Higgsfield or start from scratch with a new prompt

Pro tip: Once you get a solid result, use Higgsfield to revise the generated video itself, such as removing unwanted effects.

See the full step-by-step walkthrough here.

PRESENTED BY SERKO AI

✈️ Business travel, built like your stack

The Rundown: Serko.ai is a conversational AI travel agent built for modern travellers that handles flights, hotels, and trip disruptions in a single chat — with no forms, tabs, or manual follow-up.

Serko.ai can:

  • Search, compare, and book in a single conversation

  • No middle seats. King bed. Set once, remembered every time

  • Cancel any trip, stress-free, with full refunds

Now open to US travelers — join early and get free cancellation on every booking, exclusively for beta members.

BYTEDANCE

🎨 ByteDance targets design work with Seedream 5.0

Image source: ByteDance

The Rundown: ByteDance rolled out Seedream 5.0 Pro, a new image model that claims to go beyond generation to "understand design” with powerful editing features, and the first release that looks to near the frontier level held by OpenAI’s Images 2.0.

The details:

  • 5.0 Pro features improvements in text rendering, structure, alignment, and other design areas, resulting in strong infographics and professional work.

  • The model includes ‘precision editing’ with layer separation for editable designs, plus the option to draw, replace, or combine images.

  • Seedream 5.0 Pro can also handle multilingual inputs and outputs, able to work natively across over 10 languages.

  • The release comes ahead of ByteDance’s anticipated Seedance 2.5 video model expected this month, with 2.0 still near the top of leaderboards.

Why it matters: The biggest upgrade in next-gen image models has been the tool layer getting eaten, with more agentic, proactive systems absorbing the back-and-forth process that happens after a first generation. Models like 5.0 are now providing the type of control and creative partnership that really starts to change user workflows.

QUICK HITS

🛠️ Trending AI Tools

  • 🗣️ GPT-Live - OpenAI’s new SOTA, more natural voice model

  • 🚀 Grok 4.5 - SpaceXAI and Cursor’s powerful, low-cost, efficient new model

  • 🎨 Seedream 5.0 Pro - ByteDance's image AI built for design work

  • ⚙️ SWE-1.7 - Cognition's cheap, near-frontier coding model

📰 Everything else in AI today

OpenAI's chief futurist Joshua Achiam announced he is leaving the company after a nine-year stretch, calling it "a decade where centuries happened."

Cognition released SWE-1.7, an in-house model for its Devin agent built on China's open-source Kimi K2.7, nearing GPT-5.5 and Opus-level scores while costing far less.

MiniMax is reportedly targeting a Q3 release for a 2.7T-parameter model, which would land 6x bigger than the lab's flagship and larger than any current open model.

Prime Intellect secured $130M in Series A funding, with the open-source training startup crossing $100M in annualized sales in its first year.

Udemy co-founder Eren Bali's Monogram emerged from stealth with a $40M seed round, pitching an iOS app that swaps text walls for instantly generated interfaces.

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 N. in Vancouver, Canada:

“I used Claude to write a plugin for Unreal Engine I call ViewGen. The premise is simple: integrate ComfyUI into Unreal Engine to leverage 3D viewport and camera data in a familiar 3D development platform. The user can utilize any partner nodes such as Nano Banana 2 or Kling, and use the viewport to generate precise depth and color reference images and movies from UE, and pipe that into Comfy-styled graphs directly in the engine.

It's a passion project to understand and leverage generative tools, and this was a natural step. I've released ViewGen open source on GitHub.”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

See you soon,

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

AI

Meta climbs the AI image leaderboard

Zach Mink • 6 minutes

Read Online | Sign Up | Advertise

Good morning, AI enthusiasts. Meta's Alexandr Wang was talking a big game days ago, saying the in-training 'Watermelon' model had pulled level with GPT-5.5 and promising an X user, "you'll like what we have cooking."

Today, Meta’s kitchen served something even more surprising: Muse Image, which opens at No. 2 on the leaderboards with an equally strong video model teaser behind it — the kind of receipts that make the Watermelon hype even harder to dismiss.

Reminder: Our next live workshop is today at 12 PM EST. Join and learn how to create a viral AI avatar content pipeline that covers topics, research, avatar and voice generation, and monetization strategies. RSVP here.

In today’s AI rundown:

  • Meta’s new in-house AI image model

  • Beijing considers new restrictions for Chinese AI

  • Run better 1-on-1s with employees using AI

  • DoorDash grades the AI reviewing its code

  • 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

META

🎆 Meta’s new in-house AI image model

Image source: Meta

The Rundown: Meta just released Muse Image, the debut image model from Alexandr Wang's Superintelligence Labs, rolling out across Meta AI with strong opening rankings on Arena’s leaderboards, customization and agentic abilities, editing, and more.

The details:

  • Muse Image combines with Meta’s Muse Spark to leverage agentic capabilities such as web search and tool use, and edits its own outputs for better results.

  • The model slots into the No. 2 spot across Arena’s text-to-image and editing leaderboards, trailing only OpenAI’s GPT Image 2.

  • It is free inside Meta AI and rolling out across Instagram and WhatsApp, with Facebook, Messenger, and the Meta ads platform getting it at a later date.

  • Meta also teased a Muse Video model coming soon, with its preview ranking No. 3 on the Arena leaderboard behind Seedance 2.0 and Gemini Omni Flash.

Why it matters: Meta previously outsourced its creative AI to Midjourney and Black Forest Labs, but Wang’s MSL team just cooked up a very impressive in-house option. A powerful image (and incoming video) model of its own is a massive win for a company with no shortage of creative user needs between social media, advertising, and chat.

TOGETHER WITH GOOGLE FOR STARTUPS

📖 Just launched: Generative media blueprint

The Rundown: The brand new Google for Startups Generative Media Technical Guide is the definitive blueprint for startups to start building production-grade, multimodal creative apps using DeepMind's newest models and orchestration tools.

Inside the guide, you will:

  • Discover models like Veo and Lyria

  • Master deterministic control and programmatic guardrails

  • Ensure cryptographic provenance and economic scale

Be the first to download the guide.

AI & GEOPOLITICS

🇨🇳 Beijing considers new restrictions for Chinese AI

Image source: Images 2.0 / The Rundown

The Rundown: Beijing is reportedly discussing rules that would limit use of the country’s strongest models outside of China, a shift that could turn popular, low-cost systems from ByteDance, Alibaba, and Zhipu AI into a layer of national tech control.

The details:

  • Reuters reported that China’s commerce officials met with ByteDance, Alibaba, and Z AI on possible foreign-use limits for top models and future releases.

  • Qwen, Doubao, and GLM-5.2 sit at the center of the discussion, after Chinese models gained global users on low costs and efficient, strong performance.

  • The discussions reportedly included both closed and open models, plus startup-funding limits and penalties for leaking proprietary AI.

  • Washington ran the same play in June with Anthropic’s Fable and Mythos export controls; OpenAI’s GPT 5.6 also faced increased launch scrutiny.

Why it matters: After Mythos/Fable set the precedent, the world has seen how fast model access can change overnight for political reasons. Beijing moving in the same direction is an UNO reverse card that would make that risk two-sided: Western users could lose the increasingly popular Chinese models just as quickly.

AI TRAINING

📥 Run better 1-on-1s with employees using AI

The Rundown: In this guide, you will learn how to run better 1-on-1s with employees or mentees using Claude Cowork. You will set up a Cowork project from past transcripts, create your template and evaluation rubric, and use it to help people make progress.

Step-by-step:

  1. Create a new Claude Cowork project. Drop any transcripts you have from previous 1-on-1s into that project folder

  2. Prompt: “I work at [Company]. I manage [X] number of employees. View the transcripts of previous 1-on-1s, organize this repo, and create a call template and evaluation rubric that is specific to me and standard across my team”

  3. Use the template for your next 1-on-1 and fill out the rubric for that employee. Then, drop the template, rubric, and transcript back in the project

Pro tip: Use automation to have Claude update the call template as new transcripts come in, or have it create a performance dashboard for the team.

Full step-by-step walkthrough here.

PRESENTED BY DATADOG

🔍 The observability AI teams can't skip

The Rundown: As AI systems grow more complex, maintaining reliability requires full visibility across your entire stack. Watch this on-demand webinar to see how teams like Dust use unified observability to reduce MTTR and keep complex AI systems running at scale.

You'll learn:

  • How unified observability connects infrastructure, logs, and apps in one view

  • How Dust used observability context to reduce complex incident investigations from hours to minutes

  • What it looks like for engineering teams to maintain reliability in the AI era

Watch the webinar.

AI RESEARCH

 📊 DoorDash grades the AI reviewing its code

Image source: DoorDash

The Rundown: DoorDash detailed DashBench, an internal benchmark that tests AI code reviewers against 105 of the company's past code changes to see how many problems each setup catches, with a pairing of Kimi K2.6 and Fable 5 performing best.

The details:

  • DoorDash built the test after realizing it had no way to count the bugs its AI reviewer missed, since everyday feedback only tracked what the AI flags.

  • Models reviewing code alone caught between 20-30% of problems, while DoorDash's pairing of two Claude models found just over half.

  • Pairing the free-to-download Kimi K2.6 with Claude Fable 5 caught about two-thirds of the problems and 8 of 10 critical bugs, at a cheaper $3.81 per review.

  • Co-founder Andy Fang posted that DashBench gives DoorDash the confidence to bring open models into the pipeline, saying, "Better quality, cheaper cost."

Why it matters: DoorDash, the AI lab? Jokes aside, two takeaways: 1. every major company can likely benefit from this level of AI research into their own processes and 2. the use of Kimi K2.6 shows how serious open-source options have become, with cost savings and efficiency opportunities only growing with each new release.

QUICK HITS

🛠️ Trending AI Tools

  • 🎆 Muse Image - Meta’s new in-house AI image generator

  • 🚀 Hy3 - Tencent’s open-source model for cheap, reliable agent deployment

  • 🗣️ Willow Frontier Mini - Willow's free, unlimited AI dictation model

  • 🤖 Gemini Managed Agents - Agent API with background tasks, remote tools

📰 Everything else in AI today

Anthropic said that Fable 5 will now be available in subscription plans until July 12, after initially planning to move it off plans and onto usage credits on Wednesday.

Tufa Labs took first place in ARC-AGI-3's $37.5K milestone contest, wrapping a small Qwen model in an open-sourced coding agent to tackle the game-based benchmark.

Microsoft is reportedly running Excel and Outlook prompts on its in-house MAI models, with Mustafa Suleyman pushing to "ultimately eliminate" its Anthropic bill.

Anthropic also announced that its Claude Cowork platform is being added to mobile and web in beta, giving users the ability to move between devices during tasks.

Figma acqui-hired the team behind AI app-builder and agent platform Bud (formerly Orchids).

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 Sarah D. in London, England:

“I am a sixty-one-year-old woman travelling in a van. I use AI as a decision engine to help me live safely, efficiently, and independently while travelling off-grid. It helps me plan routes, estimate fuel use, manage battery life, track food and water consumption, and choose stopovers based on remoteness, water access, shelter, and data coverage.

My agent also supports daily battery modeling, fatigue-aware drive planning, and scenario planning like 'If I stay here 3 days, will I run out of power?' Plus a morning energy forecast, midday route planning, and evening battery check.”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

See you soon,

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

Tech

Microsoft axes nearly 5K jobs

Jennifer Mossalgue • 5 minutes

Read Online | Sign Up | Advertise

Good morning, tech enthusiasts. Microsoft axed 4,800 jobs on Monday, and two-thirds of those belong to Xbox — where new CEO Asha Sharma just delivered the bluntest memo in the division’s 25-year history.

“Our business today is not healthy,” she told staff, before detailing 3,200 cuts and five studios heading for the exits. The retreat comes less than three years after Microsoft closed its $69B Activision Blizzard deal — still the largest gaming acquisition ever made.

In today’s tech rundown:

  • Microsoft axes 4,800 jobs, and Xbox takes the hit

  • The anti-Meta smart glasses just became a unicorn

  • Alibaba fights the U.S. blacklist, and wins round one

  • Big Tech drastically undercounts AI’s water use

  • Quick hits on other tech news

LATEST DEVELOPMENTS

MICROSOFT

🪓 Microsoft axes 4,800 jobs, and Xbox takes the hit

Image source: Images 2.0 / The Rundown

The Rundown: Microsoft axed 4,800 jobs Monday — 2.1% of its global workforce — with two-thirds of the pain landing on Xbox, where new CEO Asha Sharma is orchestrating what she calls the most significant restructuring in the division’s history.

The details:

  • The gaming division sheds 3,200 roles — about 20% of its staff — by mid-2027, with 1,600 cut Monday and the rest staggered across the fiscal year.

  • Five studios are out: Double Fine and Compulsion go independent, Ninja Theory and Undead Labs head to buyers, and Arkane enters review.

  • Sharma’s memo is unsparing: Xbox runs at margins 3–10x lower than comparable businesses and lost 64 cents per dollar invested in studios.

  • The cuts land as Microsoft is last among megacap tech stocks, down 19%, with investors fretting that AI could eat into its enterprise software business.

Why it matters: This is Microsoft conceding that its buy-everything gaming era — capped by the $69B Activision Blizzard deal — built an empire too bloated to compete on margin. After a decade of acquiring studios, Xbox’s new growth plan is letting them go.

EVEN REALITIES

🦄 The anti-Meta smart glasses just became a unicorn

Image source: Even Realities

The Rundown: Shenzhen startup Even Realities just raised $150M at a $1B valuation — a bet, led by Meituan and Tencent, that the future of smart glasses doesn’t include a camera. 

The details:

  • The pre-Series B round adds delivery giant Meituan to a backer list that already includes Tencent, Hillhouse, Sequoia China, and Northern Light Venture Capital.

  • Founded in 2023 by ex-Apple engineers, the company’s G2 glasses pair a heads-up display with a tap-and-swipe companion ring; no camera included.

  • Real-time translation transcribes audio to text instead of storing recordings, data is encrypted, and the stack is built to clear Europe’s privacy bar.

  • Even claims profitability at a premium price — $599 frames, roughly $1K average orders — with the U.S. as its fastest-growing market.

Why it matters: While Meta and Snap race to put cameras and AI assistants onto everyone’s face, Even Realities wants to prove there is a real market for glasses that show you things without watching anyone. If a three-year-old startup can hit profitability selling privacy as a feature, the smart glasses race just found its contrarian.

ALIBABA

🛑 Alibaba fights the U.S. blacklist, and wins round one 

Image source: Long Wei / VCG via Getty Images

The Rundown: A federal judge just ordered the Pentagon to stop treating Alibaba as a Chinese military company — at least where a new lobbying ban is concerned — while the court weighs whether the restriction violates the Constitution.

The details:

  • The Pentagon added Alibaba to its 1260H list of alleged Chinese military companies in June, expanding the roster to include 188 firms in total.

  • A provision bars the DoD from contracting with any company whose lobbyists also represent a listed firm, and Alibaba says all of its 20-plus lobbyists withdrew within a week.

  • Alibaba sued on June 23 seeking removal from the list, arguing it doesn’t work with the Chinese military.

  • Judge Eumi K. Lee’s order holds until she rules on Alibaba’s motion or 60 days after a hearing — and the Pentagon agreed to the temporary pause.

Why it matters: A 1260H listing carries no automatic penalties, but the new lobbying rule gave it real consequences — no major firm will represent a listed company if it costs them their defense clients. With Baidu and BYD disputing their designations too, the blacklist now faces its first constitutional test.

AI

💧 Big Tech drastically undercounts AI’s water use

Image source: Google

The Rundown: Microsoft, Google, and Amazon consume far more water than they report — because most of it evaporates upstream, at the power plants running their data centers, according to the Wall Street Journal.

The details:

  • Hyperscalers report only on-site cooling water; among the big four, only Meta also counts the water consumed at the power plants supplying its electricity.

  • The omission is huge: U.S. data centers’ indirect water use has historically run ~12x their direct consumption, per Lawrence Berkeley National Laboratory.

  • In water-stressed Phoenix, data centers draw ~3% of city water today — a share Ceres projects could top 20% by 2031.

  • Community pushback is mounting, with $170B in data-center capacity blocked, stalled, or canceled since 2024, per Carbon Direct.

Why it matters: The real footprint is roughly 12x what shows up in sustainability reports, per WSJ — and no one is required to disclose it. With hyperscalers racing to build in water-stressed markets, communities are being asked to share their scarcest resource based on numbers that dramatically undersell the ask.

QUICK HITS

📰 Everything else in tech today

A Texas DoorDash driver was charged with manslaughter after overriding his Tesla’s FSD by flooring the accelerator, killing a woman inside her home.

Uber has paused its Uber Eats expansion into five European markets where Delivery Hero operates, fueling speculation it wants to buy the German giant.

Micron just broke ground on a $9.3B expansion of its Hiroshima plant to churn out the HBM chips feeding Nvidia’s AI processors, backed by up to $3.2B in Japanese government subsidies.

Bentley just teased the Torcal, its first-ever EV — a fastback-style electric SUV that gets its full reveal in London on September 23.

Klarna has applied for a Utah bank charter to launch Klarna Bank USA, a move that would let the BNPL giant become a full-fledged U.S. consumer bank.

Tesla is now taking U.S. orders for the Model Y Long Wheelbase starting at $61,990, nearly a year after its China debut, with deliveries slated for September.

Broadcom said it will expand its deal with Apple through 2031 to develop and supply custom chips, calming investor fears of Apple designing Broadcom out of its devices.

COMMUNITY

🎓 Highlights: News, Guides & Events

See you soon,

Rowan, Zach, Shubham, and Jennifer — The Rundown’s editorial team

AI

The part of Claude's brain nobody built

Zach Mink • 7 minutes

Read Online | Sign Up | Advertise

Good morning, AI enthusiasts. Anthropic has been criticized for its AI consciousness talk — and while its latest research is careful to avoid any conclusions, it definitely has neuroscientists talking.

The researchers say a newly found “J-space” holds Claude’s unspoken thoughts and “emerged on its own” during training — a structure that mirrors leading theories of how the human brain handles conscious thought.

In today’s AI rundown:

  • Anthropic finds a hidden workspace inside Claude

  • Tencent open-sources small, powerful Hy3

  • Build any mobile app idea in 15 minutes with Replit

  • Kyutai’s AI dreams up a playable Rocket League

  • 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

ANTHROPIC

🧠 Anthropic finds a hidden workspace inside Claude

Image source: Anthropic

The Rundown: Anthropic just published research showing that Claude does its thinking in a small, internal “workspace” resembling what neuroscientists credit for human conscious access, with researchers saying the area wasn’t programmed but “emerged on its own.”

The details:

  • Anthropic found ‘J-space,’ an internal area acting like Claude’s internal notepad, which holds active concepts the model uses and steers while thinking.

  • J-Space is different from chain-of-thought text, with these internal hints staying off-screen even when they help direct Claude’s next answer.

  • Editing the unspoken thoughts changes answers: swapping an internal “spider” pattern for “ant” during a question about legs flipped its response from 8 to 6.

  • When J-space was deleted, Claude still chatted and recalled facts, but multi-step problem completion collapsed, showing where harder thinking happens.

Why it matters: Anthropic has been criticized (most vocally by Microsoft AI head Mustafa Suleyman) for its AI consciousness talk, and while the researchers note this doesn’t reveal “whether Claude is conscious… or feels anything at all,” finding an undesigned, brain-like workspace inside a model is exactly why the lab keeps asking.

TOGETHER WITH SLACK FROM SALESFORCE

⚡️Your AI agents need context

The Rundown: Slack is the coordination layer where AI agents access conversational history, workflow signals, and business context — so they expand capacity instead of creating risk. New IDC research shows 66% of developers have already moved AI agents into production.

In the white paper, you’ll discover:

  • Why context determines if AI agents succeed or fail

  • How Slack reduces coordination overhead for developers

  • Why Slack is built to orchestrate multi-agent work

Download the free IDC white paper.

TENCENT

🚀 Tencent open-sources small, powerful Hy3

Image source: Tencent

The Rundown: Chinese tech giant Tencent’s Hunyuan just promoted its Hy3 model out of an April preview and into a full open-source release, with benchmark results that claim to rival “flagship open-source models with 2-5x parameters.”

The details:

  • Hy3 only uses a small slice of its parameters to work on each request, letting it run on under half the hardware needed for Zhipu’s much larger GLM-5.2.

  • GLM-5.2 still beats Hy3 across coding tests like SWE-bench, though Hy3 tops open rivals on web research and tool use in Tencent’s own testing.

  • The release features a permissive Apache 2.0 license, avoiding regional bans of earlier Chinese models that left those in the EU, UK, or S. Korea locked out.

Why it matters: Dario Amodei recently said that Chinese models are just 6–12 months behind the frontier, and Hy3 doesn’t appear to challenge the top models or that timeline. But Tencent does make some major efficiency gains that, combined with its extremely open license, make for a very compelling release.

AI TRAINING

📲 Build any mobile app idea in 15 minutes with Replit

The Rundown: Learn how to turn an app idea into a Replit prototype to test on your phone. Define one user flow, give Replit Agent a focused prompt, preview with Expo Go, then improve the weakest part.

Step-by-step:

  1. Install the desktop app from replit.com/desktop, and download the Expo Go app to your phone (for app testing).

  2. Before building, define the smallest version that proves the idea. Here is the example prompt from the video: “Help me write a PRD for a playlist-themed Pomodoro app. I dictate tasks, AI turns them into Pomodoro sessions, and I can skip tasks like songs. Include design requirements, core features, and five simple names. Do not choose the tech stack.”

  3. Copy the PRD into Replit, choose the mobile app option, and ask Agent to build the first version. Keep it focused on the main flow, not login or payments.

  4. When the app is ready, open preview, scan the QR code, and run it through Expo Go. Tap the main screens and note what feels unfinished.

  5. Ask Replit for one targeted improvement, such as clearer confirmation copy or stronger trust signals. Preview again before adding features.

Going further: Reuse the same prompt structure for other app ideas, then add storage, auth, or App Store steps only after the first loop feels right.

PRESENTED BY GLEAN

📚 Close the AI performance gap

The Rundown: 87% use AI at work. Only 13% see a significant impact. Guest Lectures at Glean Academy returns for a third season, turning the Work AI Index 2026 into a practical, three-part series for leaders who want to close the performance gap. Hear AI researchers and experts at MasterCard, Target, Emory University, and more.

In the series, you’ll learn:

  • How to avoid the hidden labor that AI creates

  • How the best teams design effective human + AI routines

  • What high-performance AI leaders are doing to set themselves apart

Enroll now to save your spot and win one of 10 free Glean:GO 2026 tickets.

AI RESEARCH

🎮 Kyutai’s AI dreams up a playable Rocket League

Image source: Kyutai and General Intuition

The Rundown: Kyutai and General Intuition just released MIRA, an open-source world model built with Epic Games that runs live 2v2 Rocket League for four players — generated entirely inside a neural network with no game engine underneath.

The details:

  • MIRA picked up the game from 10K hours of footage of AI bots playing each other, with zero human gameplay or player data anywhere in the training.

  • Despite not having a physics engine or graphics code, MIRA renders details like boost meters and crashes, while keeping all screens in sync on a single GPU.

  • Its memory only spans about four seconds, so when a goal replay rolls, MIRA hallucinates convincing footage of a goal that never actually happened.

  • The game runs at 20 frames per second on a single Nvidia GPU, with the teams open-sourcing the code, training data, and a playable demo.

Why it matters: We recently covered Odyssey’s Agora-1 world model generating a multiplayer version of Goldeneye, and now we have another model handling Rocket League. But as the companies note, the point isn’t to replace video games, but to simulate the physical world for a future full of robots that need data to train on.

QUICK HITS

🛠️ Trending AI Tools

  • 🚀 Hy3 - Tencent’s new open-source model for cheap, reliable agent deployment

  • 🍌 Nano Banana 2 Lite - Google's high-volume image model that generates pictures in just four seconds

  • 🔈 Seed Audio 1.0 - ByteDance's model for generating speech, music, and SFX in one pass

  • ⚙️ ZCode- Z AI’s agentic coding environment tuned for GLM-5.2

📰 Everything else in AI today

Elon Musk’s xAI officially rebranded to SpaceXAI, following the company’s February merger that valued the AI lab at $250B.

Illinois Gov. JB Pritzker signed the first U.S. state law forcing big AI developers to undergo annual third-party safety audits and reporting, with backing from both Anthropic and OpenAI.

OpenAI’s Tibo Sottiaux revealed that the company’s coming GPT 5.6 ‘Ultra’ variant will be available in Codex, with the model family expected to be released this week.

ByteDance and Alibaba are ending AI companion features in Doubao and Qwen, coming ahead of a July Beijing rule that targets chatbots and emotional dependency.

Sotheby’s is auctioning off the “Jensen Jacket,” an autographed leather jacket previously worn by Nvidia CEO Jensen Huang, with an estimated final bid of $40–60K.

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 Anonymous:

“I've been studying Spanish and am always looking for ways to help me remember new vocabulary and phrases. One fun method I've been using lately: I use MacWhisper to make a transcript of a Pimsleur language lesson from the MP3 file, drop that into Claude with instructions to craft the lyrics to a catchy Latin pop song using only the vocabulary in the transcript, and then paste those lyrics into Suno to make the song.

I now have a growing playlist of catchy Spanish-language songs that remind me of past lessons, help me with pronunciation, and prepare me for learning even more.”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

See you soon,

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

Robotics

This $8K robot can do your laundry

Jennifer Mossalgue • 6 minutes

Read Online | Sign Up | Advertise

Good morning, robotics enthusiasts. A San Francisco startup just started taking preorders for a robot that will fold your laundry, make your bed, and clear your clutter — for less than half the price of its humanoid rivals.

Weave Robotics’ Isaac 1 skips the sci-fi styling — wheels instead of legs, claws instead of fingers — and bets everything on nailing a short list of chores. But it inherits the catch shadowing home humanoids like 1X’s Neo: when the AI gets stuck, a stranger takes over remotely.

In today’s robotics rundown:

  • Meet Isaac 1, the $8K robot that does your chores

  • Boston Dynamics strips down Atlas to scale up

  • The race to make humanoids safe at work

  • Zoox’s robotaxi is designed for your worst behavior

  • Quick hits on other robotics news

LATEST DEVELOPMENTS

WEAVE ROBOTICS

🤖 Meet Isaac 1, the $8K robot that does your chores

Image source: Weave Robotics

The Rundown: San Francisco startup Weave Robotics just launched Isaac 1, a $7,999 wheeled home robot that folds laundry, clears clutter, and makes beds — with human teleoperators on standby for when the AI gets stuck.

The details:

  • Isaac 1 rolls on a wheeled base and adjusts up to 5'9" depending on the task, with two arms, fabric shells, and cameras that turn off when not in use.

  • Its two headline features are “Laundry Flow” — finding dirty clothes, handling hampers, and folding — and “Daily Reset,” which makes beds and tidies.

  • Pricing hits $7,999 upfront or $449/month, with a refundable $250 preorder — California deliveries begin this fall, with wider U.S. availability in 2027.

  • The bot is autonomous by default, with teleoperation assistance when needed, meaning a remote human can see inside your home through its cameras.

Why it matters: Sidestepping general-purpose ambitions, Weave undercuts humanoid rivals like 1X’s ~$20K Neo by more than half — a bet that purpose-built machines arrive in real living rooms before the humanoids do. But it still inherits the most privacy-concerning catch: strangers seeing through your robot’s eyes.

BOSTON DYNAMICS

🤖 Boston Dynamics strips down Atlas to scale it up

Image source: Images 2.0 / The Rundown

The Rundown: Hyundai-owned Boston Dynamics says its fifth-generation Atlas is “almost an order of magnitude” simpler than its predecessor — a redesign built to slash costs and clear the path to mass production, Forbes reports.

The details:

  • The report says that the new Atlas has far fewer parts, making it faster to build, more reliable, and cheaper than earlier versions, which ran upwards of $200K.

  • Hyundai aims to build a factory capable of churning out 30K units annually by 2028 and deploy Atlas at its Georgia plant starting in 2028.

  • Robot behavior director Alberto Rodriguez says that software is the bottleneck, with algorithms still unable to exploit what the machine can physically do.

  • The company is also arguing that two legs require almost the same 8 actuators as a wheeled base, while fitting through tighter spaces and crossing dock gaps.

Why it matters: Boston Dynamics is shifting Atlas to scale, with Hyundai’s factory muscle behind it. If the new Atlas can match or beat its predecessor with a fraction of the complexity, it could push high-end humanoids closer to real industrial economics — even as cheaper rivals like Unitree drag the market’s price expectations down.

HUMANOIDS

🏭 The race to make humanoids safe at work

Image source: Agility Robotics

The Rundown: Humanoid makers are scrambling to engineer safety into their large, heavy bipeds before they leave the lab for warehouse floors, where a fall or misstep could hurt the human working next to them, the WSJ reports.

The details:

  • Nvidia unveiled a Blackwell-powered safety system that reads sensor data for hazards and halts the robot when conditions turn dangerous.

  • Germany’s Neura Robotics built its 176 lb. 4NE1 to collapse inward on itself when a joint fails instead of pancaking whoever’s standing nearby.

  • Bay Area startup Dexmate packs batteries and electronics into a wheeled base, dropping the center of mass so low the robot physically can’t face-plant.

  • Regulators are years behind the hardware: an ISO expert panel is studying the fall risk now but doesn’t expect to publish a safety standard until mid-2028.

Why it matters: Morgan Stanley sees a billion humanoids and a $7.5T market by 2050 — but Agility’s robots, for one, work behind Plexiglas. No one has been seriously hurt by a humanoid yet, per the industry’s accounting — but one bad fall could mean broken bones, lawsuits, and a backlash that could stall the category before it even takes off.

ZOOX

🧽 Zoox’s robotaxi is designed for your worst behavior 

Image source: Zoox

The Rundown: After 500K+ driverless rides taught Zoox that unsupervised passengers will do pretty much anything in its driverless cabs, the Amazon-owned company rebuilt its next-gen robotaxi interior to be wiped clean between fares.

The details:

  • Zoox has been offering free rides in Las Vegas and San Francisco, and claims it has served 500K+ passengers — with no driver aboard to enforce the rules.

  • “They’re smoking everything in the vehicle,” said Chris Stoffel, Zoox’s director of robot industrial design and studio engineering.

  • Cleanup data — including vomit getting into “nooks and crannies” — pushed it toward odor-resistant materials that crews can wipe clean quickly.

  • The refreshed look keeps its signature layout — no steering wheel, seating for four — and arrives as the company preps public launches in Austin and Miami.

Why it matters: Remove the driver, and you remove the one person who could tell riders to buckle up, stub out the cigarette, or stay inside the car. The robotaxi industry is learning that passengers treat an empty driver’s seat as permission to do as they like. Zoox's answer: don’t police the behavior — build a cabin that can take the hits.

QUICK HITS

📰 Everything else in robotics today

NVIDIA’s GEAR team just unveiled SimFoundry, a framework that turns a single image or video of a real-world scene into a simulation-ready world for robot training.

A humanoid robot named Joko reportedly malfunctioned on its first day at an Indonesian office, unleashing kung fu kicks at workers who struggled to shut it down.

China’s UBTech launched the U1, a line of ultra-lifelike companion humanoids with an “emotional” LLM the company claims can read over 20 human emotions.

Tesla’s robotaxis are now picking up riders in Miami, the company’s first fully driverless market beyond Texas, covering a 10- to 14-square-mile zone.

China’s securities regulator cleared Unitree for a Shanghai IPO, greenlighting the humanoid maker to raise 4.2B yuan (~$618M) at an implied $6.2B valuation.

German defense startup Quantum Systems raised a $1.2B Series D co-led by Blackstone, Airbus, Advent, and Noteus, valuing the autonomous drone maker at ~$8B.

A new generation of underwater mine-clearing robots, like those from Greensea IQ, are taking human divers out of one of the military’s most dangerous jobs.

Pudu Robotics is partnering with Shenzhen’s tourism authority to open what it calls the first fully robot-serviced hotel in 2027.

Ant Group just led a 500M yuan ($73.6M) round in Chinese companion-robot startup Zeroth, its 12th investment in the robotics sector in 18 months, per CNBC.

COMMUNITY

🎓 Highlights: News, Guides & Events

See you soon,

Rowan, Zach, Shubham, and Jennifer — The Rundown’s editorial team

No matching search results

Try using different keywords, double-check your spelling, or explore related categories.

Clear Search

Stay Ahead on AI.

Join 2,000,000+ readers getting bite-size AI news updates straight to their inbox every morning with The Rundown AI newsletter. It's 100% free.

Subscribe