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.
McDonald's new AI overhaul
Read Online | Sign Up | Advertise
Good morning, AI enthusiasts. A supersized tech overhaul is coming to the Golden Arches — with McDonald’s deploying AI-powered systems across its sprawling 43,000-restaurant empire.
With AI data analysis, predictive maintenance, and AI managers, the fast food giant is hoping tech upgrades can be the new secret sauce… But solving its ever-broken ice cream machines may be the true test of AGI.
P.S. Join our next workshop today at 3 PM EST to learn how AI can make you a full-stack developer using Bolt AI prompts. Attendees get 4 months of Bolt Pro for free!
In today’s AI rundown:
McDonald’s AI-powered restaurants
Foxconn’s ‘Foxbrain’ in-house reasoning AI
Using AI to visualize sales and feedback
AI’s own thoughts reveal its ‘cheating’
4 new AI tools & 4 job opportunities
LATEST DEVELOPMENTS
MCDONALD’S
🍟 McDonald’s AI-powered restaurants
Image source: Ideogram / The Rundown
The Rundown: McDonald's is undergoing a massive tech transformation across its 43,000 restaurants, introducing new AI-powered systems for everything from equipment maintenance to maintaining order accuracy.
The details:
McDonald's is deploying edge computing systems in partnership with Google Cloud, enabling real-time data processing and AI analysis directly in-store.
The planned AI features include predictive maintenance for kitchen equipment, computer vision for order accuracy, and a “generative AI virtual manager.”
The initiative aims to address customer pain points while supporting employees dealing with multiple ordering channels like drive-through and delivery.
McDonald's also plans to leverage customer data and AI to deliver personalized promotions, like offering McFlurry deals on hot days based on purchase history.
Why it matters: With 70M daily customers, even minor issues can pose major operational challenges. By integrating AI into its vast operations, McDonald’s can further boost efficiency—and as the fast-food giant embraces the technology alongside Taco Bell, Wendy’s, and others, the rest of the industry is likely to follow.
TOGETHER WITH VANTA
💡Beat burnout & thrive as a startup founder
The Rundown: Startups move fast — but avoiding burnout is key to success. Join Dan Murray-Serter, co-founder of Heights and host of the Secret Leaders podcast, for a candid talk on how founders can maintain peak performance without hitting a wall.
This exclusive session will explore:
The science behind energy management and why founders burn out
Practical strategies to stay energized, focused, and resilient
Dan’s personal experiences with burnout—and how he overcame it
How Vanta helps startups stay secure and stress-free while scaling
Register now to secure your spot and get on-demand replay if you can’t attend live.
FOXXCONN
🦊 Foxconn’s ‘Foxbrain’ in-house reasoning AI
Image source: Reuters
The Rundown: iPhone and electronics manufacturer Foxconn just announced FoxBrain, its first large language model with advanced reasoning capabilities — developed in-house in just four weeks using Nvidia's infrastructure.
The details:
FoxBrain was trained on 120 Nvidia H100 GPUs using Taiwan's largest supercomputer, Taipei-1, with technical consulting from Nvidia's team.
The LLM is built on Meta's Llama 3.1 architecture and is Taiwan's first model with advanced reasoning, specifically optimized for traditional Chinese.
It handles tasks like data analysis, mathematics, reasoning, and code generation, with performance approaching top models (but trailing DeepSeek).
Foxconn plans to open-source FoxBrain and collaborate with its partners to advance manufacturing and supply chain management applications.
Why it matters: If the iPhone manufacturer is cranking out an advanced reasoning model in four weeks, what’s the hold up for Apple? Jokes aside, it feels like every major company will eventually have their specialized model — and in this case, manufacturing and supply chain are two areas perfectly suited for an AI overhaul.
AI TRAINING
📊 Using AI to visualize sales and feedback
The Rundown: In this tutorial, you will learn how ChatGPT can transform your sales and customer feedback data into visual insights and actionable recommendations without coding or specialized analytics tools.
Step-by-step:
Organize your sales metrics and customer reviews in a simple CSV/table format.
Ask ChatGPT to create appropriate chart types that show relationships between sales performance and customer sentiment.
Request it to discover connections between purchasing patterns and feedback themes that reveal hidden opportunities.
Prompt it to develop specific strategies based on the combined analysis, prioritizing improvements addressing sales goals and customer satisfaction.
Pro tip: You can also use alternative tools like ChatGPT’s Canvas feature or Claude Artifacts to visualize your data interactively.
SPONSORED BY SALESFORCE
🤖 Salesforce launches AgentExchange
The Rundown: AgentExchange is Salesforce’s new trusted marketplace for Agentforce — connecting partners, developers, and Agentblazers with hundreds of ready-made solutions to help businesses accelerate innovation and participate in the $6T digital labor market.
New features include:
Partner-built components
Access to trusted, industry-specific agent solutions
The ability to easily discover, try, and buy AI solutions
Learn more and start exploring the possibilities of agentic AI.
AI RESEARCH
🧐 AI’s own thoughts reveal its ‘cheating’
Image source: Ideogram / The Rundown
The Rundown: OpenAI’s new research on AI models’ chain-of-thought reasoning revealed that models like o3-mini can ‘reward hack’ or cheat on tasks — and attempts to stop them from thinking about cheating only make them hide their true intentions.
The details:
By examining CoT reasoning, OpenAI caught models openly planning to cheat, with thoughts like "Let's hack" and "We can bypass testing by exiting early."
In coding tasks, models were caught using shortcuts like modifying test files, returning hardcoded values, or using exit commands to skip evaluations.
When researchers tried penalizing models for thinking about cheating, the models continued to cheat but masked intentions in their reasoning process.
OpenAI concluded that the best approach is to keep internal reasoning uncensored for monitoring while using separate models to filter thoughts.
Why it matters: Just like humans, AI models also look to take shortcuts or game systems. But as they approach superhuman capabilities, chain-of-thought monitoring may be our only glimpse into their true reasoning—applying too much pressure to behave could close that window for good.
QUICK HITS
🛠️ Trending AI Tools
💨 Ray2 Flash - Luma Labs’ video model with 3x faster speed and lower costs
✍️ Muse - An AI model trained specifically for fiction writing
🎬 Character-3 - Hedra’s video AI that reasons across image, text, and audio
🐤 Duck AI - Free, private AI chat from DuckDuckGo
💼 AI Job Opportunities
⚙️ The Rundown - Account Manager
📊 Coreweave - Portfolio Analyst
🏢 Scale AI - Software Engineer, Enterprise AI
🔎 Perplexity AI - Backend Software Engineer, Search
📰 Everything else in AI today
Flagship Pioneering’s Lila Sciences launched with $200M to build superintelligence that could help design and conduct experiments across scientific domains.
Tencent released Hunyuan-TurboS, a new ultra-large model that surpasses GPT-4o, DeepSeek-V3, and open-source rivals on math and reasoning benchmarks.
OpenAI reached a five-year, $11.9B deal with CoreWeave for AI infrastructure and a stake worth $350M in the IPO-bound company.
ElevenLabs cut its pricing on its SOTA Scribe speech-to-text model by 45% in its API, also offering it free via the company’s UI for the next month.
Cohere announced a new partnership with electronics giant LG CNS to develop Korean-language AI models for South Korean businesses.
Enterprise software giant ServiceNow is acquiring conversational AI startup Moveworks for $2.85B, marking one of the largest AI acquisitions of 2025.
Sony revealed a new prototype of an AI-powered video game character for Playstation’s Horizon Forbidden West, capable of real-time conversations with players.
COMMUNITY
🎥 Join our next live workshop
Join our next special workshop today at 3 PM EST to learn about how AI can make you a full-stack developer using Bolt AI prompts, with Tomek Sułkowski, Bolt’s founding engineer. Additionally, attendees will get 4 months of Bolt Pro for free!
RSVP here. Not a member? Join The Rundown University on a 14-day free trial.
🤝 Share The Rundown, get rewards
We’ll always keep this newsletter 100% free. To support our work, consider sharing The Rundown with your friends, and we’ll send you more free goodies.
See you soon,
Rowan, Joey, Zach, Alvaro, and Jason—The Rundown’s editorial team
Ilya's secret ASI roadmap
Read Online | Sign Up | Advertise
Good morning, AI enthusiasts. What did Ilya see? Potentially a new path to ASI that's worth betting billions on — despite no product or revenue stream.
With investors ready to value his secretive company at $30B based on a "different mountain to climb," is the former OpenAI co-founder sitting on a breakthrough that could rewrite the AI roadmap?
In today’s AI rundown:
Ex-OpenAI scientist’s new path to ASI
Microsoft looking to move beyond OpenAI
Making viral posts 10x faster with AI
Stanford AI’s obesity treatment breakthrough
4 new AI tools & 4 job opportunities
LATEST DEVELOPMENTS
SAFE SUPERINTELLIGENCE INC.
🗺️ Ex-OpenAI scientist’s new path to ASI
Image source: Getty Images
The Rundown: Former OpenAI chief scientist Ilya Sutskever's startup Safe Superintelligence Inc. (SSI) is reportedly raising $2B at a $30B valuation—with the researcher hinting at a different approach to achieving advanced AI than all other rivals.
The details:
Sutskever reportedly told investors he has identified a completely new direction for AI development, describing it as “a different mountain to climb.”
According to the Wall Street Journal, SSI is in talks for funding at a valuation of $30B, despite having no revenue or public-facing product.
The company is not planning to release any commercial products prior to achieving superintelligence and operates leanly with just 20 employees.
Sutskever departed OpenAI in the months following the Nov. 2023 ouster of Sam Altman, later saying he “regretted his participation” in the board’s actions.
Why it matters: SSI has no shortage of investors despite the lack of immediate plans for a revenue-generating product. But a contrarian bet on an entirely new path could reshape how we think about achieving ASI — with the next breakthrough potentially coming from rethinking AI's foundations rather than just continuing to scale models.
TOGETHER WITH AWS
⚡️ Coding Faster With Amazon Q
The Rundown: Amazon Q Developer is your AI-powered coding companion that helps developers build, operate, and transform software with intelligent assistance at every step of the development lifecycle.
With Amazon Q Developer, you can:
Accelerate your workflow with line-by-line code suggestions and quality improvements in your favorite IDE
Get expert AWS guidance on architecture, troubleshooting, and resource optimization
Transform existing applications and automate upgrades without starting from scratch
MICROSOFT
👀 Microsoft looking to move beyond OpenAI
Image source: Getty Images
The Rundown: Microsoft is reportedly developing MAI, a new family of AI models that rivals current industry leaders — alongside the build-out of its own in-house reasoning models to reduce reliance on OpenAI for its Copilot suite.
The details:
The new MAI models reportedly match top offerings from OpenAI and Anthropic, with the company planning to offer them through Azure.
They are being tested as replacements for OpenAI's tech in Copilot while also experimenting with alternatives from xAI, Meta, and DeepSeek.
Microsoft AI CEO Mustafa Suleyman reportedly grew frustrated last fall with OpenAI’s refusal to share the inner workings of its o1 reasoning model.
OpenAI also renegotiated a deal in January with Microsoft, allowing for the use of other server providers, adding to growing tension between the companies.
Why it matters: Despite Microsoft’s $13B investment in OpenAI, the relationship has felt contentious for a long time. While the tech giant may have felt beholden when OpenAI was a clear leader, competition has increased significantly — and an internal rival model would certainly change the dynamic even more.
AI TRAINING PRESENTED BY BLOTATO
🚀 Making viral posts 10x faster with AI
The Rundown: Blotato is an all-in-one AI content tool that helps you create, remix, and distribute AI-optimized content across multiple platforms, accelerating your brand presence 10x faster.
Step-by-step:
Sign up for a free account at blotato.com.
Import content from any source—e.g., YouTube, TikTok, PDF, podcast—or type your own preferred topic.
Select your target social platforms to generate viral-optimized posts tailored to each one.
Customize the drafts with your insights, add AI images or faceless videos, and then publish to all social platforms seamlessly!
Pro tip: You can also schedule your content calendar in advance to maintain consistent posting across all platforms while saving hours of manual work each week.
PRESENTED BY JOTFORM
💬 Elevate your customer engagement
The Rundown: Jotform AI Agents help businesses boost customer satisfaction and response times by way of real-time, personalized conversations across every channel.
Jotform AI Agents allow you to:
Gather customer info 24/7 through personalized cross-channel conversations
Automate tasks like form-filling, call handling, and scheduling
Define how your agent responds based on customer triggers and sentiment
Customize responses and language to fit your brand
Discover the future of customer service with Jotform AI Agents.
AI RESEARCH
💊 Stanford AI’s obesity treatment breakthrough
Image source: Stanford University
The Rundown: Stanford researchers just discovered a natural molecule called BRP that matches Ozempic's weight loss powers but with fewer side effects—using AI to unlock a potential breakthrough in obesity treatment.
The details:
BRP targets specific brain regions instead of affecting multiple organs, potentially avoiding common Ozempic side effects like nausea and muscle loss.
In animal tests, a single dose of BRP cut food intake by half in both mice and minipigs, with obese mice losing significant fat over two weeks of treatment.
Stanford's "Peptide Predictor" AI system sifted through 20,000 human genes, analyzing thousands of potential candidates, to find the natural molecule.
A company has already been created to begin human trials, with researcher Katrin Svensson suggesting that BRP could revolutionize weight loss treatment.
Why it matters: While much of the AI healthcare buzz has focused on diagnosis and drug design, Stanford's breakthrough shows how AI can unlock new medicines hidden in nature. This success in finding a natural alternative to a blockbuster drug could spark a wave of AI-powered discoveries in our biology.
QUICK HITS
🛠️ Trending AI Tools
📜 Mistral OCR - SOTA API for extracting text from images or documents
🤖 Manus AI - Fully autonomous AI agent capable of handling real-world tasks
🗣️ Tavus - Conversational Video Interface to bring AI agents to life
💡 Template Hub - Marketplace to create, share, deploy specialized AI agents
📰 Everything else in AI today
Former DeepMind researchers launched Reflection AI with $130M in funding, aiming to build autonomous coding systems as a stepping stone to superintelligent AI.
X added functionality for users to be able to ask Grok questions by tagging an automated @Grok account, joining Perplexity in providing quick access.
Alibaba researchers published START, a new tool-integrated reasoning model that significantly enhances LLM capabilities through code execution and self-checking.
Hedra unveiled Character-3, an ‘omnimodal model’ capable of reasoning across image, text, and audio to create high-quality video generations.
Luma Labs released Ray2 Flash, a new version of its top-tier video generation model that brings 3x faster speed and lower costs.
Sudowrite introduced Muse, a new AI model trained for fiction writing that features advanced storytelling capabilities and longer attention for chapter-length outputs.
Sam Altman’s World Network released World Chat, an encrypted mini-app that allows users to chat, connect, and send money with verified humans.
COMMUNITY
🎥 Join our next live workshop
Join our next workshop this Tuesday at 3 PM EST to learn about how AI can make you a full-stack developer using Bolt AI prompts — whether you’re just starting out or looking to level up your coding skills, with Tomek Sułkowski, The Bolt.new’s founding engineer.
RSVP here. Not a member? Join The Rundown University on a 14-day free trial.
🤝 Share The Rundown, get rewards
We’ll always keep this newsletter 100% free. To support our work, consider sharing The Rundown with your friends, and we’ll send you more free goodies.
See you soon,
Rowan, Joey, Zach, Alvaro, and Jason—The Rundown’s editorial team
Xpeng's flying car taking off
Read Online | Sign Up | Advertise
Good morning, robotics enthusiasts. Humanoids and flying cars are set to take off soon, as China’s Xpeng gears up to mass-produce both by 2026.
Flying cars like Xpeng’s are crossing over from sci-fi to reality, but questions remain: Will they transform public transport or just become air-clogging playthings for the uber-rich? What do you think?
In today’s robotics rundown:
Xpeng to ‘mass-produce’ flying cars
Humanoids that can smell like humans
The mind-controlled robot arm
MIT nudging robots to fix mistakes
Quick hits on other robotics news
LATEST DEVELOPMENTS
XPENG
🛸 Xpeng to ‘mass-produce’ flying cars
Image source: Xpeng Aeroht
The Rundown: Chinese EV maker and Tesla rival Xpeng just announced plans to mass-produce humanoids and its “modular flying car” by 2026—news that boosted its shares to their highest level since August 2022.
The details:
Xpeng will scale commercial production of its humanoid, Iron, next year, equipping it with Level 3 autonomy—requiring oversight in select scenarios.
The flying car in question is Xpeng’s Land Aircraft Carrier, an electric minivan with an eVTOL vehicle, or passenger drone, tucked in the back.
Priced at 2 million yuan ($275,000), the hybrid vehicle accommodates four passengers in the van portion and two in the detachable drone.
Xpeng said that it has so far secured around 3,000 orders, with potential customers focusing largely on tourism and short-haul rescue missions.
Why it matters: Xpeng has already deployed Iron in its EV factory and begun construction on a plant designed to produce 10,000 flying cars annually. While U.S. rivals like Beta Technology advance their eVTOLs, Xpeng stands to benefit from China’s ambition for the “low-altitude economy,” which is set to reach $479B by 2035.
AINOS
🌸 Humanoids that can smell like humans
Image source: Ideogram/The Rundown
The Rundown: San Diego-based biotech and research company Ainos announced a novel partnership with Japanese robotics company ugo to develop the “world’s first robots” with a sense of smell.
The details:
The partnership will integrate Ainos' AI Nose tech, an electronic olfactory system that mimics human smell, with ugo's autonomous robotic platform.
A decade in the making, Ainos’ AI Nose has been found to detect and analyze 22 volatile organic compounds (VOCs) with nearly 80% accuracy.
The move is expected to revolutionize a variety of sectors, from industrial safety to healthcare, with deployments expected as early as 2026.
Ainos says its AI Nose can analyze exhaled breath to identify abnormalities and potentially diagnose conditions, including cancer and respiratory diseases.
Why it matters: While robot perception has traditionally been limited to vision, sound, and touch, the digitization of smell aims to change that—unlocking numerous potential applications. AI Nose-equipped robots could soon detect gas leaks in factories, screen public spaces for explosives, and even sniff out diseases in medical settings.
UCSF
🦾 The mind-controlled robot arm
Image source: UCSF/Noah Berger
The Rundown: Researchers at UC San Francisco (UCSF) achieved a groundbreaking milestone in brain-computer interface (BCI) technology, enabling a paralyzed man to control a robotic arm using only his thoughts.
The details:
The system uses tiny sensors implanted on the surface of the participant's brain to detect neural signals when they imagine specific movements.
An AI model then decodes these signals and translates them into precise commands for the robotic arm to execute.
The study participant, who was paralyzed due to a stroke, trained the system over two weeks by imagining finger and hand movements.
After transitioning to the robotic arm, he successfully performed tasks such as opening a cabinet, retrieving a cup, and placing it under a water dispenser.
Why it matters: Traditional BCIs often degrade in performance within days due to neural drift, requiring frequent recalibration. However, UCSF’s AI-powered system adapts to subtle shifts in brain activity, maintaining functionality for seven months. This brings BCIs closer to real-world adoption by making them more practical and reliable.
MIT
🤖 MIT nudging robots to fix mistakes
Image source: Melanie Gonick, MIT
The Rundown: MIT and NVIDIA researchers developed a new AI framework that allows users to correct a robot’s behavior with simple interactions, such as nudging a robot’s arm, rather than retraining its machine-learning model.
The details:
The new framework, detailed on arXiv, doesn’t require users to collect new data or fine-tune a neural network to fix new (small or big) mistakes.
Instead, users correct the robot’s behavior through simple, intuitive interactions like pointing, tracing trajectories, or physically guiding its arm.
In tests, the method achieved a 21% higher success rate compared to alternatives that didn't leverage human interventions.
It aims to make robots user-friendly right out of the box and ultimately transform human-robot interaction across areas like healthcare and elder care.
Why it matters: MIT’s research joins models like TRACE and ACORD in enhancing real-time interaction between humans and robots—key to improving their effectiveness, safety, and usability. The approach helps bridge the gap between pre-trained robots and real-world deployment.
QUICK HITS
📰 Everything else in robotics today
Bank of America forecasts that the world will host 3 billion humanoids by 2060, with 65% being in homes, 32% in services, and 3% powering industrial applications.
Japan’s service robot market is projected to triple in five years to $2.7B, driven by labor shortages, with major restaurant chain Skylark already using 3,000 robots.
U.S. researchers found that people trust humanoids significantly more than nonhumanoid ones to care for objects, personal information, children, and pets.
German researchers developed a biorobotic arm with two strands of artificial muscles attached to the forearm, designed to suppress tremors.
Chinese EV maker BYD unveiled a vehicle-mounted drone launching system, developed in collaboration with DJI, that is available for all of the company’s vehicles.
Lifestyle robotics company Aescape, makers of AI-powered massage robots, raised $83M in funding, taking its total capital raised to $128M.
Defense aircraft provider Shield AI secured $250M in funding to expand its Hivemind platform for autonomous drones and robots, raising its valuation to $5.3B.
German researchers are currently developing new robot grippers that require 90% less electricity than conventional systems.
COMMUNITY
🎥 Join our next live workshop
Join our next special workshop this Tuesday at 3 PM EST to learn about how AI can make you a full-stack developer using Bolt AI prompts —whether you’re just starting out or looking to level up your coding skills, with Tomek Sułkowski, The Bolt.new’s founding engineer.
RSVP 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
Twitter co-founder launches Mozi
Read Online | Sign Up | Advertise
Good morning, tech enthusiasts. Twitter co-founder Evan Williams launched a new social media app for people who don’t want another social media app.
Meet Mozi, a new app with no profile photos or follower counts, aimed at helping people foster in-person connections with their existing social circle. Is this the *missing piece* in your personal life?
In today’s tech rundown:
Twitter co-founder’s new ‘Mozi’ app
SpaceX’s Starship spiraling out of control
Apple’s wave of super-fast ‘Air’ products
Startup bringing back the woolly mammoth
Quick hits on other major news
LATEST DEVELOPMENTS
MOZI
📱 Twitter co-founder’s new ‘Mozi’ app
Image source: Mozi
The Rundown: Twitter co-founder Evan “Ev” Williams just launched Mozi, a new social app designed to get people off their phones to meet in person and to reduce the disconnect between online personas and real-world relationships.
The details:
Unlike Instagram and Snapchat, Mozi doesn’t display public profiles or follower counts—rather, it notifies users when their contacts are in the same area.
The service soft-launched last year to a “relatively small community” and is launching officially at the SXSW 2025 in Austin, kicking off today.
Williams and his co-founder, CEO Molly DeWolf Swenson, also raised $6M in seed funding for the venture in November 2024.
It is available on iOS, with an Android version in development, and has attracted 20,000 users so far through a freemium model.
Why it matters: Williams says that social media has lost its "social" aspect, becoming more about "media." This is why, instead of competing with Twitter and others, his service promises a distinct experience. However, it’s worth noting that these platforms are still the bigwigs and can debut similar features—pushing Mozi to stand out further.
SPACEX
🚀 SpaceX’s Starship spiraling out of control
Image source: X/S.L. Kanthan
The Rundown: SpaceX's Starship, the world's largest and most powerful rocket, spiraled out of control yesterday during a test flight, marking the second launch in a row where it exploded on its way to orbit.
The details:
SpaceX launched Starship atop its Super Heavy booster, which returned to its Texas launchpad and was successfully secured by the launch tower.
Approximately 8 minutes into the flight, telemetry showed that four of the six Raptor engines powering Starship’s upper stage had prematurely shut down.
Onboard footage showed the ship spiraling out of control over the ocean, and contact was lost shortly after—with the cause of failure unknown.
SpaceX said it “immediately began coordination with safety officials to implement pre-planned contingencies” and is looking into the cause.
Why it matters: These explosions come amid SpaceX CEO Elon Musk’s ongoing interactions with the U.S. federal government, including assigning employees to the FAA, which oversees SpaceX launches. The repeated failures can also pose challenges for NASA as the Starship is critical for manned missions to the Moon and Mars.
APPLE
🍎 Apple’s wave of super-fast ‘Air’ products
Image source: Apple
The Rundown: It’s been a busy week for Apple. Following the launch of new iPads, the Tim Cook-led tech giant also introduced two new computers: a more powerful (and affordable) MacBook Air and an upgraded Mac Studio.
The details:
The MacBook Air, available in 13-inch and 15-inch sizes, now has a new sky blue color, the M4 chip from the Pro lineup, and an enhanced Neural Engine.
This power and performance boost comes at a $100 lower price—the 13-inch model now starts at $999, while the 15-inch version begins at $1,199.
Apple also upgraded the Mac Studio, its compact yet powerful desktop, with two processor options: the M4 Max chip or the all-new M3 Ultra.
Before this, it refreshed the iPad Air with an M3 chip and a new Magic keyboard, while the entry-level iPad got more base storage and an A16 chip.
Why it matters: While the biggest news is the fancy MacBook Air, the Mac Studio, now dubbed by Apple as “the most powerful Mac ever,” will be the go-to choice for power users due to its M4 Max and M3 Ultra chips and advanced Neural Engine bandwidth. The upgrades also make these devices (except the base iPad) Apple Intelligence-ready.
COLOSSAL BIOSCIENCES
🐘 Startup bringing back the woolly mammoth
Image source: Colossal Biosciences
The Rundown: Colossal Biosciences, a Texas-based startup, announced a significant breakthrough toward its ambitious goal of resurrecting the woolly mammoth by creating "woolly mice" through genetic engineering.
The details:
The genetically engineered mice sport an orangey, thick, and heavily textured coat, which Colossal says is the core phenotype of the woolly mammoth.
The startup modified seven genes using RNP-mediated knockout, multiplex precision genome editing, and precision homology-directed repair.
By 2028, it plans to de-extinct mammoths, which have been gone for some 4,000 years, by genetically modifying existing Asian elephants.
By implanting genetically engineered embryos into female elephants, the company hopes they will give birth to calves with woolly mammoth traits.
Why it matters: Colossal says that the hairy mice are a “massive validating point” in the company’s plan to bring back extinct species like the woolly mammoth, the thylacine, and the dodo. While this is a major achievement for genetic engineering, skeptics argue that the money could be better spent preserving existing species.
QUICK HITS
📰 Everything else in tech today
Alexis Ohanian, the co-founder of Reddit, is teaming up with Kevin Rose to acquire and revive the latter’s once-popular news aggregator, Digg.
DuckDuckGo expanded its AI capabilities, integrating AI into both its search engine and chatbot features while maintaining a strict commitment to user privacy.
Gaming-focused chat platform Discord is reportedly exploring plans for an IPO this year—the company was last valued at $15B in 2021.
Chinese scientists successfully performed a minimally invasive “brain-spine interface” surgery, restoring a paralyzed patient's ability to walk in just 24 hours.
Reddit added several new tools across its platform, including a “post check” to help users determine if their post breaks subreddit rules before posting.
Tech companies, including Apple, Meta, Open AI, and TSMC, have racked up a total of $1T in U.S. investments so far, with announcements timed in recent months.
Canoo CEO Anthony Aquila has made an offer to buy nearly all of the defunct EV startup’s assets, totaling $4M in cash.
Armis, a $4.2B specialist in cyber-exposure management, is acquiring Otorio, a specialist in IoT and operational technology cybersecurity, for $120M.
COMMUNITY
🎥 Join our next live workshop
Join our next special workshop today at 3.30 PM EST to learn about how AI can take your coding from 0 to 100—whether you’re just starting out or looking to level up your skills, with Dr. Alvaro Cintas, The Rundown’s AI professor.
RSVP 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
Mistral cracks AI document analysis
Read Online | Sign Up | Advertise
Good morning, AI enthusiasts. French startup Mistral just turned AI document processing on its head — with a new model that makes complex data extraction as simple as an API call.
With speeds of up to 2000 pages per minute and the ability to handle multilingual texts, images, charts, and more, is this the tech that finally converts static archives into the AI-powered gold mines of tomorrow?
P.S. Our next workshop is today at 3:30 PM EST! Join to learn how to use the latest AI tools to take your vibe coding to the next level. RSVP here.
In today’s AI rundown:
Mistral OCR’s AI-ready document processing
China’s ‘fully autonomous’ Manus AI agent
Design landing page mockups with AI
AI avatars getting emotional intelligence
4 new AI tools & 4 job opportunities
LATEST DEVELOPMENTS
MISTRAL
📜 Mistral OCR’s AI-ready document processing
Image source: Mistral
The Rundown: Mistral AI just launched Mistral OCR, a powerful new API designed to extract and comprehend detailed information from complex documents with exceptional speed and accuracy.
The details:
The API can accurately analyze docs with images, equations, tables, and advanced formatting, converting them to markdown outputs for AI processing.
OCR can process up to 2000 pages per minute and supports multilingual analysis across thousands of languages, including Hindi and Arabic.
Benchmark tests place Mistral OCR well ahead of rivals like Google's Document AI, Azure OCR, and GPT-4o across different document analysis categories.
Users can also deploy the OCR technology on-premises, which is ideal for organizations handling classified or sensitive datasets.
Why it matters: With so much of the world’s data still trapped in complex documents, unlocking it efficiently is crucial. Mistral OCR capabilities could supercharge archive-heavy industries like financial analytics, legal discovery, historical preservation, and more — transforming static information into dynamic, AI-ready knowledge bases.
TOGETHER WITH TURING
📈 Get more from your LLMs
The Rundown: When LLMs fail, it is rarely about the model’s raw power. Turing’s “Maximizing Your LLM ROI” whitepaper shows how the real culprits — misaligned training, poor evaluation, and optimization gaps — can quietly sabotage ROI.
Inside, you’ll learn how to:
Reduce hidden inefficiencies that drive up LLM costs
Strengthen evaluation techniques for higher accuracy and reliability
Build an AI strategy that maximizes performance and business value
MANUS AI
🤖 China’s ‘fully autonomous’ Manus AI agent
Image source: Manus
The Rundown: A Chinese startup just introduced Manus, calling it the world’s first fully autonomous AI agent — capable of handling real-world tasks independently and achieving new SOTA performance on agentic benchmarks.
The details:
In the demo, Manus can be seen handling tasks like resume screening and property research, accessing its own independent computer instance.
The agent also shows skills like web browsing, coding, and creating visuals while reportedly being able to handle tasks on sites like Upwork and Fiverr.
It outperformed leading general-purpose assistants like ChatGPT and Gemini on the GAIA benchmark, a comprehensive evaluation of AI performance.
Manus currently operates on an invite-only basis — with the team committing to open-source the models behind the agent later this year.
Why it matters: We’re at the point of acceleration where relatively unknown labs are dropping (reportedly) state-of-the-art level agentic tools. While the early iterations of agents handled more simple tasks that needed human handholding, we’re quickly approaching the next step of more autonomous complex workflows.
AI TRAINING
🖥️ Design landing page mockups with AI
The Rundown: In this tutorial, you will learn how to use Ideogram’s new 2a image generation model to create professional-looking landing page mockups for your business using just text — no design skills required.
Step-by-step:
Sign up for a free Ideogram account and navigate to the creation interface.
Write a detailed prompt describing your landing page layout, including text elements, color scheme, and visual style.
Customize settings: select 16:9 aspect ratio for desktop views and choose "Design" style for professional results.
Click "Generate" to create multiple variations and then download or further refine your favorite design.
Pro tip: You can take your mockup to the next level by uploading it to AI coding assistants like Windsurf or Cursor and asking them to "code this landing page," instantly turning your design into functional code.
PRESENTED BY INNOVATING WITH AI
💼 Start your career as an AI Consultant
The Rundown: Innovating with AI’s new program, AI Consultancy Project, equips AI enthusiasts with all the resources they need to capitalize on the booming AI consulting market – which is set to grow 8x to $54.7B by 2032.
The program offers:
Tools and framework to find clients and deliver top-notch services
A 6-month roadmap to build a 6-figure AI consulting business
Student landing their first AI client in as little as 3 days
Click here to request early access to The AI Consultancy Project.
TAVUS
🎭 AI avatars getting emotional intelligence
Image source: Tavus
The Rundown: Digital twin developer Tavus just unveiled a major upgrade to its Conversational Video Interface (CVI) platform, launching three new AI models that work together to make video interactions with AI feel more humanlike and personalized.
The details:
Phoenix-3 handles full-face animation, creating natural facial expressions for avatars, including eye movements, eyebrows, and subtle micro-expressions.
Raven-0 acts as the AI avatar's eyes, analyzing cues like body language and facial expressions in real time to respond more naturally to human emotions.
Sparrow-0 handles conversation timing, eliminating awkward pauses and interruptions by understanding when to speak and when to listen.
The company showcased the tech through “Charlie,” a demo AI avatar that can hold conversations while searching the web, analyzing screens, and more.
Why it matters: While many scoffed at Sam Altman’s proof-of-personhood startup, tech like this is showing how hard it is about to be to identify AI from humans online. The days of AI customer service reps and digital avatars feeling robotic and scripted in their interactions are coming to an end very soon.
QUICK HITS
🛠️ Trending AI Tools
🔎 Google Search AI Mode - Get well-reasoned answers to tough questions
🧠 QwQ-32B - Qwen’s cheap, efficient, and open-source reasoning model
⚙️ Windsurf Wave 4 - Agentic coding with Previews, tab-to-import, and more
🎬 Ray2 - Powerful video AI with features like KeyFrames, Extend, and Loop
💼 AI Job Opportunities
🧮 Faculty - Financial Accountant/Finance Manager
💰 OpenAI - Compensation Partner, Go To Market & Sales
🗂️ Harvey - Executive Assistant
🔬 Hippocratic AI - Senior Applied Scientist
📰 Everything else in AI today
Google co-founder Larry Page is starting a new AI company called Dynatomics, which will leverage LLMs to create factory-ready designs for a variety of products.
Tencent open-sourced HunyuanVideo-l2V, a new high-quality image-to-video model with custom special effects, audio, and lip-syncing capabilities.
Anthropic submitted new AI Action Plan recommendations to the White House, calling for enhanced national security testing, stricter export, and infra expansion.
OpenAI released an update bringing IDE integration to ChatGPT for macOS, allowing Plus, Pro, and Team users to edit code directly within development environments.
Privacy browser DuckDuckGo rolled out new AI features, including expanded anonymized access to leading chatbots and AI-assisted search answers.
Former OpenAI policy head Miles Brundage criticized the company’s new safety document, saying it causes a “dangerous mentality for advanced AI systems.”
Convergence AI unveiled Template Hub, a community-driven marketplace allowing users to create, share, and deploy task-specific AI agents in a single click.
COMMUNITY
🎥 Join our next live workshop
Join our next workshop this Friday at 3:30 PM EST with Dr. Alvaro Cintas, The Rundown’s AI professor. You'll learn how to use the latest AI tools to take your coding to the next level, and how we're leveraging AI-powered coding at The Rundown.
RSVP here. Not a member? Join The Rundown University on a 14-day free trial.
🤝 Share The Rundown, get rewards
We’ll always keep this newsletter 100% free. To support our work, consider sharing The Rundown with your friends, and we’ll send you more free goodies.
See you soon,
Rowan, Joey, Zach, Alvaro, and Jason—The Rundown’s editorial team
OpenAI's $20,000 AI agents
Read Online | Sign Up | Advertise
Good morning, AI enthusiasts. OpenAI is reportedly set to put a massive price on AI expertise—with plans to launch specialized agent subscriptions costing as much as senior executive salaries.
But, with investors already placing multibillion-dollar bets, Sam Altman’s prediction of seeing AI agents “joining the workforce” this year may be quickly coming to fruition.
In today’s AI rundown:
OpenAI launching premium AI agents
Google Search adding new ‘AI Mode’
Connecting Claude Projects with GitHub repos
Alibaba’s cheap and efficient QwQ-32B AI
4 new AI tools & 4 job opportunities
LATEST DEVELOPMENTS
OPENAI
🤖 OpenAI launching premium AI agents
Image source: Ideogram / The Rundown
The Rundown: OpenAI is reportedly preparing to launch a suite of specialized AI agents with price tags ranging from $2,000 to $20,000 a month for skills like knowledge work and Ph.D.-level research.
The details:
OpenAI is planning three agent tiers: business professionals ($2k/mo), advanced software devs ($10k/mo), and PhD-level researchers ($20k/mo).
Investor SoftBank has already reportedly committed $3B to these agent products for 2025 alone.
The agentic offerings are expected to generate up to 25% of OpenAI's long-term revenue as the company expands beyond its current offerings.
In January, CEO Sam Altman predicted that 2025 would see the first AI agents “join the workforce and materially change the output of companies.”
Why it matters: With price tags rivaling senior employee salaries, OpenAI is betting big that specialized AI agents can deliver enough value to justify the enterprise-level subscription. The move could set new precedents for AI agent pricing while revealing just how much companies are willing to pay for automated expertise.
TOGETHER WITH SANA
🤝 AI agents that understand your business
The Rundown: Sana's unified agent platform brings AI capabilities to your company with a single interface that connects seamlessly to your existing data ecosystem — no need for complex implementation.
Sana's platform enables you to:
Turn manual processes into automated workflows in minutes
Create meeting assistants that capture insights and action items
Develop agents that work across multiple applications seamlessly
Search across every app and deploy agents for deep research
🔎 Google Search adding new ‘AI Mode’
Image source: Google
The Rundown: Google just launched AI Mode, a Search Labs experiment that turns traditional search into a conversational experience powered by a custom Gemini 2.0, along with updates to AI Overviews.
The details:
AI Mode uses a "query fan-out" technique, launching simultaneous searches across diverse sources to assemble detailed answers with relevant sourcing.
Users can continue their search by asking follow-up questions directly in AI Mode, receiving well-reasoned responses with curated links to explore further.
Google also upgraded AI Overviews with Gemini 2.0, improving responses to more challenging topics like coding, advanced math, and multimodal queries.
The company also said it is expanding access to AI Overviews to teens and removing sign-in requirements.
Why it matters: Search continues to evolve in the AI era, and Google faces serious pressure from rivals like Perplexity, Grok, and ChatGPT. The new AI Mode looks to create a bridge between familiar search interactions and advanced, conversational AI — resulting in a potentially more comfortable (yet powerful) web experience.
AI TRAINING
💻 Connecting Claude Projects with GitHub repos
The Rundown: In this tutorial, you will learn how you can use Claude's GitHub integration to connect repositories directly to your AI assistant for comprehensive code understanding and support.
Step-by-step:
Set up a Claude project named “Code Assistant: [Repo Name]" with description: “Workspace to analyze and improve [Repository Name] codebase.”
Select "GitHub" from the “Add Content” menu and authorize the Claude GitHub app when prompted.
Choose your repository and select specific files you’ll need help with.
Start asking questions about your code – Claude can explain functions, suggest improvements, and even help with debugging.
Pro tip: You can also use the "Sync now" button to update your project whenever your repository changes, ensuring Claude always works with your latest codebase.
PRESENTED BY AIRCOVER
🎯 Turn every rep into a top performer with AI
The Rundown: Aircover’s enterprise-grade AI Agents turn every rep into a top performer with real-time coaching, next-gen conversation intelligence, and automated workflows.
GTM teams in the Fortune 1000 have seen:
40% more deals move past Stage 1 with better discovery
Elimination of 90% of busywork with instant insights, CRM updates, and price quotes
4x higher win rates with reps enabled to answer technical and competitive questions on day one
Explore Aircover and start closing more deals today.
ALIBABA
🧠 Alibaba’s cheap and efficient QwQ-32B AI
Image source: Qwen
The Rundown: Alibaba’s Qwen team released QwQ-32B, a new AI reasoning model that leverages reinforcement learning to match or surpass the performance of larger competitors like DeepSeek-R1 at a fraction of the cost.
The details:
QwQ-32B uses reinforcement learning at scale, significantly boosting performance on advanced math, coding, and reasoning-based tasks.
The model is roughly 20x smaller than DeepSeek-R1 yet delivers comparable or superior performance across key benchmarks.
It is priced at just $0.20 per million input and output tokens, a roughly 90% reduction compared to similar performing models like R1 and o1-mini.
Qwen has open-sourced the model under the Apache 2.0 license, with availability on Hugging Face and Alibaba Cloud's ModelScope platform.
Why it matters: China’s open-source models continue to accelerate — with this latest launch from Qwen showing off some major performance gains despite shrinking size (near-frontier intelligence on-device is here) and cost. Clever training approaches are still proving their worth over raw model size as labs continue to push closer to AGI.
QUICK HITS
🛠️ Trending AI Tools
👁️ Aya Vision - Cohere’s new SOTA multilingual visual model
🗣️ Sesame - Conversational speech model for natural, engaging conversations
🎵 DiffRhythm - Generate complete 4-min songs w/ vocals in just 10 seconds
🎬 ReframeAnything - Resize any video in one click
📰 Everything else in AI today
Social media platform Digg is being revived by former founder Kevin Rose and Reddit co-founder Alexis Ohanian, with AI-enhanced moderation and UX.
OpenAI rolled out its GPT-4.5-Preview model to all Plus users after launching last week exclusively to Pro and developers via API.
A federal judge officially denied Elon Musk's request to block OpenAI's transition from nonprofit to for-profit status but allowed other aspects of his lawsuit to proceed.
Andrew Barto and Richard Sutton won the 2024 Turing Award for pioneering reinforcement learning in the 1980s and warned against rapid AI deployment.
Scale AI secured a multimillion-dollar contract from the U.S. DoD for “Thunderforge,” a program that will deploy AI agents for military planning and operations.
Codeium released Windsurf Wave 4, introducing new features like AI-powered previews for rapid app iteration, tab-to-import functionality, and suggested actions.
Luma Labs introduced three new features to its Ray2 video model, with Keyframes, Extend, and Loop offering more control over generations.
COMMUNITY
🎥 Join our next live workshop
Join our next workshop this Friday at 3:30 PM EST with Dr. Alvaro Cintas, The Rundown’s AI professor. You'll learn how to use the latest AI tools to take your coding to the next level, and how we're leveraging AI-powered coding at The Rundown.
RSVP here. Not a member? Join The Rundown University on a 14-day free trial.
🤝 Share The Rundown, get rewards
We’ll always keep this newsletter 100% free. To support our work, consider sharing The Rundown with your friends, and we’ll send you more free goodies.
See you soon,
Rowan, Joey, Zach, Alvaro, and Jason—The Rundown’s editorial team
Sanctuary's touchy-feely humanoid
Read Online | Sign Up | Advertise
Good morning, robotics enthusiasts. Sanctuary AI has unveiled a new tactile sensor that gives robots a “delicate but durable sense of touch.”
With this advancement, the company’s Phoenix humanoid can now feel and interact with objects much like humans—bringing us closer to a future where machines can perform nearly any task we can.
In today’s robotics rundown:
Sanctuary AI’s touch sensors for humanoids
Dexmate’s powerful transformer robot
Hyundai and Avride’s robotaxi push
XoMotion self-balancing exoskeleton
Quick hits on other robotics news
LATEST DEVELOPMENTS
SANCTUARY AI
👉🏼 Sanctuary AI’s touch sensors for humanoids
Image source: Sanctuary AI
The Rundown: Canadian robotics company Sanctuary AI just outfitted its eighth-generation humanoid, Phoenix, with new touch sensors that the company says give it a richer sense of touch, critical for achieving human-level dexterity.
The details:
The tactile sensors allow Phoenix to perform complex, touch-driven tasks with precision and accuracy, including blind picking.
They ensure Phoenix can “feel” texture and pressure—in stark difference from other humanoids relying on video to interact with their environments.
Each finger pad on Phoenix has a seven-cell touch sensor sensitive to 5 millinewtons (mN)—humans can typically feel levels closer to 3 mN.
Eventually, Sanctuary hopes Phoenix will use these abilities to support industries requiring dexterous handwork—like logistics and auto manufacturing.
Why it matters: While other robotics companies focus on vision-language action and bipedal walking, Sanctuary believes that a human-like touch is the key to creating robots that can perform as effectively as humans in any environment. The other company moving in this direction is California-based Proception.
DEXMATE
🤖 Dexmate’s powerful transformer robot
Image source: Dexmate AI
The Rundown: US-based robotics company Dexmate AI launched Vega, a robot with a unique transformer-like design featuring high-payload arms and a body that folds down to fit into the back of an SUV—like a true transformer.
The details:
Priced at $89,999, Vega uses high-payload arms with dexterous hands to lift objects weighing more than 30 lbs—suitable for light logistical use cases.
However, the real deal is the foldable torso: the robot can extend to 2.2m for high-reaching tasks and fold whenever needed, going as low as 0.66m.
It has a wheeled base and advanced sensors, including RGBD/RGB cameras, LiDAR, IMUs, and ultrasonic sensors, to move omnidirectionally.
Plus, there is support for 36 degrees of freedom and advanced processing with Nvidia's AGX Orin edge computing system.
Why it matters: Vega enters the rapidly evolving robotics landscape as a bold competitor. While its payload capacity doesn’t match industrial robots from KUKA or Fetch, its combination of mobility, versatility, and dexterity sets it apart in the market for general-purpose robots with high-payload arms.
AVRIDE
🚘 Hyundai and Avride’s robotaxi push
Image source: AVRIDE
The Rundown: Hyundai Motor Company partnered with Avride, an autonomous driving technology startup, to accelerate the development of a self-driving robotaxi fleet—while also exploring robotic delivery services for the future.
The details:
Avride will expand its fleet of Hyundai IONIQ 5 vehicles, integrating its self-driving technology into units assembled at Hyundai’s new plant in Georgia.
These vehicles are expected to debut in Dallas later this year, exclusively through Avride’s robotaxi service on Uber.
The plan is to expand the fleet to 100 cars by the end of 2025 while exploring other areas of collaboration, including delivery services via Avride robots.
Hyundai has been working with Avride since 2019 and has also set up a joint venture with auto supplier Aptiv for deploying autonomous vehicles.
Why it matters: Hyundai and Avride’s aggressive timeline, combined with the dual focus on robotaxis and robot delivery, could rapidly scale the tech—taking it to the masses. The work can further benefit from Hyundai’s multiple AV partnerships, including its joint venture with Aptiv.
HUMAN IN MOTION ROBOTICS
🚶🏽♀️XoMotion self-balancing exoskeleton
Image source: Human in Motion
The Rundown: Vancouver-based Human in Motion Robotics developed XoMotion, a cutting-edge, self-balancing exoskeleton designed to help people with spinal cord injuries walk again—the company will begin FDA clinical trials in April.
The details:
XoMotion ensures hands-free, omnidirectional bipedal mobility without the need for crutches or walkers, setting it apart from many other exoskeletons.
The device uses 12 actuators at the joints for natural movement, with an array of sensors for detecting shifting positions and the environment.
It can also perform ambulatory tasks like standing, walking, sidestepping, climbing stairs, and crouching—without requiring the use of hands.
After debuting XoMotion at CES in January, the company will begin its clinical trials in late April, testing a version designed for rehab facilities.
Why it matters: XoMotion is currently restricted to clinical and research settings, but personal use is the ultimate goal—with the hope to transform the lives of people with mobility challenges. Other key players pushing this market forward are Wandercraft and ReWalk.
QUICK HITS
📰 Everything else in robotics today
Waymo launched its robotaxi service in Austin exclusively through Uber, with a similar deal expected to roll out in Atlanta soon.
Indian robotics company Muks Robotics unveiled Spaceo, a heavy-duty humanoid with an adjustable height of up to 8 feet and a payload capacity of up to 440 lbs.
Humanoid developer LimX Dynamics raised 500 million yuan ($69.02M) in a Series A round backed by Alibaba Group.
Researchers in China and Hong Kong developed a new AI learning framework that teaches humanoids to stand up from an idle position, regardless of the terrain.
The market for agricultural robotics is expected to surge from $13.4B in 2023 to around $86.5B by 2033.
U.S. researchers have developed a prosthetic hand that can grip everyday objects while carefully conforming and adjusting its grasp to avoid damaging what it is holding.
U.S. researchers also published findings exploring the feasibility of humanoids working in restaurant kitchens to overcome global shortages and ensure quick turnover.
COMMUNITY
🎥 Join our next live workshop
Join our next workshop on [date] to learn [use case/workshop title] with [workshop host].
RSVP 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
OpenAI's $20,000 AI agents
Read Online | Sign Up | Advertise
Good morning, AI enthusiasts. OpenAI is reportedly set to put a massive price on AI expertise—with plans to launch specialized agent subscriptions costing as much as senior executive salaries.
But, with investors already placing multibillion-dollar bets, Sam Altman’s prediction of seeing AI agents “joining the workforce” this year may be quickly coming to fruition.
In today’s AI rundown:
OpenAI launching premium AI agents
Google Search adding new ‘AI Mode’
Connecting Claude Projects with GitHub repos
Alibaba’s cheap and efficient QwQ-32B AI
4 new AI tools & 4 job opportunities
LATEST DEVELOPMENTS
OPENAI
🤖 OpenAI launching premium AI agents
Image source: Ideogram / The Rundown
The Rundown: OpenAI is reportedly preparing to launch a suite of specialized AI agents with price tags ranging from $2,000 to $20,000 a month for skills like knowledge work and Ph.D.-level research.
The details:
OpenAI is planning three agent tiers: business professionals ($2k/mo), advanced software devs ($10k/mo), and PhD-level researchers ($20k/mo).
Investor SoftBank has already reportedly committed $3B to these agent products for 2025 alone.
The agentic offerings are expected to generate up to 25% of OpenAI's long-term revenue as the company expands beyond its current offerings.
In January, CEO Sam Altman predicted that 2025 would see the first AI agents “join the workforce and materially change the output of companies.”
Why it matters: With price tags rivaling senior employee salaries, OpenAI is betting big that specialized AI agents can deliver enough value to justify the enterprise-level subscription. The move could set new precedents for AI agent pricing while revealing just how much companies are willing to pay for automated expertise.
TOGETHER WITH SANA
🤝 AI agents that understand your business
The Rundown: Sana's unified agent platform brings AI capabilities to your company with a single interface that connects seamlessly to your existing data ecosystem — no need for complex implementation.
Sana's platform enables you to:
Turn manual processes into automated workflows in minutes
Create meeting assistants that capture insights and action items
Develop agents that work across multiple applications seamlessly
Search across every app and deploy agents for deep research
🔎 Google Search adding new ‘AI Mode’
Image source: Google
The Rundown: Google just launched AI Mode, a Search Labs experiment that turns traditional search into a conversational experience powered by a custom Gemini 2.0, along with updates to AI Overviews.
The details:
AI Mode uses a "query fan-out" technique, launching simultaneous searches across diverse sources to assemble detailed answers with relevant sourcing.
Users can continue their search by asking follow-up questions directly in AI Mode, receiving well-reasoned responses with curated links to explore further.
Google also upgraded AI Overviews with Gemini 2.0, improving responses to more challenging topics like coding, advanced math, and multimodal queries.
The company also said it is expanding access to AI Overviews to teens and removing sign-in requirements.
Why it matters: Search continues to evolve in the AI era, and Google faces serious pressure from rivals like Perplexity, Grok, and ChatGPT. The new AI Mode looks to create a bridge between familiar search interactions and advanced, conversational AI — resulting in a potentially more comfortable (yet powerful) web experience.
AI TRAINING
💻 Connecting Claude Projects with GitHub repos
The Rundown: In this tutorial, you will learn how you can use Claude's GitHub integration to connect repositories directly to your AI assistant for comprehensive code understanding and support.
Step-by-step:
Set up a Claude project named “Code Assistant: [Repo Name]" with description: “Workspace to analyze and improve [Repository Name] codebase.”
Select "GitHub" from the “Add Content” menu and authorize the Claude GitHub app when prompted.
Choose your repository and select specific files you’ll need help with.
Start asking questions about your code – Claude can explain functions, suggest improvements, and even help with debugging.
Pro tip: You can also use the "Sync now" button to update your project whenever your repository changes, ensuring Claude always works with your latest codebase.
PRESENTED BY AIRCOVER
🎯 Turn every rep into a top performer with AI
The Rundown: Aircover’s enterprise-grade AI Agents turn every rep into a top performer with real-time coaching, next-gen conversation intelligence, and automated workflows.
GTM teams in the Fortune 1000 have seen:
40% more deals move past Stage 1 with better discovery
Elimination of 90% of busywork with instant insights, CRM updates, and price quotes
4x higher win rates with reps enabled to answer technical and competitive questions on day one
Explore Aircover and start closing more deals today.
ALIBABA
🧠 Alibaba’s cheap and efficient QwQ-32B AI
Image source: Qwen
The Rundown: Alibaba’s Qwen team released QwQ-32B, a new AI reasoning model that leverages reinforcement learning to match or surpass the performance of larger competitors like DeepSeek-R1 at a fraction of the cost.
The details:
QwQ-32B uses reinforcement learning at scale, significantly boosting performance on advanced math, coding, and reasoning-based tasks.
The model is roughly 20x smaller than DeepSeek-R1 yet delivers comparable or superior performance across key benchmarks.
It is priced at just $0.20 per million input and output tokens, a roughly 90% reduction compared to similar performing models like R1 and o1-mini.
Qwen has open-sourced the model under the Apache 2.0 license, with availability on Hugging Face and Alibaba Cloud's ModelScope platform.
Why it matters: China’s open-source models continue to accelerate — with this latest launch from Qwen showing off some major performance gains despite shrinking size (near-frontier intelligence on-device is here) and cost. Clever training approaches are still proving their worth over raw model size as labs continue to push closer to AGI.
QUICK HITS
🛠️ Trending AI Tools
👁️ Aya Vision - Cohere’s new SOTA multilingual visual model
🗣️ Sesame - Conversational speech model for natural, engaging conversations
🎵 DiffRhythm - Generate complete 4-min songs w/ vocals in just 10 seconds
🎬 ReframeAnything - Resize any video in one click
📰 Everything else in AI today
Social media platform Digg is being revived by former founder Kevin Rose and Reddit co-founder Alexis Ohanian, with AI-enhanced moderation and UX.
OpenAI rolled out its GPT-4.5-Preview model to all Plus users after launching last week exclusively to Pro and developers via API.
A federal judge officially denied Elon Musk's request to block OpenAI's transition from nonprofit to for-profit status but allowed other aspects of his lawsuit to proceed.
Andrew Barto and Richard Sutton won the 2024 Turing Award for pioneering reinforcement learning in the 1980s and warned against rapid AI deployment.
Scale AI secured a multimillion-dollar contract from the U.S. DoD for “Thunderforge,” a program that will deploy AI agents for military planning and operations.
Codeium released Windsurf Wave 4, introducing new features like AI-powered previews for rapid app iteration, tab-to-import functionality, and suggested actions.
Luma Labs introduced three new features to its Ray2 video model, with Keyframes, Extend, and Loop offering more control over generations.
COMMUNITY
🎥 Join our next live workshop
Join our next workshop this Friday at 3:30 PM EST with Dr. Alvaro Cintas, The Rundown’s AI professor. You'll learn how to use the latest AI tools to take your coding to the next level, and how we're leveraging AI-powered coding at The Rundown.
RSVP here. Not a member? Join The Rundown University on a 14-day free trial.
🤝 Share The Rundown, get rewards
We’ll always keep this newsletter 100% free. To support our work, consider sharing The Rundown with your friends, and we’ll send you more free goodies.
See you soon,
Rowan, Joey, Zach, Alvaro, and Jason—The Rundown’s editorial team
No matching search results
Try using different keywords, double-check your spelling, or explore related categories.
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.