- NATURAL 20
- Posts
- Claude Becomes a Coworker, Google Powers Siri, & OpenAI's "Sweet Pea"
Claude Becomes a Coworker, Google Powers Siri, & OpenAI's "Sweet Pea"
PLUS: Salesforce Brings Autonomous AI Agents to Slack, Boston Dynamics Reveals Fully Electric Atlas Robot and more.
Wes Roth
January 14, 2026
In partnership with
Help us make better ads
Did you recently see an ad for beehiiv in a newsletter? We’re running a short brand lift survey to understand what’s actually breaking through (and what’s not).
It takes about 20 seconds, the questions are super easy, and your feedback directly helps us improve how we show up in the newsletters you read and love.
If you’ve got a few moments, we’d really appreciate your insight.
Hey — quick check-in.
This week’s theme feels like AI graduating from “cool demo” to “real-world systems”: tools made for creators, models made for clinics, and infrastructure made for communities.
Let’s dive in!
Today:
Claude Becomes a Coworker, Google Powers Siri, & OpenAI's "Sweet Pea"
Google Upgrades Medical AI with 3D Imaging & Dictation
Microsoft Commits to "Community-First" Energy & Water Use
Veo 3.1 Brings Character Consistency and 4K to AI Video
Salesforce Brings Autonomous AI Agents to Slack
Boston Dynamics Reveals Fully Electric Atlas Robot
Claude Code just EVOLVED
Claude Code is now Co-work, bringing advanced coding and folder tasks to everyone on Mac. Users give a folder; the AI plans, asks approval, then writes, edits, and browses like a tireless helper. Google launched a shared checkout system so shopping bots can buy straight from Shopify, Target, Walmart and more, signaling AI-driven buying.
Apple will power future Siri features with Google’s Gemini, prompting monopoly fears from Elon Musk. China’s DeepSeek says its coming V4 beats Western models in coding tests. Leaks suggest OpenAI’s first gadget will be “Sweet Pea,” AirPod-like AI earbuds, with 50 million units planned for 2026 release.
Google’s latest Veo update is basically: “stop fighting the tool, start telling stories.”
The big upgrades: more consistent characters/backgrounds, more expressive clips even with short prompts, and native vertical (9:16) so you’re not cropping your life away for Shorts/TikTok/Reels.
On top of that, there’s 1080p + 4K upscaling, and it’s rolling out across the places people actually work: Gemini app, YouTube, Flow, Google Vids, Gemini API, and Vertex AI.
Also worth noting: Google is leaning into provenance—videos get SynthID watermarking, and the Gemini app can help verify whether a video was generated with Google AI.
This one is quieter, but it’s a big deal if you care about AI doing useful things beyond chat.
MedGemma 1.5 (4B) updates Google’s open medical model with better support for a wider range of medical imaging, including: CT, MRI, and whole-slide histopathology—plus chest X-ray time series review, anatomy localization in X-rays, and extracting structured data from lab reports.
Google also shares early benchmark lifts: CT finding classification improves (61% vs 58%), MRI improves more (65% vs 51%), and histopathology slide “fidelity” (ROUGE-L) jumps (0.49 vs 0.02) in one internal setup.
Then there’s MedASR, an open medical speech-to-text model tuned for medical dictation—meant to pair with MedGemma when you want “speech in, structured text out, reasoning on top.”
They’re also trying to pull builders in with a Kaggle MedGemma Impact Challenge offering $100,000 in prizes.
Microsoft is basically admitting what everyone already feels: AI infrastructure is exploding, and communities are nervous—about power, water, and whether locals actually benefit.
They’re launching a “Community-First AI Infrastructure” initiative with five promises:
They’ll pay their way so data centers don’t raise residential electricity prices
Minimize water use and replenish more than they use
Create jobs for residents
Add to the local tax base (schools, hospitals, parks, libraries)
Invest in local AI training + nonprofits
Under the hood, they point to tactics like working with utilities on rate structures, paying for transmission/substation upgrades, and even contracting new generation capacity (they cite 7.9 GW contracted in the Midwest MISO market).
On water: they commit to 40% improvement in water-use intensity by 2030, and describe new closed-loop cooling designs where potable water isn’t needed for cooling in some deployments.
They also highlight workforce programs like expanding their Datacenter Academy with community colleges/vocational partners.
🧠RESEARCH
Researchers created "VideoDR," a new test requiring AI to watch videos and simultaneously search the web for answers. They discovered that complex autonomous AI agents often get distracted and lose focus during long tasks. Surprisingly, simpler step-by-step methods currently perform just as well as these advanced "agent" systems.
"BabyVision" is a benchmark designed to test if AI models possess basic visual skills—like understanding shapes—without relying on text descriptions. Results show that top models, including Gemini 3, score lower than six-year-old children. This proves that despite advanced language skills, AI still lacks fundamental human-like vision.
A new system called "PaCoRe" helps AI solve difficult problems by exploring many potential solutions at once instead of just one at a time. By coordinating these parallel thoughts, a smaller AI model outperformed the massive GPT-5 on math tests, proving that organized "thinking" beats raw size.
🛠️TOP TOOLS
Each listing includes a hands-on tutorial so you can get started right away, whether you’re a beginner or a pro.
BypassGPT – AI Content Humanizer / AI Writing Tool - web and mobile “AI text humanizer” that rewrites AI‑generated text to read more human and score higher on AI‑content checks.
Call Annie – AI Assistant - mobile and web app that lets you practice speaking with AI tutors over real‑time video calls.
Cami AI – AI Messaging Assistant - messaging-first AI assistant that worked inside apps like WhatsApp and LINE, offering text answers, image generation, and voice transcription with customizable “personas.”
📲SOCIAL MEDIA
🗞️MORE NEWS
Salesforce Slackbot Announcement Salesforce has launched a redesigned smart assistant inside Slack that can handle tasks like scheduling meetings and drafting documents. Unlike older bots, this AI can read your team’s messages and files to understand exactly what you are working on. The feature is now available for businesses on higher-tier plans to help employees save time on daily chores.
Boston Dynamics New Atlas Robot Boston Dynamics unveiled a new, fully electric version of its human-like robot, Atlas, which is now ready for factory work. They have partnered with Google to give the robot a smarter brain, allowing it to better understand and react to its surroundings. The company says every robot they can build for 2026 has already been sold to customers.
Meta Cuts Reality Labs Division Meta is planning to fire about 10% of the staff in its Reality Labs division, which builds virtual reality headsets and the "metaverse." This move comes as the company tries to cut costs and shift its main focus toward artificial intelligence. Reports suggest the decision follows a poor public reaction to their latest AI model, Llama 4.
Alibaba-Backed PixVerse The AI video platform PixVerse has raised $60 million in new funding led by the Chinese tech giant Alibaba. This tool allows users to instantly turn text and images into high-quality videos and has already attracted over 100 million users. The company plans to use the money to make their video generation even faster and expand into gaming and advertising.
Anthropic C-Suite & Mike Krieger Mike Krieger, the co-founder of Instagram, is steering new product development at the AI company Anthropic. He is leading a push to create AI tools that can write computer code and complete complex work projects on their own, rather than just chatting. This strategy aims to help software engineers and businesses automate harder tasks in the coming year.
What'd you think of today's edition? |
Reply