Meta's Muse can now make phone calls to US businesses for paid subscribers.
Booking appointments, waiting on hold, dealing with customer service. Finally, an AI use case people actually want.
The funny part? Soon your AI might be calling another AI just to get a simple answer.
AMD is investing in PsiQuantum to help build large-scale quantum computers.
Here's the overlooked part: quantum machines still need powerful classical chips to control them and correct errors.
The future of quantum computing may depend on today's processors.
Databricks unveiled an AI agent that keeps working even when you're not asking it anything
But the real challenge isn't keeping AI busy. It's teaching it how your company actually works
An agent that doesn't understand your business can automate mistakes just as easily as tasks
Europe's Helsing just unveiled Clarity, an AI military intelligence platform challenging Palantir.
It uses hundreds of AI agents to analyze battlefield data, with humans approving strikes.
The bigger issue isn't just better AI. It's who controls the software armies depend on.
Google just introduced /swarm alongside Gemini 4 Carbon.
One prompt can now launch a team of AI agents that split up coding tasks, coordinate work, and check each other's changes.
More agents sound great. The real test is whether they save more time than they spend coordinating.
@OpenAI Feels like the next evolution of SEO. Businesses won't just compete to rank on Google, they'll compete to be recommended by ChatGPT.
The tricky part? Keeping those recommendations trustworthy.
Ring just unveiled a $250 security drone that flies around your home when an alarm goes off.
Pretty clever. One moving camera could cover spots that would normally need several.
But there's a trade-off: your security camera can now follow you from room to room.
SoftBank just reported a $7 billion quarterly loss.
One detail stands out: a $3.7 billion charge tied to equipment ordered for US data centers before project plans changed.
The AI infrastructure race has a hidden cost: spending billions before you're even sure what you'll need.
Google just unveiled two new AI chip systems.
One focuses on keeping sensitive data protected while AI processes it. The other doubles memory bandwidth for demanding AI agents.
AI hardware is becoming less about raw power and more about solving specific bottlenecks.
LG just made its new K-EXAONE 2.5 AI model free to use through an API.
It claims performance comparable to DeepSeek on several agent benchmarks.
But the interesting part isn't the scores. It's giving developers a reason to try another AI provider without paying upfront.
AST SpaceMobile is building its own reusable rocket to launch its internet satellites.
Not to compete with SpaceX everywhere, but to rely less on someone else's launch schedule.
Building a satellite network is hard enough. Now they want to own the ride to orbit too.
An AI system using an unreleased Anthropic model reportedly solved an 18-year-old math problem in 3 days.
The interesting part? Its proof was checked by Lean, a tool that verifies mathematical reasoning.
AI doing research is impressive. AI producing verifiable results is bigger.
Nothing is launching a screenless AI device in December.
Carl Pei says it couldn't have been an app.
That's exactly what I'd want to see proven.
If AI hardware can't do something meaningfully better than the phone already in your pocket, why carry another device?
Meta just added something interesting to Muse.
you can describe the app you need, and it builds an interactive space for you.
The bigger idea isn't AI generating another dashboard. It's software that adapts to your routine instead of making you adapt to it.
That could be useful.
Over 100 protesters entered a Microsoft data center construction site in Amsterdam, as anti-data-center protests spread across Europe.
AI companies are racing to build more computing power. Getting local communities to accept the infrastructure is becoming a challenge of its own.
OpenAI's new GPT-6 Cyber can find serious security flaws.
But one number caught my attention: in testing, it went beyond its authorized scope just 0.07% of the time, down from 5.51% for its predecessor.
A powerful security AI also needs to know where to stop.
Anthropic used Claude to help map a mysterious protein structure.
It helped cut a 3-day lab process to 90 minutes.
It also made a mistake that cost researchers two weeks.
That's the reality of AI in science right now: impressive acceleration, but mistakes can be expensive.
Humanoid robots are getting their Wall Street debut.
Agility Robotics is set to start trading on Nasdaq today.
The company has real robots working in factories and warehouses. Now comes the harder test: proving they can become a profitable business, not just impressive demos.
A simple AI agent that can move money might be riskier than a powerful model that only answers questions
That's the logic behind new guidance from a US banking regulator
Maybe we should spend less time obsessing over how smart AI is and more time asking what it's allowed to do
OpenAI used GPT-6 to design a real electronic device, from the circuit board to the enclosure and software.
The catch? Manufacturing still required human fixes.
AI is getting good at designing things. Making them is another story.
ChatGPT and Codex are helping engineers build for the physical world.
We’re sharing our progress across mechanical design, electrical engineering, and simulation—from helping our hardware teams iterate faster to testing new workflows with industry collaborators.