His mother fell at home, broke her arm and spent 10 minutes on the floor waiting for help
So he built this.
No smartwatch.
No emergency button.
No cameras watching every room.
Just a small AI robot that listens for the sound of a fall, uses radar to confirm someone is there, and alerts a caregiver in seconds.
Watch 0:20 — Pebble starts moving through the home on its own.
At 0:34, the system switches from normal to emergency.
He built the first working version in just two months.
The goal?
Not to prevent every fall.
To make sure nobody has to lie there alone, waiting for someone to notice.
One year ago, this farmer started using GPT.
Today, his farm is doing $30M in revenue.
No huge tech team.
No expensive automation startup.
No army of engineers.
Just GPT helping him diagnose problems, write code and automate parts of the greenhouse he used to manage manually.
The result?
One farmer can suddenly operate more like an entire technical team.
And this is after just one year of using AI.
It’s hard to imagine what farms like this could look like 5 years from now.
A 4-person startup in LA is building drones designed to blend into the sky like real birds
Watch 0:30
One founder worked on Mars and lunar missions at NASA JPL.
The other built Starship hardware at SpaceX.
No loud quadcopter.
No obvious drone silhouette.
No giant defense contractor.
At 0:48, it shows what they are striving for.
The goal?
Make reconnaissance look like wildlife.
Honestly, if one of these flew over you, would you even look twice?
OpenAI made a $230 keyboard for controlling AI agents
It sold out in about 12 hours.
No screen.
No GPU.
No new model.
Just 13 keys, a joystick and a dial that lets you control Codex without constantly jumping between tabs.
The keys even change color depending on whether your agents are thinking, running, waiting or done.
The result?
One sold for $1,250 on eBay. Another was listed for $1,850.
The AI hardware era hasn’t even really started.
People are already collecting the artifacts.
China temporarily closed part of its coastline during military drone exercises
-One man decided the swimming ban didn’t apply to him
No lifeguard ran after him
No patrol boat was sent out
No officer even had to step onto the beach
An AI-controlled drone found him first
It detected him entering the restricted area, flew over the water, locked a spotlight on him and ordered him through a loudspeaker to turn back
No human was actively flying the drone
The AI was handling the patrol, spotting violations and deciding when to intervene on its own
One system
Miles of coastline
Almost no manpower
We thought AI surveillance would mean smarter cameras
Turns out it might just fly
A small team wanted to clean rivers without spending millions on some overengineered robot
No giant factory
No huge robotics team
No “AI will save the planet” pitch deck
Just a few AI tools, a simple idea, and a machine designed to do one job really well
The result?
A boat that drives through polluted water and literally eats the trash in front of it
Sometimes the best tech isn’t the most advanced.
It’s the thing that actually works
This robot grows instead of walking
FiloBot moves by 3D-printing its own body in real time, extending itself toward light and through tight spaces like a climbing plant
No wheels
No legs
No traditional navigation
It just keeps building more of itself until it gets where it needs to go
Honestly, this feels like a more interesting direction for robotics than another humanoid doing a backflip
A father and son in South Africa spent two years building this drone
No $50M seed round
No huge engineering team
No “AI-powered” landing page promising to change the world
Just constant testing, broken prototypes, and four versions of the same idea
The result?
A custom-built drone that hit 408 mph
That’s faster than a lot of supercars
Tech loves talking about moving fast
These guys took it literally
@j_soph48 Not much at full send 😅 This build is optimized almost entirely for speed. I haven’t seen a confirmed flight-time figure for the 408 mph setup though.
My 2026 coding stack:
Claude — writes the code
GPT — explains why it doesn’t work
Local model — saves me from going bankrupt
Kimi — fixes what all three confidently got wrong
Balance is more important than perception and strategy — it's quite counterintuitive; you usually think that robots will optimize for attack rather than for staying on their feet. It turns out that the entire struggle is not for the best strike, but for who can remain upright the longest
China has robot MMA now, and it looks way better than I expected.
These are full-size EngineAI T800 robots fighting inside an actual cage.
One of them got its head kicked off.
It kept fighting.
The interesting part is that teams use the same robot platform, so the competition comes down to software, control, balance and engineering.
Real Steel is slowly becoming a benchmark.
At first this just looks like another crazy Unitree robot demo
But the more interesting part is what’s happening behind the hardware
NVIDIA is already using Unitree robots as part of its Physical AI stack
Their new Isaac GR00T reference humanoid combines a Unitree body with Jetson Thor and NVIDIA’s robot foundation models.
And GR00T N1.5 has already been trained on the Unitree G1, reaching 98.8% success on one manipulation test after 1,000 demos.
That’s why Unitree is interesting.
They’re not just building robots that move well.
They’re slowly becoming one of the bodies AI labs can plug a brain into