Pinned Tweet
I'm not here to look smart.
I'm here to:
- Build things
- Learn out loud
- Connect with real ones
If that's you too, let's connect @ANMOLSAINII
Great decision by a Great Leader 🫡
JUST IN: Elon Musk officially signs petition to reinstate Jensen Huang as Nvidia $NVDA CEO.
Humanoid robots do not need to look human to become economically useful.
Gemini Robotics 2.5 can perform tasks such as folding clothes, tying bows, handling pastries with tongs and pipetting liquids.
Google evaluated it across more than 200 tasks and reports a 30% to 100% higher success rate than previous baselines, depending on the task.
But the more interesting development is how Google is improving these robots.
Its Gemini 4 Argon model can create simulated practice environments from real-world camera footage, generate thousands of practice scenarios and analyze failed attempts to suggest improvements.
Google also tested the system with Apptronik and Mercedes-Benz, where an Apollo 2 humanoid robot achieved a reported 96% success rate while sorting automotive parts.
The bigger development is not just better robots.
It is AI helping robots learn new physical skills faster.
That could make automation more flexible across industries.
Planning, 3D asset creation, coding, visual inspection, debugging, and revisions can be divided between specialized agents.
The value is shifting from generating code to completing workflows.
Carbon made a pelican demo so you don't have to 😅
fun test of long-horizon coding and the new /swarm feature in Google @Antigravity.
Builder and critic subagents ran iterative loops across Blender and Three.js - checking their own renders to build a playable 3D game of a pelican biking around our Mountain View campus.
15
Manus just made its AI video editor more useful for professional creators.
You can now export Manus Studio projects to Premiere Pro, DaVinci Resolve, CapCut, and Final Cut Pro.
This means creators can use Manus to generate and work on their videos, then continue editing in the software they already use.
That is an important move.
AI video tools have mostly competed on how impressive their generated videos look.
But professional production involves much more than generation. Editors need control over cuts, pacing, captions, sound, transitions, and the final presentation.
Manus is moving towards a workflow where AI handles more of the initial production while humans retain control over the finishing process.
The interesting question is no longer just which AI can generate the best video.
It is which AI can fit into the actual production workflow.
1
109
OpenAI just introduced GPT-6 Cyber, a specialized AI model built for cybersecurity research.
It is already being used by security teams at Mozilla, Palo Alto Networks, Calif, Cisco and OpenAI.
The model also powered OpenAI's recent Linux kernel research, where an automated pipeline identified 2,308 candidate security findings, reproduced 1,724 vulnerabilities and generated 948 local privilege escalation exploits.
But the bigger story is not the number of vulnerabilities it found.
It is the ability to automate multiple stages of security research, from discovering a weakness to validating it and helping develop a patch.
This changes the economics of vulnerability research.
Security teams could investigate more software, test more potential weaknesses and respond faster.
But attackers can benefit from the same underlying capabilities.
That is why OpenAI is initially restricting access to selected cybersecurity customers.
The real competition is becoming a race between discovering vulnerabilities and fixing them.
307
Elon Musk demonstrated a futuristic 3D rocket-engine interface in 2013.
Now, Perplexity's Computer has recreated it using nothing but the original YouTube video as its reference.
The result is a browser-based 3D wireframe of SpaceX's Merlin 1D engine containing 7,677 segments.
But the interesting part is how it works.
You can open your hand to grab and rotate the engine, close your fist to release it, and use two hands to zoom, move and roll the model.
It uses Three.js for 3D rendering and MediaPipe for hand tracking through a webcam.
Think about what happened here.
An AI agent did not just generate a webpage from a text description. It studied an existing software demonstration and recreated its visual structure and interactive behavior.
That is a much more interesting direction for AI coding.
The next step is making this reliable enough for complex, real-world software.
37
The RAM shortage could have consequences beyond expensive gaming PCs.
Think about the number of products that depend on memory.
Laptops, smartphones, tablets, gaming consoles, workstations, servers, and AI accelerators all need different types of memory.
When AI infrastructure absorbs a larger share of manufacturing capacity, other technology markets can face pressure.
That could affect product pricing, upgrade cycles, and the specifications manufacturers choose for future devices.
There is another consequence.
If local AI becomes more capable but the hardware required to run it becomes more expensive, consumers could find themselves increasingly dependent on cloud-based AI services.
That outcome is not guaranteed.
But it is worth watching because the economics of hardware availability can influence which AI model people actually get to use.
AI infrastructure is becoming a political negotiation.
Meta President Dina Powell McCormick shared an interesting example at the WSJ Tech conference.
A newly elected governor who made data centers a major campaign issue reportedly asked Meta to meet after inauguration.
The reason?
The governor did not want the investment and jobs to go elsewhere.
This reveals an important tension in the AI infrastructure race.
Governments want the economic benefits of AI investment, but they also face pressure over electricity costs, water consumption, local infrastructure, and community impact.
For companies like Meta, getting access to land and GPUs is only part of the challenge.
They also need local governments willing to approve projects and communities willing to accept them.
The next AI infrastructure battle may be fought as much in government offices as in engineering labs.
48
What do Steve Jobs, Walt Disney, Leonardo da Vinci, and Frank Lloyd Wright have in common?
They kept working until the very end of their lives.
Brian Chesky shares why these people are his heroes:
"Walt Disney was dying of lung cancer... He was plotting out Disney World."
For some founders, building isn't just a career. It's a part of who they are.
And perhaps the biggest measure of ambition isn't how much you achieve, but whether you still have something you desperately want to create.
30
Brian Chesky thinks there's another way.
"You're like a golf instructor and you watch their swing."
He explains that founders need to observe, correct, and repeat until the right approach becomes muscle memory.
He once pushed Airbnb's team through 70 iterations of a single press release to demonstrate what excellence looked like.
But here's the distinction:
Founder mode isn't about controlling every decision forever.
It's about getting deeply involved long enough to establish standards that the entire organization can eventually maintain.
The goal isn't to make yourself indispensable. It's to make your standards independent of you.
1
31
Airbnb's CEO lost his entire executive team because he wanted to run the company differently.
Brian Chesky said:
"The entire exact team left."
Instead of backing down, Chesky decided to get deeply involved in every detail of the business.
His philosophy?
"You don't have to convince everyone. You just have to convince the people that are willing to stay and you have to convince enough people to join."
He didn't try to win over the entire organization. He focused on building a small team willing to execute his vision.
The bigger lesson: Great companies aren't always built by following conventional management wisdom.
Sometimes, founders have to rebuild the organization around the way they believe the business should operate.
Conviction attracts the right people. Results prove whether that conviction was right.
14
"I believed in myself at a very small scale."
Airbnb CEO Brian Chesky reflects on his early days as a founder.
He could see himself leading 5 or 10 people. But he struggled to imagine leading thousands.
Interestingly, his co-founders Joe Gebbia and Nathan Blecharczyk saw something bigger in him than he could see in himself.
There's an important lesson here: Sometimes, the people building alongside you recognize your leadership potential before you do.
You don't always need to see the entire destination to start building toward it.
1
49
Airbnb CEO Brian Chesky shared an interesting leadership practice:
"I probably sucked all the options out of the room."
After proposing a new company tagline, Chesky realized his confidence might have discouraged his creative team from disagreeing.
So he changed his position the next day.
He gave them two weeks to come up with something better, explicitly telling them he believed a better tagline existed.
There's a deeper leadership lesson here:
When a CEO speaks with too much certainty, employees can mistake conviction for a final decision.
The best leaders don't just ask for honest feedback. They actively make room for disagreement.
Because the biggest risk in a leadership team isn't a lack of smart people.
It's a room full of smart people who stop challenging the person at the top.
15
“I don't even know how to quit. That seems like defeat.”
Airbnb CEO Brian Chesky says his father drilled one lesson into him growing up: never quit.
But the most interesting part of his story is what happened during COVID, Airbnb's darkest moment.
His board continued believing in him when things looked bleak.
Founder resilience isn't just about refusing to give up. It's also about having people who believe in your vision when you struggle to believe in it yourself.
13
Brian Chesky admitted:
"I can do anything, therefore I can do everything."
After Airbnb became a global phenomenon in just 18 months, its founders began assuming that building anything else would be just as easy.
But they underestimated the difficulty of scaling while starting new things.
They lacked the leadership, structure, and discipline to support that ambition.
There's a lesson here for every successful founder:
Being good at building one great company doesn't mean you're automatically good at building everything else.
Success can give you confidence. But unchecked confidence can make you forget what made you successful in the first place.
15
Airbnb CEO Brian Chesky explains why most people misunderstand Founder Mode.
“Founder mode is a way to teach people by staying in the details by letting go only after it's muscle memory.”
The idea isn't that founders should control every decision forever.
It's that you cannot delegate a standard your team hasn't learned to recognize.
Chesky argues that people see excellent teams operating independently but miss what made that excellence possible: founders getting deeply involved in the details first.
Delegation creates freedom. But teaching people what excellence looks like creates the ability to use that freedom well.
That's the difference between micromanagement and building a high-performing organization.
12
Anmol.SI retweeted
Most startups think about monopolies too early.
Sam Altman thinks they're asking the wrong first question.
When asked why Y Combinator used to push founders to think about how their startups could eventually become monopolies, Altman clarified his approach:
"I'd always start with just like what is the product? Why do users need it? How you going to get more users?"
His argument is that competitive advantage matters, but it shouldn't come before product-market fundamentals.
A startup's first priority should be building something people genuinely need, figuring out how to attract users, and creating a reason for them to stay.
Only then does the question of long-term competitive advantage become increasingly important.
The deeper lesson: A moat cannot compensate for a product nobody wants.
Founders should focus on earning demand before figuring out how to defend it.









