
Building the agentic OS @RomeAILab / Chief of @rspack_dev / ex-@BytedanceTalk / opinions are my own
Earth
Joined April 2012
- Tweets537
- Following334
- Followers1.9K
- Likes656
Ray Zhang retweeted
Every year I read the Nobel Prize in Physics headline, nod, and understand none of it.
This year my agent made me a 2-minute film, and now I actually get it.
This year's winner buried 5,160 sensors more than a mile deep in the ice under the South Pole, to catch particles that pass through almost everything.
Why ice? Why there?
To catch a ghost particle, you need a trap that's gigantic, crystal clear and pitch dark. The ice deep under the South Pole is all three: a billion tonnes of it, with the air bubbles squeezed out by the weight above and no sunlight reaching down.
Watch 👇
Ray Zhang retweeted
A small milestone: 10M weekly downloads for Rspack
Thank you for building with us ❤️
Ray Zhang retweeted
Many asked me how the video was made. I said you could set up the open source Rome OS repo and get the app.
But... that's probably too much work for some people. So I built a hosted version that turns any question into an explainer video, so you can try it out quickly.
evan.romeos.cc/full/apps/ink…
I am also using it to create a "quantum physics for kids" series (the photo is a poster made by Rome). Will share more in a followup.
Excited to see your ideas!
Opus 5.5 is so good at making hand-drawn whiteboard animations
This one shot video explaining Leidenfrost effect perfectly. Voice over and music added all in one pass.
Maybe we can revolutionize education. Great work by @AnthropicAI
Tried this with mRNA vaccines too. Made the whole thing with @RomeAILab .
Rome + Opus 5.5 might actually be a pretty great tutor for your kids.
Opus 5.5 is so good at making hand-drawn whiteboard animations
This one shot video explaining Leidenfrost effect perfectly. Voice over and music added all in one pass.
Maybe we can revolutionize education. Great work by @AnthropicAI
Ray Zhang retweeted
Opus 5.5 is so good at making hand-drawn whiteboard animations
This one shot video explaining Leidenfrost effect perfectly. Voice over and music added all in one pass.
Maybe we can revolutionize education. Great work by @AnthropicAI
Ray Zhang retweeted
Opus 5.5 is kind of amazing. @claudeai
It one-shot this product explainer and somehow got really close to what I had in my head.
Absolutely love the art style and the music.
This is my favorite @RomeAILab Rome explainer so far.
Ray Zhang retweeted
Computer use is slow?
Not true if the agent can learn the efficient and reliable primitives to do the repeated tasks
Ray Zhang retweeted
Long YouTube videos @YouTube are great — until I realize I don’t have 3 hours to watch one.
So I built a little app on @RomeAILab for myself.
I paste in a YouTube link and get the content in the formats I actually like to use: summaries, mind maps, slides, and a searchable transcript.
I basically customized the whole thing around how I like to learn.
Here’s a quick demo 👇
Fun fact about @RomeAILab:
Once you install the Terminal app, Rome can basically become your remote dev server — run any agent you want, from anywhere.
Skills keep getting more powerful—and more complex. Plain text isn’t always the best interface for them.
I’ve always wanted a product that could turn any complex skill into a simple, interactive app. So when I found a skill for labeling Linear issues, I turned it into an app with Rome.
Everything in the video below came from a single prompt
If you have trouble keeping up with your LinkedIn connections, @RomeAILab can help.
Here’s what @zhangfan_x did: the Rome agent that automatically opens LinkedIn, reads your messages, drafts replies, and waits for your approval before sending them.
This will definitely save you a lot of time.
Hi! I just shipped People in Rome. One person, one history, and a memory that grows about them, across every app your friends message you on. Everything stays local unless you choose to share it.
It's a primitive the OS exposes to your agent and every app. Build on it: a personal CRM, a nudge when someone important needs a reply, an app that prepares a gift before a friend's birthday.
Come build with us: github.com/rome-os/rome
Ray Zhang retweeted
GPT-6 Astra is a beast. Give it a Zillow listing. It can 3D model the house based on the listing photos and create a cool promotion video.
The video was just created in one shot and there are still some wrong details. But I am sure it will be much better if I ask it to polish further. OpenAI is so back.
One of my favorite @RomeAILab user cases: “I don’t want to check mortgage rates every day. Just tell me when it’s time to refinance.”
One sentence, and a mortgage monitor app just pops up.
It watches rates continuously and emails them the moment the rate drops below their target.
That’s Rome.
👇🔗links in comment.
Replying to @zoolsher @RomeAILab
I kept missing good refi windows because I wasn’t watching mortgage rates every day. So I built a little app on @RomeAILab that does it for me.
I set a target rate, and when rates drop below it, I get an email. That’s it. No more checking rates every day. No more refi FOMO.
Fun fact about 🤗 @huggingface
Everyone talking about KD invested in Huggingface as early as March, 2017.
Of course KD has good investment tastes, but people should be reminded that Huggingface were an AI chat bot company at the time and that's why it's name is so cute. Huggingface only started getting known in the AI community when Google launched BERT and a few months later Huggingface open sourced the PyTorch version. Then came the famous Transformers library and the rest is history.
What I learned is that the starting point doesn’t matter as much. Be in the right place, do the right thing, deliver value, and build a vibrant community. The rest will come naturally.
What is an AI Native company?
A company that has never been polluted by traditional ways of working.
We’re not just building @RomeAILab — we’re running our company on it.
As an AI-native company, we use Rome every day to help us get work done. We’ve open-sourced some of the apps we use internally so you can see how we actually work with our agents.
github.com/rome-os/rome-apps

















