On August 10, the day before Grok Bot launched, Lauren Tan had 31,289 followers on X. This morning she has 198,231. That is more than six times as many in ten weeks, and at her current pace of about 3,500 a day she crosses 200,000 this weekend.
In that same stretch, Elon Musk has reposted, quoted, or replied to her at least 21 times. She has been at Cursor for less than seven months and at SpaceX for two. She has not even been there a year, which at most tech companies means she has not hit the standard one-year vesting cliff on her stock yet. Wild. Her talk on shipping 2,500 PRs in a month has 3,944,957 views. Her name now shows up close to 4,000 times a day on X.
Most people who use Grok Bot have never heard of her. They should know who she is. She rebuilt Grok Bot from the ground up, she is the face of it, and her story starts with a teenager in Singapore, a copy of Photoshop, and an Excel macro. This is that story.
Her X avatar is a cartoon of her eating a baked sweet potato. Hence poteto.
2. 📈 The chart
The line is almost flat for weeks. On July 19 she had 29,280 followers, and she was adding a few hundred a week. Then Grok Bot launched on August 11 and the curve bent. 44,444 by August 20. 85,360 by September 1. 105,853 by September 15. 142,749 by October 1. Her biggest single day was October 8, at more than 6,000 new followers.
She marked one milestone herself. On August 24 she passed 50,000 and wrote a long thank-you to X. That post explains how she got here better than anything I could write, and Elon reposted it.
3. 🇸🇬 A nobody in Melbourne
Lauren Tan grew up in Singapore. At her very first conference talk, EmberConf 2015, she told the room her dad gave her a copy of Photoshop when she was 15. She taught herself design from tutorials on the web, then taught herself to code.
She did not study computer science. She went to Monash in Australia for finance, and her first program was an Excel macro. At React Conf 2025 she told the story this way:
“I learned how to code through Excel... I went to college for finance, not computer science. So my introduction to programming was going to this financial modeling class where my professor taught us how to do Visual Basic, VBA macros for Excel. And then I was like, wow, this is so much fun.”
Lauren Tan, React Conf 2025 livestream, 626 views
Before the big jobs there was a failed startup, ThePriceGeek, a price lookup tool for eBay and Craigslist that peaked at about $1,000 a month before the money ran out. She told that story on The Changelog in 2020.
Then came the move that set up the rest of her career. She blogged about learning the Ember framework and tweeted the posts. The EmberConf organizers, @wifelette and @tomdale, found her through those tweets and invited her to speak. In the 50K post above she describes herself back then as “a nobody in melbourne who was still learning how to code in js.” She had never been to a conference, let alone spoken at one.
That talk got her a job at DockYard and a move to Boston in 2015. Another talk got her into Netflix in 2016, building the studio tools behind shows like Stranger Things. She was a tech lead there, then an engineering manager for about two years, then went back to writing code because she missed it. Along the way she started hiring-without-whiteboards, a list of companies that skip whiteboard interviews. It has 52,746 stars.
Every job, she says, came through relationships she built on X. Her line: “i owe my entire career to it.”
4. ⚛️ The React years
In 2020 Dan Abramov, whom she had met at a conference talk in London, asked if she wanted to join the React team at Facebook. She spent the next six years there, much of it on the React Compiler, the tool that makes React apps faster automatically. She was the sole author of the React Compiler Beta announcement and one of three authors of React Compiler 1.0 in October 2025. She co-presented the 1.0 keynote on stage at React Conf.
Her thinking had been heading toward compilers for a decade. In her first talk in 2015, she explained reactive UI with a spreadsheet: two cells and a formula. In 2024 she shipped a compiler that works out those dependencies by itself. Her one-line summary from the React Conf 2024 stage:
“Computers, it turns out, are way better at this type of granular tracking and optimization than we are.”
Lauren Tan, React Conf 2024, “Forget About Memo”, 14,104 views
Here is the twist that makes the rest of this story. In October 2025, at that same React Conf, she was the cautious voice on AI:
“Personally, I think that the best use of AI is really as a pair, someone you're pair programming with, and not a tool to just replace the act of writing code, or worse, outsourcing your thinking... I've got to ship like 100 PRs this week and barely understand what I'm doing...”
Lauren Tan, React Conf 2025, via a re-upload with 599,774 views
Ten months later she was shipping more than 2,000 PRs a month. What changed was not the hype. It was how much she trusted her agents.
5. 🔥 Burned out, then Cursor
By late 2025 she was done with big-company life. She wrote about it in July:
“i knew i wanted to leave meta as early as last year. but i had no idea what was next. i wasn't having fun at work any longer and every hour spent there felt like agony... at @cursor_ai i voluntarily work 12 hours+ 7 days a week. why? because it's fun!”
Lauren Tan, X, 1,274 likes
She took a month off, built side projects, and joined Cursor the week of March 24, 2026. She was supposed to join the cloud agents team. Instead, because she knew React better than almost anyone, she was pulled onto the new agents window, which was days from launch and had serious performance problems. The fixes were slow and manual. In her Compile talk she says she got so frustrated she stopped and thought: “wait, we have agents, what am I doing?”
So she built skills to let agents drive the app, take their own performance traces, and verify their own fixes. That frustration is where everything else in this story comes from.
6. 🐶 From Binnie to Grok Bot
Inside Cursor she built a Slack bot named Binnie, with a little dog as its avatar. Binnie took incoming bug reports, spun up its own computer in the cloud, reproduced the bug, and fixed it. Her reason was refreshingly honest:
“I'm a fairly lazy person. We were getting so many bug reports for Cursor, and I wanted to just fix them all. So I started thinking about how would I make it so my agents could just automatically fix bugs for me while I sleep.”
Her teammate Roshan Sadanani calls it “a really famous Slack bot” that still lives in their Slack, and says Binnie was part of the motivation for Grok Bot: what if everyone could have a bot like that, with its own name, its own computer, and its own routines.
Grok Bot launched in beta on August 11. The first client was a fast prototype. Over the next weeks she tore it down and rebuilt it on a framework she designed for agents to write, codenamed Dune. Her launch post said so:
That refactor took more than 600 PRs, by her count. Dune is strict on purpose. It bans code comments and React's useEffect in CI, keeps Electron's main and renderer processes apart through the import graph, and makes the shortest path the correct one, because agents always take the shortest path. Her description of the result:
“I have a ton of constraints and CIs. It's actually very annoying to write code in GrokBot, but agents absorb all of that annoyance.”
Lauren Tan, Maven workshop on Cursor's agents, 33,614 views
That is why I call her the architect behind Grok Bot. She did not build it alone, and she is the first to say so: “I wouldn't say I built it all.” There is a whole team, including designer Peng Zheng. But the architecture is hers, and so is much of the idea of a bot as a colleague. Peter Yang introduced her on Behind the Craft as Grok Bot's engineering lead. Today her Grok Bot agents fix Grok Bot's own bug reports while she reviews their code. Asked what her job is now, she said she is “almost like a manager for my digital colleagues.”
7. 🚀 2,500 PRs and the trust curve
On September 21 she posted the talk that made her famous outside the AI crowd. It was supposed to be her talk at Cursor Compile London. She skipped the event to run a Grok Bot livestream, so she put the talk on X for free instead.
It is her most-liked post ever: 18,484 likes and 3,944,957 views. The PR counts are her own: about 1,000 in July, about 2,000 in August by the talk, and 2,500 in the post. She is careful to say she never set out to hit a number. The talk is really about trust.
Her model is a curve. At the start you run one to five agents and watch every conversation, because you do not trust them yet. You earn trust with verification skills that let agents run the app, take traces, and prove their work. Only then do you scale up, and eventually you let agents merge their own PRs while you sleep. Her most-repeated line:
“You can't spawn 100 agents when you don't even trust the output of one agent.”
Lauren Tan, Maven workshop, 33,614 views
She fixes agent mistakes in five layers, strongest first: make the mistake impossible in the codebase, then fail the build with lints and CI, then rules and review bots, then skills, and only last a human style guide. Her reasoning is that agents copy whatever patterns they see. Or, as she put it: “Before AI slop, we had human slop.”
That one workshop, recorded on Maven with Colin Matthews the day after Grok Bot launched, has been re-uploaded by at least seven channels with no connection to her. Together those copies have about 870,000 views.
8. 🌌 Grok Bot Galaxy and the Elon effect
From September 15 to 17, she, Matt Palmer, and Roshan Sadanani ran Grok Bot Galaxy: 72 hours, livestreamed, trying to build a company from zero using Grok Bots. About 70,000 people signed up to watch. Her “going live” post did 38.7 million views, and Elon quoted it:
His quote, “Live demo of building a company with @Grok @Bot!”, has 24,408 likes and 29,990,695 views. That week was her biggest ever by views.
He kept coming back. By my count from a full X archive search, Elon has interacted with her at least 21 times between August 24 and October 15. At least eleven reposts of her own posts, on everything from her 50K thank-you to Grok Bot in Microsoft Teams to Team Bots, which he reposted the day it launched. At least four reposts of other people's posts about her. Four quote posts. Two replies, including a heart on one of her posts. I caught two of the reposts live, as notifications on my phone:
On October 4 he quoted a post that called her “the SpaceXAI engineer who leads Grok Bot” with three words:
That one has 5,198,579 views. And when Elon called the new Tesla Roadster a candidate for the greatest product demo of all time, she quoted him with “hey @bot please preorder the new Tesla Roadster for me immediately, thanks.” He quoted her right back with a single emoji, 🔥. That got 12,305 likes.
9. 🥔 pstack, the potato sack
Her open-source work spread just as fast. pstack is the set of agent skills she uses every day, published in Cursor's official plugins repo (12,195 stars). The name is a joke. Garry Tan has gstack. In her Maven workshop she says they share a last name, have no relation, and she wanted to poke fun. The P is for potato. Her README sets the tone:
“there's a growing sense that ai writes too much slop code. i agree. i don't want to ship like a team of twenty slop artists. throughput without quality is not a goal i aspire to. if you want to go fast, go deep first.”
Lauren Tan, pstack README, GitHub, 12,195 stars
Matt Pocock named her first in his list of top three skill makers. That post has 7,526 likes:
# Cursor
/add-plugin pstack
# Claude Code
/plugin marketplace add cursor/plugins
/plugin install pstack@cursor-plugins
# Then put poteto-mode in front of anything that needs rigor
/poteto-mode this pr has a subtle bug where the scroll drifts every 750ms even when idle. repro first, then fix and verify.
/poteto-mode i'm going to bed. land the stack even if ci flakes. i want everything merged by morning.
The community ported pstack to every other harness within weeks. The biggest port, michael-denyer/pstack-claude, has 2,733 stars. If you want to try it, these are the install lines and two of her example prompts, straight from the README:
One thing in the README stopped me. pstack has no planning skills, because, in her words, “i don't believe in planning. the best spec is code.” My own rule number one is a plan.md before anything. I think hers works because her codebase already does the job my plan does. When Dune makes the wrong move impossible, the plan is built into the repo.
10. 🎯 “You should be more ambitious”
Her second most-liked post is two short lines: “you should be more ambitious / way more ambitious.”
It has 15,368 likes. It is also a fair summary of the last seven months. A self-taught engineer who once thought she might not be cut out for engineering anymore joined a startup in March, watched it get acquired by SpaceX in August, rebuilt the architecture of a flagship product, ran a 72-hour livestream that Elon quoted to 30 million people, and went from 31,000 followers to 198,000 in ten weeks.
Today she asked her new audience a question that tells you where she is headed next: “is chat the final form of agent ui?” It got 936 replies within hours. The person who made agents trustworthy enough to run overnight is now asking what the screen should look like when you wake up. If you use Grok Bot, she is a big part of why it works the way it does. Follow @poteto.
