About Lauren Tan
Lauren Tan, known online as poteto, is an engineer on the core team behind Grok Bot, SpaceXSI's product for creating AI teammates. She previously worked on the React Compiler and on Cursor, and she is the author of pstack, an open-source plugin of agent skills that encodes how she wants coding agents to do engineering work.
She grew up in Singapore, studied accounting and finance at Monash University in Australia, and taught herself to program after finding she enjoyed building the spreadsheets and macros in a financial modeling course more than the finance. Her first conference talk came from an invitation by the Ember community after she blogged about what she was learning.
In November 2026 she reported shipping 3,000 pull requests to production in a month in an AI Engineer talk; on Y Combinator's Lightcone podcast the same month, her colleague Roman Ugarte put her rate at about 4,000 a month.
Career Highlights
- Core team member on Grok Bot, taken from first line of code to launch in about seven weeks
- Author of pstack, a set of agent skills for rigorous engineering workflows, used across her company
- Worked on the React Compiler
- Self-taught developer with a degree in accounting and finance from Monash University
Notable Positions
On Corrections as Code
Tan argues that every time an engineer corrects an agent, the correction should become a deterministic check (a lint rule, a type, an API that makes the mistake impossible) instead of another line of prose in a rules file. Agents copy what the codebase already does, so bad patterns have to be removed, not just discouraged. See Software Factory.
On Verification Over Code Review
Her team's code review has shrunk to checking verification artifacts: videos, screenshots, diagrams and formal models. Agents run the application themselves, the team model-checks user flows with TLA+ and proves properties of a virtualization library in Lean. She says most of her token usage goes to verification. See Closing the Loop.
On What Automation Should Touch
She points her automated pipelines at maintenance, performance and bug fixes, and keeps feature decisions with people. Her reasoning is that software built for humans still needs an intentionally designed experience.
Key Quotes
- "So whenever you're correcting your agent, think about how you can make that a deterministic check instead of adding another line to your rules or your agents.md." (AI Engineer talk)
- "if your inner coding loop produces slop, then automating that loop just produces more slop." (AI Engineer talk)
- "I've gotten to a point where I essentially have my agents merge their own pull requests." (Lightcone)
Related Reading
- Lauren Tan: 11,000 Agent PRs a Month, 0.5% Reverted - Full breakdown of the AI Engineer talk
- Roman Ugarte and Lauren Tan on Grok Bot's AI Coworkers - The Lightcone conversation
- Factory Engineering - Building the system that builds the product