Lauren Tan
Software engineer on the SpaceX AI GrokBot team, known on platform X by the handle Poteto. A practitioner in the AI Agentic Coding domain, Lauren Tan is recognized for large-scale AI agent engineering methodologies and contributions to the discipline of building and operating AI agents at scale.
Core Facts
Pstack does not cover the maintenance phase, and Lauren did not provide skills for collecting issues and feeding them back into the planning phase.
Lauren Tan open-sourced a repository called Pstack, building a complete skill system around the AI software development lifecycle.
Timeline (last 90 days)
Lauren Tan was formerly a member of the Meta React team and is now an engineer at Cursor (renamed SpaceX AI).
Lauren Tan pins her hopes on programming languages that are more oriented toward Agents (such as Bend, which combines programming with mathematical proofs), but believes that day has not yet fully arrived.
Lauren Tan believes this "lights-out" automatic PR merging method is highly dependent on the verifiability of the work and is not applicable to domains where verification is difficult, such as healthcare, law, and finance.
Lauren Tan's Agents are set to fully autonomous mode: each PR triggers a strict verification loop, generating multiple verification Agents to perform fuzz testing, fixing issues on their own and repeating until mergeable.
Lauren Tan reviews merged PRs by sampling spot checks rather than fully blocking review, and focuses on error patterns that Agents repeatedly make in order to fix the environment (such as adjusting lint rules, type constraints, or skills).
Lauren Tan currently has more than 10 coordinator Agents, each responsible for a different domain (performance, user bug fixes, exploring rewriting the application in different languages, etc.).
Lauren Tan extracts the deterministic parts into custom CLIs (such as scripts that interact with Playwright and the Chrome DevTools Protocol) to avoid each Agent repeatedly building verification scripts.
Lauren Tan uses a hill climbing approach, letting Agents autonomously iterate and improve based on scoring criteria within the verification loop.
Lauren Tan points out that environment elements such as codebase structure, constraint rules, and verification tools are entirely within engineers' control and have a huge impact on Agent performance.
Lauren Tan believes the ideal Agent working environment should be likened to a "Michelin kitchen," emphasizing craft and quality rather than mechanical replication.
36 more timeline events
All Facts (20)
Pstack does not cover the maintenance phase, and Lauren did not provide skills for collecting issues and feeding them back into the planning phase.
65%VerifiedLauren Tan open-sourced a repository called Pstack, building a complete skill system around the AI software development lifecycle.
65%UnverifiedLauren Tan points out that environment elements such as codebase structure, constraint rules, and verification tools are entirely within engineers' control and have a huge impact on Agent performance.
50%UnverifiedLauren Tan believes this "lights-out" automatic PR merging method is highly dependent on the verifiability of the work and is not applicable to domains where verification is difficult, such as healthcare, law, and finance.
50%UnverifiedLauren Tan was formerly a member of the Meta React team and is now an engineer at Cursor (renamed SpaceX AI).
50%UnverifiedLauren Tan reviews merged PRs by sampling spot checks rather than fully blocking review, and focuses on error patterns that Agents repeatedly make in order to fix the environment (such as adjusting lint rules, type constraints, or skills).
50%UnverifiedLauren Tan currently has more than 10 coordinator Agents, each responsible for a different domain (performance, user bug fixes, exploring rewriting the application in different languages, etc.).
50%UnverifiedLauren Tan extracts the deterministic parts into custom CLIs (such as scripts that interact with Playwright and the Chrome DevTools Protocol) to avoid each Agent repeatedly building verification scripts.
50%UnverifiedLauren Tan's Agents are set to fully autonomous mode: each PR triggers a strict verification loop, generating multiple verification Agents to perform fuzz testing, fixing issues on their own and repeating until mergeable.
50%UnverifiedLauren Tan uses a hill climbing approach, letting Agents autonomously iterate and improve based on scoring criteria within the verification loop.
50%UnverifiedLauren Tan uses a dozen or so coordinator Agents to merge more than 2,500 PRs into production every month.
50%UnverifiedLauren Tan believes the ideal Agent working environment should be likened to a "Michelin kitchen," emphasizing craft and quality rather than mechanical replication.
50%UnverifiedLauren Tan built an open-source skill system called PStack, in which the verification skill is the most critical component.
50%UnverifiedThe talk quoted Lauren Tan: "If an agent can verify its own work, nothing else matters; if it can't, nothing else matters either."
50%UnverifiedLauren Tan previously worked at SpaceX and Meta, and is a member of the React core team.
50%UnverifiedLauren Tan experienced burnout on Meta's React team and took a month off to recover.
50%UnverifiedLauren Tan once worked at Netflix and held a manager position.
50%UnverifiedLauren Tan went to Australia to study finance and graduated with honors.
50%UnverifiedLauren Tan once worked on Meta's React team.
50%UnverifiedLauren Tan (online handle Potato) has joined Cursor.
50%