One Engineer Ships 2,000 Pull Requests Monthly Using Custom AI Tool

This approach reflects a shift toward engineer-owned quality gates in fast-moving AI teams

Lauren Tan, an engineer on the Grok team at SpaceXAI, has revealed how her personal agent workflow tool, pstack, enables her to ship nearly 2,000 pull requests to production each month. Previously employed at Cursor and Meta, Tan shared her approach in a recent guide, highlighting the role of automated verification in maintaining confidence at such a high volume. Her workflow allows her to average close to 100 PRs per working day, a pace made possible by integrating real-time checks directly into her development process. The core of Tan’s method lies in pstack, a custom-built agent that automates verification steps typically done manually. By embedding validation into the PR lifecycle, she reduces reliance on lengthy review cycles while ensuring code quality remains high. Tan emphasized that verification is not a bottleneck but the foundation enabling speed, allowing her to catch issues early without slowing down deployment.

This approach reflects a shift toward engineer-owned quality gates in fast-moving AI teams. How Verification Enables High-Volume Shipping Tan explained that pstack handles tasks like linting, test execution, and security scanning automatically upon PR creation. If any check fails, the system blocks merging and provides immediate feedback, preventing flawed code from advancing. She noted that this automation frees her to focus on writing code rather than managing process overhead. The tool also learns from past failures, adapting its checks over time to catch recurring issues more effectively. This feedback loop has significantly reduced regression rates in her workflow. What Would Happen Without Automated Checks? Without pstack’s verification layer, Tan said she would be unable to sustain her current output due to the cognitive load of manual validation. She estimated that manual checks would add at least 10 minutes per PR, making 2,000 monthly shipments impractical.

Instead, the tool’s real-time responses keep her workflow fluid, turning verification into a seamless part of development. This balance, she argues, is what makes extreme velocity sustainable rather than reckless. Frequently Asked Questions How does pstack differ from standard CI/CD pipelines? pstack operates as a personal agent integrated into Tan’s workflow, providing instant, contextual feedback directly in her development environment, unlike traditional pipelines that often run later and require separate monitoring. Can other engineers replicate this workflow using existing tools? Tan suggests that while pstack is custom-built, similar results can be approached by combining lightweight automation scripts with early-stage verification hooks in popular IDEs and version control systems. What impact has this workflow had on team practices at SpaceXAI? Her approach has inspired discussions about decentralizing verification, though adoption varies across teams due to differing project constraints and regulatory requirements in aerospace-adjacent AI systems.