In this video, the creator explores Pstack, a plugin developed by Lauren Tan (known as Potato on Twitter), an engineer at Cursor and Grok Bot. Pstack, also referred to as “potato mode,” is designed to improve coding efficiency by focusing on writing less but higher-quality code through a structured stack of skills. The plugin operates by selecting appropriate playbooks based on the task described in plain language, guiding the coding process step-by-step with embedded engineering principles and verification loops to ensure code quality and reliability. The creator emphasizes that Pstack encourages deep, thoughtful coding rather than brute forcing large amounts of code, which has helped Potato achieve remarkable productivity.
The core mechanism of Pstack involves invoking “potato mode” with a task description, which then selects a relevant playbook—such as bug fixes or prototyping—and executes a series of skills and steps to complete the task. These playbooks include clear to-do lists and verification processes where the agent tests its own work, collects evidence like videos or traces, and fixes issues autonomously. Pstack also incorporates principles like laziness protocol, bias towards deletion, and smallest change that solves the problem, which guide the agent’s decision-making to produce clean, maintainable code. Additionally, Pstack supports multi-model collaboration, using different AI models for various tasks to optimize performance.
The creator tests Pstack within Grok Bot, a clean and user-friendly interface that recently became accessible with their subscription. They install the Pstack plugin and run it on a game project called Token Burn, a platformer puzzle game involving AI agents and token-based mechanics. The video demonstrates how potato mode plans and executes the development of the game’s first stage in multiple small, verifiable steps, each accompanied by tests and commits. The process includes planning, coding, testing, and bug fixing, all guided by Pstack’s playbooks and principles. The creator notes the smooth operation of Grok Bot but also mentions the lack of visibility into the underlying tool calls and model usage, which contrasts with more transparent agents like Hermes.
During the development of Token Burn, the creator observes how Pstack’s structured approach leads to clearer commit messages, better code organization, and a focus on proof through real artifacts rather than mere claims. The agent uses different playbooks for planning, feature development, and bug fixing, applying foundational engineering principles throughout. The video also highlights the verification loop where the agent reproduces and addresses a bug autonomously. While the UI and experience in Grok Bot are praised for their simplicity and speed, the creator expresses some reservations about the loss of control and transparency compared to other agent frameworks, especially regarding model selection and usage limits.
In conclusion, the video provides a comprehensive first look at Pstack and its integration with Grok Bot, showcasing how the plugin can enhance coding workflows by combining multi-model AI collaboration, engineering principles, and verification processes. The creator expresses interest in exploring Pstack further in other environments and using specific skills like architect and interrogate. They invite viewers to share their experiences and advice on using Pstack and Grok Bot, acknowledging that this exploration is an initial step toward understanding these powerful tools. The video ends with reflections on the balance between ease of use and control in AI-assisted development.
