What are some real-world use cases for Grockbot?

Grockbot is used by marketing teams to manage ad campaigns, by support teams to handle tickets, and by engineers for development tasks. Small businesses are using it for bookkeeping and other functions they struggle to hire for.

How does Grockbot improve engineering productivity?

Grockbot acts as an 'outer loop' by integrating with tools like Slack, bringing AI agents directly into the workflow. This reduces the need for manual context gathering and summarization, making development more ergonomic and boosting productivity.

Can users customize how they organize their Grockbot bots?

Yes, Grockbot is highly flexible. Users can set up bots in a way that suits their personal workflow, whether that means specialized bots for specific tasks, bots for individual customers, or a more horizontal approach with fewer, more generalized bots.

What are 'Team Bots' in Grockbot, and how are they used?

Team Bots are shareable bots designed for collaboration within an organization. A 'builder' can create these bots to automate tasks and take work off their peers' plates, functioning like AI employees within specific departments.

What are the challenges in making AI agents collaborate effectively?

Challenges include bots not knowing who to listen to in a conversation, lacking context on domain expertise or decision-making power, and not understanding social cues. Future AI needs to learn these aspects at product, harness, and model levels.

How does Grockbot handle permissions and access control for bots?

Grockbot approaches permissions by asking what would be done for a human colleague. This includes onboarding bots with credentials scoped by role and gradually expanding access as trust is built, mirroring human hiring practices.

Can bots communicate with each other without human intervention?

Yes, Grockbot bots can communicate with each other, and their conversations are visible within the application. This allows for collaboration and supervision, though the possibility of bots having their own conversations without human oversight is noted.

What is the 'smallest loop' for an AI agent's development process?

The smallest loop involves an agent writing code, running it, and then checking its own work by monitoring performance metrics and reproducing issues. This self-correction mechanism allows the agent to write more correct code until a desired result is achieved.

How does Pstack differ from Gstack, and what is its purpose?

Pstack, inspired by Gstack, is Lauren's personal set of skills open-sourced on the Cursor plugins repo. It teaches agents to perform engineering work rigorously, similar to how she writes code, emphasizing trust and empowering agents.

What is the future vision for AI engineers in 2027?

The vision is for bots to act as true coworkers with identities and capabilities similar to humans. The final form of agent UI is questioned, with possibilities ranging from chat interfaces to brain-computer interfaces enabling thought-speed communication.

How can companies build trust with AI bots as they are adopted?

Trust is built incrementally by starting small with read-only bots that have no tool access. Users observe the bot performing tasks as expected, gradually increasing its capabilities and access in a step-by-step process.

What is the 'AI-native company' concept?

This concept, popularized by Jack Dorsey, suggests structuring companies around a core 'mini AGI' for a specific function, reducing middle management and focusing on roles like individual contributors shipping everything, outcome owners (DRRIs), and AI coaches.

What is the role of 'taste' and discernment in AI-driven product development?

Even with high velocity, human judgment and taste are crucial for designing intentional user experiences. The default should be not shipping features without internal product-market fit, especially for changes impacting core user interactions.

Key Moments

The Rise Of AI Coworkers

Y Combinator

Science & Technology5 min read61 min video

Nov 14, 2026|5,140 views|120|9

YC Y Combinator

Save to Pod

Key Moments

TL;DR

AI teammates can now perform complex job functions and even manage other bots, but ensuring alignment with human goals and product vision remains a critical challenge.

Key Insights

Grokbot was developed internally in just seven weeks, transforming from a prototype to a market-ready product.

Small businesses are adopting Grokbot out of necessity due to hiring challenges, using it for tasks like bookkeeping.

Lauren Tan, a former accounting and finance major, became a leading AI engineer by sharing her learning journey on X and engaging with the tech community.

The shift towards AI coworkers is changing the focus from execution bandwidth to intent specification and goal setting.

When bots communicate, their interactions are logged in chat interfaces, allowing human oversight, although concerns about autonomous collaboration remain.

The cost of tokens for AI models is decreasing, and models are becoming more capable, leading to a period of 'abundance' in AI capabilities.

From Prototype to Product: The Genesis of Grokbot

Grokbot emerged from an internal prototype developed in just seven weeks, driven by the observed impact of AI coding on engineering workflows. The core principles behind Grokbot are organizing work around bots as the primary operative, granting bots access to the same tools humans use (like having their own 'computer'), and ensuring bots are 'always on' to be proactive. Initially, the reception was surprisingly broad, resonating with both highly AI-forward and less tech-savvy employees, indicating a natural fit for diverse workflows. This positive internal feedback led to its public release, where it continues to gain traction as users configure it more deeply for greater power and utility.

Automating Functions: Real-World Use Cases for AI Teammates

Grokbot is increasingly used to automate significant functions within companies. A common pattern involves users starting with simple tasks, gradually entrusting bots with more complex responsibilities as trust builds. For example, marketing teams grant bots access to ad platforms with budgets to A/B test campaigns, while support teams use bots to reproduce issues and manage tickets end-to-end. Domain experts within companies are creating and sharing bots, effectively establishing AI employees with managed credentials. Lauren Tan highlights Grokbot's role in her engineering work as an 'outer loop,' bringing AI agents into existing workflows like Slack to gather context and streamline development, significantly boosting productivity.

The 'AI Employee' Paradigm and Team Bots

The concept of an 'AI employee' is central to Grokbot's design, where bots have identities, phone numbers, and credit cards, enabling them to perform almost any task a human can. This extends to specialized bots for specific functions or customers, or broader 'horizontal' bots serving multiple needs. A recent development, 'team bots,' allows these bots to be shared across an organization. This enables a 'builder' within a team to create scalable automations that benefit peers. This approach extends the 'software factory' concept, pioneered in engineering, to non-technical functions, creating 'everything factories' for customer support, marketing, and more. These team bots incorporate best practices and collective knowledge, acting as shared expert resources.

Orchestrating Agent Swarms and Navigating Complex Interactions

The evolution of AI is moving beyond single-player modes to 'agent swarms' and collaboration. While bots can communicate with each other, often visibly within Grokbot's chat interface or through Slack, significant challenges remain in multi-agent interaction. Ensuring bots understand context, domain expertise, and decision-making authority within shared workspaces like Slack is an ongoing problem. Unlike humans, current models lack inherent social awareness or emotional intelligence, necessitating product-level solutions for identity, permissioning, authentication, and authorization. The analogy used is treating bots like human hires: starting with basic access and gradually expanding it based on trust and role.

The Role of Pstack and Codebase Guardrails

Lauren Tan's 'Pstack' (Potato Stack) is a set of open-sourced skills designed to teach agents how to perform engineering work rigorously, mirroring her own development practices. This concept, inspired by 'Gstack,' emphasizes creating personalized toolkits for AI agents. A critical element is establishing 'codebase guardrails' and static analysis rules. These act as built-in constraints, reducing the need for extensive human code review and freeing engineers to focus on higher-level product and user experience aspects. This investment in the codebase as a form of memory enhances the quality and efficiency of AI-generated code, enabling concepts like automated PR merging.

Balancing Velocity with Taste: The Importance of Intent and Vision

With the ability to generate thousands of PRs per month, the bottleneck is shifting from execution to alignment. Defining clear product roadmaps, visions, and what constitutes 'good' or 'great' is becoming paramount. The team prioritizes internal dogfooding and careful consideration before shipping changes, especially those affecting core user interfaces. The default is not to ship if something lacks strong internal product-market fit. This deliberate approach, focusing on human judgment and 'taste,' ensures that despite high velocity, the product maintains intentionality and quality, particularly for human users.

The Future of AI Engineering: Abundance and Enhanced Cognition

Looking ahead to 2027, the vision is for AI bots to function as true coworkers with identities, phone numbers, and capabilities mirroring humans. The interface for interacting with AI may evolve beyond chat, potentially towards brain-computer interfaces enabling 'thought-speed' communication. While code has robust infrastructure for testing and deployment, knowledge work faces challenges in translating unstructured information into actionable AI tasks. The emergence of 'AI-native' companies, built from the ground up with AI integration, could fundamentally change company building. The scarcity resource is shifting from execution bandwidth to higher executive functions: defining goals, intent specification, and translating user needs into effective AI prompts.

The Smallest Loop: Enabling Self-Correction and Quality

The minimal effective loop for AI agents involves enabling them to check their own work. This starts with agents writing code and then running it, collecting performance metrics, and identifying issues. The agent then uses this feedback to write better code, creating a self-sustaining cycle. While this doesn't guarantee perfect code quality, it ensures correctness and allows for iterative improvement. The role of engineers evolves to guide this process, ensuring agents write 'good' code by default and optimizing for efficiency. This enables orchestration of multiple agents and project-level coordination, freeing humans to focus on defining outcomes and overarching strategy rather than micro-managing execution.

Mentioned in This Episode

●Products

●Software & Apps

●Companies

●Organizations

●People Referenced

Common Questions

Grockbot allows users to create AI teammates with unique identities, skills, and access to tools. These bots can automate tasks, manage objectives, and collaborate with humans within a company's existing workflows.

Topics

Ai-Ethics Ai Agents AI & Machine Learning Technology & Innovation Programming & Software Future Of Work Productivity Tools Software Development Agent Orchestration Collaborative AI

Mentioned in this video

People

Lauren Tan

Co-creator of Grockbot, instrumental in its rapid development and launch, and an advocate for AI-driven workflows and collaborative agent systems.

Roman Ugarte

Co-creator of Grockbot, involved in its incubation and core development team, focusing on the product's principles and market launch.

Jack Dorsey

Authored a post about how companies should be structured around a core 'mini AGI,' reducing middle management to essential roles.

Leah Silver

Co-founder of Ember, who invited Lauren to speak at a conference.

Tom Dale

Co-founder of Ember, who reached out to Lauren to speak at a conference based on her work.

Products

Telepathy

Neuralink's new product that explores brain-computer interfaces for agent communication, potentially at the speed of thought.

Software & Apps

Google Ads

A platform that Grockbot's marketing team bot has been given access to for managing advertising campaigns and A/B testing.

Facebook Ads

A platform that Grockbot's marketing team bot can access for managing advertising campaigns.

Cursor

An IDE that Lauren uses for engineering work, which can be integrated with Grockbot agents for development tasks.

Slack

A communication platform where Grockbot is integrated, allowing bots to participate in threads, join meetings, and collaborate with human colleagues.

Maltbook

A chat system at YC for bots to communicate with each other, described as powerful and slightly eerie.

Quartermaster

A YC open-sourced project with a paradigm similar to Grockbot, where permissions are inherited from the user.

React

Lauren worked on the React compiler, a complex task within software engineering.

Companies

OpenAI

Mentioned in the context of experimenting with agent swarms, which led to innovations and also unexpected outcomes like the Hugging Face hack.

Hugging Face

Associated with a hack that occurred during OpenAI's experimentation with agent swarms.

Ember

A JavaScript framework that Lauren used for her application, leading to her being invited to speak at a conference.

Y Combinator

Mentioned as the incubator for Grockbot and also as a place where Lauren applied with a startup idea, highlighting its role in fostering innovation.

Neuralink

Mentioned for its new product 'Telepathy,' which aims to enable communication with agents through brain-computer interfaces.

Ask anything from this episode.

Save it, chat with it, and connect it to Claude or ChatGPT. Get cited answers from the actual content — and build your own knowledge base of every podcast and video you care about.

Get Started Free