We Built Grok Bot. Here Are Our 14 Best Bots | Peng Zheng & Lauren Tan

Summary

Peng Zheng and Lauren Tan demonstrate how they use Grok Bot as a system of persistent, specialized agents for work and personal tasks. Rather than treating each chat as disposable, they give bots distinct roles, tools, skills, and memories, then create project-specific or group chats where those bots can collaborate.

Peng’s Bots and Workflows

• Chief-of-staff bot: Routes tasks to the right bot and chains actions together. Examples include buying 3D-printing filament and updating a Notion inventory, or researching, writing, and posting a Facebook Marketplace listing—with a plan to lower the price over time if it does not sell.
• Email and calendar tasks: The bot can identify inbox tasks, assist with payments, and create or edit calendar events and bookings, sometimes by operating through a logged-in interface.
• Designer and writer bots: The designer helps brainstorm and modify Figma designs through an MCP connector, using the project’s design system and tokens. Peng may create an initial keyframe or system, then have the bot expand it into more screens. The writer edits emails and Slack messages to match his needs.
• Bots in group chats: A PM, designer, and engineer bot can discuss a product idea from their respective roles. Users can address a specific bot with an @ mention or let bots decide whether their input is relevant.
• Creative publishing pipeline: Peng describes turning a street photo into a stylized 3D miniature and developing a workflow that takes a photo and location details, generates day- and night-mode images, removes the background, and publishes the result to his website. He uses the bot as a lightweight publishing interface instead of building a CMS.
• Other uses: A podcast bot converts article links into audio, illustrating how bots can bridge different content formats.

Lauren’s Engineering and Personal-Assistant Workflows

• Computer and virtual-machine control: A local execution daemon lets an agent operate on Lauren’s computer, subject to granted permissions. She uses this to test and improve support for a Linux distribution in a virtual machine, and to find and remove files taking up disk space.
• Testing agents’ work: Lauren uses cloud agents to evaluate changes to her plugin and skill stack. An assessment process checks whether a change achieves its intended goal before accepting it.
• Engineering delegation: A chief of staff passes work to a technical lead, who breaks tasks down and delegates them to engineer bots and cloud agents. Different models and reasoning levels can be chosen for different tasks. The team’s codebase is designed to be agent-friendly, enabling some pull requests to be reviewed and merged with limited direct inspection.
• Dr. Eggbot: A bot that designs and audits other bots. It reviews bot setups and conversation transcripts, then recommends new bots, skills, or procedure changes. It also considers whether frequently triggered routines are unnecessarily costly.
• Personal assistant: Lauren’s assistant booked a multi-city work trip—including flights and hotels—after she provided conference details. It presented options for her approval before booking. She also uses it for shopping and food ordering, and it can send text reminders.
• Social-media bot: A bot checks X mentions and gathers relevant feedback, such as bug reports or user confusion, which Lauren can pass to engineering bots without manually copying the context.

Product Concepts and Practical Guidance

• Agents as reusable roles: Grok Bot separates reusable capabilities and memory into persistent bot identities, while allowing separate chats for individual projects. Bots can collaborate and pass context between tasks.
• A more familiar interface: The guests describe the product as an attempt to make working with agents feel more like delegating to people with recognizable roles, rather than requiring users to understand model-focused technical concepts.
• Build trust gradually: Lauren recommends observing a bot perform a task, correcting it, and turning the refined process into a reusable skill. Once it performs reliably, users can automate it as a routine. Both guests compare this to training and onboarding an employee: start with simple tasks, test the bot’s limits, then grant more autonomy.
• Keep approval where it matters: Lauren’s travel workflow includes a human confirmation before purchases. The examples show that autonomy can be adjusted according to task risk and user confidence.

Guides and Demonstrations

The video includes practical demonstrations of bot setup, multi-agent group chats, design and publishing workflows, computer-control and engineering tasks, travel booking, and a gradual method for building reusable skills and routines. It does not present a formal product review or benchmark.

Main Speakers and Sources

Peter Yang (host/interviewer); Peng Zheng (design, Grok Bot); Lauren Tan (engineering, Grok Bot).