Watch on YouTube: https://www.youtube.com/watch?v=1cWZ7Ekff6E
Channel: Greg Isenberg
Using Grokbot as a Personal AI Orchestrator
In a recent conversation, Greg Isenberg interviews Ross Mike about treating Grokbot not as another chatbot thread, but as a personal AI orchestrator. Ross’s central argument is that Grokbot’s contextual bot system and mobile user experience create a superior interface to the internet compared with standard, single-thread AI conversations. Rather than dumping every task into one general assistant, the approach involves spinning up project-specific bots, giving each one focused context, and retiring them when the work is done.
What the Video Covers
The discussion walks through several practical deployments. Engineering work is handled by having Grokbot orchestrate Claude Code Opus 55 sessions through T3 code and Tail Scale, rather than writing code directly inside the chat. Design is locked in Paper—a Figma alternative with MCP integration—before any code is generated, which is intended to prevent “AI slop” and keep a human in the loop. For content creation, a dedicated bot scrapes X twice daily to surface topic ideas, titles, and resources tied to a specific niche. A finance advisor bot connects to Plaid, reviews transactions, and is positioned to block purchases that fall outside stated goals. Grokbot is also described handling real-world commerce: negotiating with vendors, emailing sellers, and completing purchases, with a PS5 Pro bought from eBay offered as an example. Finally, Ross recommends running security reviews using your best model or an “obliterated model” loop, particularly for apps that touch payments or sensitive user data.
Step-by-Step Walkthrough
If you want to adopt this system, the notes suggest the following sequence:
- Create a Grokbot account and set up one bot per serious project instead of using a single general thread. Delete bots when they are no longer useful to avoid bot fatigue.
- Connect Grokbot to Claude Code and use it as an orchestrator for engineering sessions. The Claude Code plan referenced costs $200 per month and includes roughly $8,000 to $12,000 in API token credits.
- Set up Paper and connect its MCP to your agent so designs are reviewed and approved before code is generated.
- Connect Plaid to a finance or advisor bot so it can review bank transactions and flag or block off-goal spending.
- Configure a content bot to scrape X twice daily and deliver video or article ideas, titles, and supporting resources based on your niche.
- Assign a real-world task to your agent, such as negotiating a purchase or emailing a vendor, to test the system outside of pure chat.
- Run security reviews with your best model or an obliterated model loop for any application handling payments or sensitive user data.
- Keep human approval as a required feedback loop for both design and functionality; do not move straight from prompt to deployed code.
Tools and Numbers Named
The conversation references a specific stack and several price points worth noting:
- Grokbot — the central orchestration interface.
- Claude Code — $200/month plan, with approximately $8,000–$12,000 in API token credits.
- Opus 55 — the model used for engineering sessions.
- Paper — a Figma alternative with MCP integration for design review.
- T3 code and Tail Scale — used alongside Claude Code for engineering orchestration.
- Plaid — bank account connection for the finance advisor bot.
- Shopify — mentioned in the context of commerce tasks.
- X/Twitter API — used by the content bot for twice-daily scraping.
- Mobin — referenced as a design resource.
- Bloomberg terminal — cited at roughly $30,000 as a comparison point for a custom trading app.
Caveats and Hype Check
The tone of the conversation is strongly promotional. Phrases such as “by far, in my humble opinion, the best one” and “it seems irresponsible to not pay for that” suggest near-magical productivity gains. No specific revenue or income claims are made, and the notes do not show verified revenue from apps built using this method. One security claim deserves caution: Ross references an “obliterated model” with guardrails removed but says he cannot show it due to “science stuff,” which is an undocumented claim. Readers should treat the security advice as directional rather than proven. Additionally, Greg Isenberg’s channel often features tool recommendations that can be affiliate-adjacent, though no direct affiliate links appear in the transcript.
It’s by far, in my humble opinion, the best one.
Unless you’re in dire financial need, it seems irresponsible to not pay for that and take advantage until the wheels fall off.
Concrete Next Actions
- Sign up for Grokbot and create a dedicated bot for your most active project.
- Evaluate whether the $200/month Claude Code plan fits your engineering budget before connecting it as an orchestrator.
- Set up Paper with MCP integration and make design approval a mandatory step before code generation.
- Connect Plaid to a finance bot and define clear spending rules so it can review transactions effectively.
- Launch a content bot that scrapes X twice daily and outputs topic ideas, titles, and resources for your niche.
- Test a real-world delegation, such as asking your agent to email a vendor or negotiate a purchase.
- Schedule a security review for any payment- or data-sensitive application before deployment.
Taken together, the video presents Grokbot as a project-based operating layer for AI work rather than a replacement for individual judgment. The most defensible takeaway is to isolate context by project, keep humans in the approval loop, and treat expensive tool stacks as experiments until your own workflow proves the return.