Dan Martell ordered lunch with seven words. He told GPT Astra to get his usual from DoorDash, and within seven minutes the tool had pulled up his order history, identified the restaurant, discovered his credit card was expired, updated the payment details, and scheduled the delivery for noon. No clicking. No switching apps. No copy-pasting. Just a result. For anyone still manually toggling between browser tabs to get basic business tasks done, that single seven-minute sequence is either a wake-up call or a glimpse of what is already sitting on their desktop unused. Martell, who spent dozens of hours testing every major AI tool across his portfolio of companies, has settled on a two-tool stack he believes puts the people using it in a genuinely different category from those who are not.
Why GPT Astra is nothing like the chatbot you deleted
The version of ChatGPT most people remember sat on top of a computer and waited to be asked something. GPT Astra runs the machine. It opens files, navigates apps, reads documents, and executes workflows while the user is doing something else entirely. Martell describes the experience as watching a ghost employee clicking around inside your computer while you work alongside it.
Setting it up takes three steps: download ChatGPT, go to Settings, then Integrations, and toggle on ‘any app,’ which is the computer-use flag that gives Astra access to the full desktop environment. There is also a floating icon option he calls the ‘show pet’ setting that keeps a voice prompt accessible from inside Slack or any other active window, so the tool is reachable without switching contexts.
To get results out of it, Martell uses what he calls the MAPS framework: Mission (what you are trying to achieve), Ask (the specific action), Parameters (the files, rules, and limits), and Shape (the exact format of the output and how you want to be notified). A real example from his team: gather every receipt from the last 30 days, open each file, rename it to the date and vendor, ignore travel receipts, add up the total, and send it to the chief of staff on Slack. The tool does it. No cron jobs. No API configuration.
His events coordinator Rachel runs eight large events a year by herself using this setup. When she met with events teams of 12 and 25 people, their first question was how she managed the volume alone. She showed them her AI stack.
Grok Bot turns one assistant into a whole department
If GPT Astra is the generalist, Grok Bot is the specialist layer underneath it. Martell frames it as the difference between a general practitioner and a brain surgeon. Built by Elon Musk’s xAI and first tested internally across his own companies, Grok Bot is an agent orchestration platform designed so that anyone can build and deploy specialized bots using plain English, no technical background required.
Each bot gets its own virtual computer, its own role, and its own task list. Martell currently runs 37 bots: a PR bot, a revenue bot, a finance bot that lives inside his Slack, his mailbox, and his finance tools simultaneously, and a chief-of-staff bot that coordinates all the others. To build a new one, a user creates it, names it, assigns a role, and then uses a ‘teach a task’ feature where they perform an action in a browser, hit stop, and the bot learns to replicate that workflow going forward. Martell wrote about this approach in his book under the name the camcorder method: record yourself doing the thing, hand the recording to someone else. Now the someone else is the bot.
As he put it directly: ‘We own the brain and then we rent the models. Whoever’s got the best model, I pay you for it.’
The combined workflow looks like this: GPT Astra acts as the chief of staff sitting in the office, handling whatever it can locally and handing off the rest to the specialized Grok bots running in the cloud. A user tells Astra to create a financial report from the gathered receipts, open Grok Bot, feed those receipts to the finance agent, generate a visual summary of the last 30 days, and send it to the CFO on Slack. The whole sequence runs without the user touching a second application.
The system that makes the next tool irrelevant to worry about
Martell’s third layer is the one he considers most important for long-term leverage: a centralized second brain that every AI tool can connect to on day one. He built a product called Apex for this purpose. It aggregates every meeting, chat, decision, report, and external input, including favorited YouTube videos, into a single knowledge base. When a new AI model emerges and becomes the best option, connecting Apex to it takes minutes, and the model starts with full context rather than as a blank slate.
The practical result is tool agnosticism. Claude, Gemini, a model nobody has heard of yet: it does not matter which one leads in three months because the knowledge does not live inside any of them.
The broken hand and the bot that covered for it
Martell’s right hand is in a cast throughout the entire walkthrough. He punched a tree while mountain biking with his kids, hitting a drop, losing control, and failing to get his hand off the handlebar before the trunk connected. He noted, without irony, that GPT Astra and Grok Bot kept his output running regardless.
The stack he describes is already doing the same for Rachel, who is running eight events a year alone, and for a team member who connected HubSpot to Grok Bot without touching a single setting manually: GPT Astra opened HubSpot, located the configuration fields, pulled the credentials from the password manager, and handed everything to Grok Bot while the person watched.
The seven-minute DoorDash order is still the clearest proof of concept. Seven words in. Lunch on the way.
