The 2 AI Tools Running My Companies Right Now (And Why I Don’t Care When They Change)

One drives my computer, one runs a team of 37 bots, and one layer underneath means I never start from zero when something better shows up.

The first thing I asked it to do was order lunch.

I typed one line: order me my usual from this restaurant on DoorDash. That’s it. It opened DoorDash, went through my order history, figured out what “my usual” meant, and put the order in. Then it hit a wall. My credit card on file had expired.

So it went and updated the card. I’d given it access, it asked me to approve, I said yes, and it scheduled the delivery for noon. The food showed up at noon.

Seven minutes, start to finish, and I never touched a single app.

That was day one with the newest version of ChatGPT, and it’s the reason I’m writing this. Everybody has an opinion on which AI is best right now. Claude, ChatGPT, Gemini, some cheap model out of China nobody’s heard of. And every one of those answers expires in a few months, because the models keep leapfrogging each other.

I’ve spent dozens of hours testing every major tool across my companies. Here’s what I’m running today, how I get the tools working together, and the one thing I built so that whichever tool wins next, I can switch the same afternoon.

ChatGPT Astra: the one that drives your computer

I’ll be honest, I was not a fan of ChatGPT for a while. I hate to say it, but it’s back.

Most AI tools sit on top of your computer like an app. You get an answer, then you copy it, paste it, and carry it into some other tool yourself. GPT Astra runs the actual machine. It opens files, uses apps, and clicks around in the background while you keep working. It’s like watching a ghost employee on your computer.

People ask me if that’s the same as Claude Code or Codex. It’s way different. Those work through APIs and connectors. Astra works the way a person works. Whatever you can do on your computer, it can do.

Here’s what that means for your team: if the person on the receiving end of your work isn’t using a tool like this, they can’t keep up with what you can produce now.

How to set it up:

1. Download ChatGPT again. If you deleted it, go get it back.
2. Turn on computer use. Go to Settings, then Integrations, and toggle on “any app” so it can use every app on your machine.
3. Click “Show Pet.” It puts a small icon with a voice prompt on your screen, so even when you’re deep in Slack you can talk to it and hand off work. It’s the old Microsoft paperclip, except it works.

How to prompt it: use the same framework I’ve been teaching for years. I call it MAPS.

• Mission. What you’re trying to achieve.
• Ask. What you want done, stated very clearly.
• Parameters. The files, the rules, the limits.
• Shape. Exactly what you want back, and how you want to be notified.

The difference now is that you can hand it the whole workflow in one go. Here’s a real one:

I want every receipt from the last month gathered and filed. Open my receipts on my computer, read each one, rename the file to the date and the vendor, ignore any travel receipts, add up the total, and send it to my chief of staff through Slack.

That used to be four tools and a pile of copy and paste. Now you hit enter.

Rachel runs about eight big events for us, as one person. When she meets the top event teams in the industry, teams of 12 and 25, they ask how she does it alone. She just shows them her AI: Astra plus the next tool on this list.

The play that pays for itself: ask it to go through your email, find every subscription notification, and work out what you’re actually paying for. Then have it log into each account as you and negotiate a discount through the support chatbot. My friend Nick told me this saved him $30,000 a month in his business.

You can also talk to it while it runs. Tell it to skip something, change direction, or stop.

If you’re still doing a task the AI could do, you’re the bottleneck. Worse, you’re close to being replaced.

GrokBot: a team of specialists that never clocks out

Astra is the generalist, great at doing things on your local machine. But what about the work that needs to be nerdy, specific, and really good at one thing?

That’s GrokBot. It runs a team of hyper-specialized bots that hand work off to each other to finish entire tasks. Every bot gets its own virtual computer, the same way you’d hand a new hire a laptop on day one. And you set everything up in plain English. “Remind me at 8am to check this Slack post, and if I don’t respond, remind me again.” Done.

If you’ve ever tried to install OpenClaw, Hermes, or another agent platform, you know you basically need a computer science degree, and you’ll probably get hacked along the way. GrokBot is the simplest agent platform I’ve used. If Apple built AI agents, it would feel like this.

Most AI gets you 90% of the way and leaves you to finish. The first time GrokBot takes something to 100%, done the way you would have done it and written in your voice, it feels awesome. After that you just keep handing it more.

I’ve spent more time in GrokBot over the last two weeks than any other AI platform. I held an all-hands and asked everyone on my team to install it within 24 hours. It was built by Elon’s team at xAI, and his own employees tested it first. There’s a reason those companies are so productive.

A couple of the bots I run:

• My sales bot. Overnight, it finds every sales bottleneck and constraint across all my companies, consolidates it into a high-level overview for me, and messages each leader directly to ask if they’re aware and what they’re doing about it.
• My finance bot. It lives in my inbox, my Slack, and my finance tools. It grabs everything and does the work.

Add a PR bot, a sysadmin bot, a revenue bot, and I’m at about 37.

How to set up your first one:

1. Create a bot and give it a role. Most people start with a chief of staff that coordinates the others, or a simple inbox manager.
2. Not sure which bots you need? Ask it. Tell it this:

Interview me. Ask me about the context of my life, or look at my emails and my Slack posts, and tell me which bots I should create.

3. Give each bot a name, a role, and tasks that belong only to that role.
4. Put related bots in a group chat so they share context. It works like text messages, so you treat the bots like people and remember what each one owns.

Now the fun part. I tell my chief of staff bot, “Clear my inbox, categorize all my email, and draft replies for the ones that need a response.” It pulls in my inbox bot, my copywriting bot, and the context bot that knows enough to write accurate replies, and my inbox gets handled. Then I add, “Do this every morning at 8am and give me a brief on what needs my attention.”

No cron jobs, no Vercel instances, no API keys. Is it running on my computer or in the cloud? Who knows.

Pro tip: click the plus icon and choose “Teach a task.” It opens a computer, you do the task yourself in the browser, you hit stop, and the bot learns how to do what you just showed it. If you’ve read my book, you know this as the camcorder method. You record yourself doing the thing, then give the recording to somebody else. Now the somebody else is a bot.

Make Astra the boss of the bots

Here’s where it gets good. Stop picking between the two and have them work together.

ChatGPT Astra is my point of contact, and it runs the bots for me. GrokBot is an app on my computer, so Astra opens it and talks to my bots, which work on their own machines in the cloud, 24/7. If a bot needs something on my laptop, Astra grabs it and hands it off.

Think of it like an office. Astra is the chief of staff sitting next to you, hearing everything you say. It does some of the work and hands the rest to the designer, the copywriter, the social media person, the office admin. Those are your GrokBots.

Most days I say one thing to ChatGPT: use GrokBot, coordinate with my bots to get this done, and report back if you need anything from me.

And you can do it from your phone. In the ChatGPT app there’s a Remote Access option in the left pane. Connect your laptop and you’re talking to your phone while it drives your laptop. I’m on the beach, it’s in the office. Don’t tell anybody.

A real example:

Create a financial report with the receipts we found earlier. Open GrokBot, feed the receipts to my finance agent, have it build a visual report of the last 30 days, and send it to Jared on Slack.

It opens GrokBot, briefs the finance agent, and the report lands in my CFO’s DMs.

Somebody on my team wanted to connect HubSpot to GrokBot, and it was asking for configuration settings he didn’t understand. So he opened Astra and said, “Go see what GrokBot is asking for and connect it. Figure it out.” Astra read what GrokBot needed, used the password manager to log into HubSpot, found the settings, handed them over, and hit submit. He never had to learn a single technical term.

Your job is to be the director. I’ve been saying that for three years.

The part that makes every tool swappable

If you’ve done everything above, you’re already ahead of 99% of people using AI.

But in three months, or three weeks, or three days, something better will come out. People are already on me in my Instagram comments. “What happened to Claude? Why are you so hot and heavy on ChatGPT now?”

I can’t predict the future. I can predict the process.

About a year and a half ago, every new AI tool meant the same wall. Reconnect everything, copy all my knowledge back in, migrate my preferences and memories. I used to make videos teaching people how to migrate. Eventually I thought, there has to be a better way.

Think about what a new tool doesn’t know on day one. Everything you’ve ever said. Your clients. How you make decisions. Your relationships. Who is Mark? If I search your email you probably have 1,700 Marks. Which one matters? A new model starts dumb until you feed it all of that, every single time.

What you need is a second brain. All the context in your head, saved in one place that any AI can connect to. Then a new tool shows up already knowing you.

You can build one yourself with Obsidian, Google Drive, or Notion, and I’ve made whole videos on how. But it takes time, it can cost more in tokens than it gives back, and it breaks. Worse, you might think it’s making good decisions when it’s missing half the context.

None of the AI companies are incentivized to let you own your data. They want you locked in and dependent on their platform.

That’s why I built Apex (apex.host), one of my companies at Martell Ventures. Every meeting, chat, decision, and report lives there. If I favorite a YouTube video, it pulls that in too. And while ChatGPT, Claude, and Gemini keep passing each other for the top spot, I just connect my brain to whichever one is winning and it does work for me out of the box.

It’s Neo in The Matrix. They plug him in and he says, “I know kung fu.” Boom. It just knows.

I own the brain and I rent the models. Whoever has the best model this month, I’ll pay them for it. The brain stays mine.

The mindset underneath all of it

What ties it together is what I call the AI mindset: the habit of handing work to the AI first. I still see it on my own team. People do work by hand because they never stop and say, “Wait, I don’t have to do this.”

You can own a NASA supercomputer and still use it for basic math. That’s what most people are doing with AI right now.

Work alongside these tools and they’ll help you solve the biggest problems in your business. So don’t just play with them. Commit to learning them and hand them the work.

Do that, and you’ll crush it in this new world.

-DM