Most people use an AI chatbot as a smarter search box. Grok is different because it can reach real things: it can browse the web, read what it finds, and drive its own computer to fill in forms and use apps. That gap, between talking about the work and doing it, is what makes it useful for starting a business instead of just planning one.
If you came from the video, that's the idea. Here's the part that doesn't fit in sixty seconds: which parts of a business Grok can carry, what you still have to own, and the briefs I'd paste in. I'll run all of it on one real example so nothing stays abstract.
The one job to give it first
The biggest mistake I see is handing Grok the whole business at once. "Start me a company" gets you a generic plan and nothing you can act on. Give it one job with a clear finish line, watch it do that, then give it the next. Here's the sequence I'd run. Each block is a separate ask, and you check the output before moving on.
Have it research real demand
Before you build anything, ask Grok whether people actually pay for this. Point it at listings, local groups, and reviews, and tell it to quote what it finds. You want the prices real people charge and the complaints they leave, not a confident guess.
Have it shape the offer
Give the research back and ask for three specific offers with prices, based only on what it saw. If it can't tie a price to a real listing it found, that's a sign it's inventing, and you send it back.
Have it build the front door
A booking form, a simple page, or a profile where your customers already are. Grok can draft the copy and, using its own computer, fill in the setup. You do the account and the login yourself, then hand it back.
Have it run the repeatable admin
The scheduling, reminders, follow-ups, and review requests. This is the part that eats your evenings and the part Grok is good at. Wire it to your calendar and messages so it drafts and you approve.
Have it review the numbers weekly
Once money is moving, a standing ask: what came in, what's booked, what's slow, what to try next. Small and specific beats a dashboard you never open.
Run it on one real thing: a dog-walking side hustle
Say you want to walk dogs and offer drop-in pet visits in your neighborhood. It's a good test because the hard parts are demand and admin, not the walking. For research, I'd have Grok pull up profiles within a few miles: what they charge for a 30-minute walk versus a drop-in, and what their reviews complain about. The complaints are the gold. If three profiles get dinged for "never confirmed the booking" or "didn't send photos," you've found your opening, and the offer writes itself: every walk ends with a photo and a note on how they did.
Then the admin, which is where Grok earns its keep. A booking comes in, it checks your calendar, drafts a confirmation, and sets a morning reminder. After the walk it drafts the photo message and, a day later, a review request. You tap approve. Every Friday it reports how many walks, how much came in, which slots stayed empty, and one thing to test. It keeps the numbers honest; you decide what to do about them.
The brief I'd paste in
Here's the first ask, the research one, written the way I'd actually send it. Fill in the brackets and it works for a dog-walking hustle or anything else you're testing. The specifics matter: telling it to quote sources and admit gaps is what keeps it from making things up.
Job: research demand for [my service, e.g. dog walking and drop-in pet visits] in [my area].
Use the web. Look at real providers, listings, and local groups.
Bring me back:
- 5 real providers, with what each one charges and for what.
- The exact complaints from their reviews (quote them, don't summarize).
- What people ask for that no one is offering.
Rules:
- Quote your sources and link them. If you can't find something, say "I couldn't find this," don't guess.
- Don't recommend a price yet. Just show me what's already out there.Once the research holds up, the next brief turns it into the running business. This is the one I'd keep and reuse every week.
You help me run [my business, e.g. a dog-walking service].
My offers and prices: [ ].
How I sound to customers: [warm, short, reliable].
When a booking comes in:
- Check my calendar for a clash and flag it if there is one.
- Draft a confirmation for me to approve.
- Set a reminder for the morning of.
After the job:
- Draft a follow-up message (include a photo if I add one).
- One day later, draft a review request.
Never do without my sign-off: send a message, take a booking, quote a price, or spend money.
Every Friday, give me: bookings this week, money in, empty slots, and one thing to try next.What Grok can carry, and what you still own
Here's the honest line to hold onto. Grok can carry the research, the drafting, and the repeatable admin. It cannot carry the judgment, the trust, or the actual service. If you're walking dogs, the walk is yours, and so is every decision a customer would feel. Knowing which is which is the whole skill. My guide on which parts of your business should AI handle goes deeper on drawing that line.
What goes wrong, and what to watch for
It invents demand that isn't there. Ask it to research a market and it may return tidy numbers and confident prices that trace to nothing. That's not lying, it's filling a gap you left open. The fix is in the brief: make it quote and link real listings, and treat any figure without a source as a draft. If it can't show you where a price came from, it doesn't count.
You hand it the login and forget to take it back. When Grok uses its own computer to set up a profile, you're signing into your real accounts inside a window it can see. Do the sign-in yourself, let it do the one narrow task, and close that access when it's done. Never leave a payment account or bank login open to an agent working on its own.
It sends something before you've read it. The expensive mistakes are the ones that reach a customer: a wrong price, a booking you can't honor, a message in the wrong tone. Keep send, book, quote, and pay as things it drafts and you approve, and put that block in the tool's actual permissions, not just the prompt. A prompt can be talked around; a permission can't. This is the same lesson as building your first AI agent.
You automate before there's anything to automate. It's tempting to spend the first week wiring up a beautiful system. Don't. Get one real customer, do the job by hand, and only hand Grok the parts that are already boring and repeatable. Automating a business you haven't proven is just a tidier way to have no sales.
Grok can take a real weight off the admin and research. It can't make people want what you're selling, and it can't be the reason they trust you. A hands-off business that also makes money is mostly a fantasy: the tools that run themselves are the boring middle, and the parts that win customers still need a person. Keep yourself in the loop where it counts.