"Game changer!" "My new obsession!" "Mind blowing!"
If you use X and your algorithm feeds you an endless stream of AI content, you've probably seen these comments or similar when it comes to users posting about Grok Bot.
It's the new agent tool released August 11 by SpaceXAI that has taken the AI world by storm.
(SpaceX absorbed xAI earlier this year so the company is now called SpaceXAI)
I was intrigued by Grok Bot, so I tried it. Below I will tell you how my first 24 hours went.
But before I go any further, the basics: What is it?
Grok Bot is an AI "teammate" (commonly called an AI agent) that signs into your apps on its own cloud computer and takes action on your behalf. Engineers have been building agents since around 2023. But Grok Bot is one of the first pitched squarely at "regular people" who don't code for a living. It's a ready-made, "plug-and-play" agent for normies.
How does it compare to other AI tools?
Unlike chatbots like Claude, Gemini or ChatGPT, where you type a question and wait for an answer, Grok Bot can autonomously operate your apps on its own, like Gmail, LinkedIn and other sites, and keep working after you close your laptop. (It sifted through tens of thousands of my emails in minutes. More on that later).
So with all the Grok Bot hype on X and elsewhere I decided to try it.
(Disclaimer: This review is only from my first 24 hours of use. It's not intended to be the definitive review).
Onboarding
Signing up was the easy part. Figuring out by which platform was the part that made my head want to explode. Get access to Bot by … Signing up to Grok Super Heavy, the top Grok tier? An existing X Premium+ account? Super Grok or Super Grok Plus tiers? What if you subscribe to two of these already, how do you link accounts?
And there's more: SpaceXAI recently acquired Cursor (the AI code editor that writes and edits software), and Grok Bot is their first collaboration, so there is access to Bot if you sign up for certain tiers of Cursor. But which ones? And is Cursor worth it?
It was so confusing I had to ask Grok 4.5 for help. It even said the process could be frustrating and confusing. Not direct quotes, but close. You get the picture. For a company that can send a rocket to space and perfectly land it back on earth, it should not be this hard!
(Since I wrote this, they've simplified access and dropped the entry price, which tells you they knew it was a mess too.)
After you find out what path is best for you to get access the actual signup and onboarding process was easy. The Bot interface (both on desktop and app) is clean and very user friendly. It's a genuinely pleasant product to use.
The first 24 hours
The good…
After I set up my account, my first use of Grok Bot was to have it do a "triage" of my personal email. And with more than 140,000 emails, I needed it. Grok 4.5 confirmed this would be a good first task. So I created a Bot, named it "Subscription Auditor" and the directions I gave it were to search back in my personal email the last 6-12 months and produce a clean, organized overview of everything I subscribe to. Do not unsubscribe, cancel, or change anything, I ordered. Just show me the list and recommendations. And then I let it loose and walked away from my laptop.
After about 15 minutes it produced a spreadsheet of all my paid newsletters/Substacks, how much I pay, in what frequency and when they expire, plus all recurring other monthly items I pay for that pass through my email, and recurring "other" marketing emails. It then gave me a list of what I have and have not opened recently, and gave me suggested items I could safely delete and/or unsubscribe from. And … said: "Tell me which to unsubscribe or cancel and I'll do only those." This is the power of the Grok Bot: it not only tells you what the choke points are, but then says it will unchoke it for you.
Then I took it a step further, and instructed Bot to go back into the email, and do a sweep on the past 30 days and drill down on all the AI-related newsletters I subscribe to (there are about seven) and provide me a succinct summary of each one. And also key quotes from each one. And on what day. It did. It took about 10-15 minutes.
It was pretty amazing stuff. Doing this sort of audit of my inbox would have taken me many hours, maybe days. Grok Bot did it in a matter of minutes, following instructions perfectly and speaking to me in clear and concise language.
So I drink a lot of espresso. I subscribe to Illy espresso pods that are delivered to my house. I experimented more, asking the Bot to search my email and tell me in 2025 how much I spent on Illy pods. Less than four minutes later Bot came back with an answer: $1,045.01 and every date I ordered a new batch, every order number and what each charge was.
I then asked Bot to do the same, but to check how much I spent at my local wine store in 2025. Again, less than four minutes later it came back with the answer ($219.25), but also the date of purchase, exact price, and exactly what bottle of wine I bought.
If I was a small business owner I could see how having this scan emails for orders etc could be powerful tool.
One other thing I like: Mid task you can keep feeding the Bot instructions and it will read, respond and execute while continuing the current task.
The … not so good …
Remember how I said Bot told me, "Tell me which (newsletters or subscriptions) to unsubscribe or cancel and I'll do only those." I decided to take it up on that offer. I picked SeeClickFix emails in my inbox (they are citizen complaint automated messages from the city I live in; "So and so is requesting the pothole get fixed" etc). I have thousands in my inbox dating back many years. I instructed Bot to go through my email and delete all of them to start. It said, "sure, on it." I closed my laptop and left. I monitored progress on the Bot app on my phone. I assumed it would take a few minutes. I was wrong. It was working for a long time. Like hours. I thought it froze. It didn't. I was less than 24 hours into experimenting with Bot and it had consumed nearly half my weekly usage. I asked Bot how much longer until it was done. It said it had deleted 2,623 emails but still had many thousands more to go. Ouch! Lesson: Grok Bot is token hungry on labor intensive, mundane tasks.
I decided to stop the mass email delete experiment. I could do it myself in Gmail. Lesson: Bot is great at diagnosing your inbox and scanning for information and condensing it. Making it delete thousands of individual emails is not what it's for.
So then I asked Grok for other ways to test out Bot. It suggested LinkedIn. There is a lot of interesting content in my LinkedIn feed but I rarely have time to save it and read it later. So I created a new Bot with exactly this task: browse my LinkedIn and help me organize what I might find most useful for reading later. (I specifically told it not to like, comment, connect, message or post anything. Read only. This is very important when testing Grok Bot).
This is an "easy" Bot task, so I wanted to see the results.
It failed. It could never get off the starting line. In order for the Bot to get access to LinkedIn it opens the website in the cloud and you have to sign in, and then the Bot takes over from there. I was able to sign in to the first part of LinkedIn, but then came a CAPTCHA, the prove-you-are-human puzzle of "click all the boxes that show a traffic signal." It failed every time. I troubleshot with Grok and it basically said those image puzzles are hard to complete over remote control, and suggested a couple workarounds. I tried them. They didn't work. There is a way around this, but I didn't have the time or headspace to figure it out. So the Grok Bot failed. To be fair, this is not entirely a Grok flaw. Much of the web is built to block exactly this kind of automated access, and agents are running head-first into those defenses.
Then I took a screenshot of one LinkedIn account I follow and told Bot once it got access to my LinkedIn to run a scan of the person's content for me. Bot didn't wait to get access. It immediately crawled the web and found the person and gave me a summary of his work and writings. But that I could have done in Claude, Grok, or ChatGPT, so there was no value added to having Bot do it. As of this writing, I have yet to be able to link Bot to LinkedIn.
In conclusion…
In 24 hours I got a peek of where Grok Bot meets or exceeds expectations and where it falls short. It's a technology with incredible potential if used with proper safeguards. I plan to keep testing it as 24 hours is very much first impressions analysis.
Is it a "game-changer" or just a very fast research assistant with a login problem?
I hate to end on a cliche, but as we say in the news business, "stay tuned." But given Grok Bot has SpaceX engineering behind it coupled with a ton of excess compute to handle demand, I would not bet against it.
PS:
As a working journalist, you might be wondering if I see any Grok Bot use cases for newsrooms? Easy answer: No. Maybe one day, but not right now.