I’m always chasing better tools. Notes apps, workflows, optimizations, and now, personal agents.
It’s an easy trap to fall into - always building the “custom” thing and sacrificing the work as a result. There are some amazing tools out there - but there are also some amazing memes about us endless optimizers.
The problem is that endless optimizations and custom setups can distract from the real magic: doing the work.
So far every “personal agent” has felt the same - configuring, provisioning, delegating, and triaging instead of doing useful things.
I’ve spent more time debugging flaky deployments or ignoring weekly iMessage check-ins than I did using all of these new tools that are supposed to make life easier.
And all I really wanted was an agent that could do the same things I do - the manual things that live on my computer - without configuring a remote server, managing service accounts, or creating security headaches.
So when I joined @Cursor_ai, I was surprised to find a new tool designed not for coding, but for general knowledge work.
https://x.com/bot/status/2087224798078517251
It’s called Grok Bot and it’s an agent with a computer.
It's the first tool I've used that feels how personal agents should - magical.
An agent with a computer
Each Grok Bot is an agent with a computer. It runs in the cloud, so it’s always on. It can do anything I can: navigate web pages, open apps, access the file system, and take screenshots.
You might be asking “why can’t it run on my computer?” Well, Grok’s computer becomes your computer as you use it… and it can access your local files if you’re chatting via desktop.
Technically, Grok Bot runs on a persistent Linux VM (if you care about that sort of thing). Practically, it feels like a really fast “always on” remote desktop. Every “bot” you create lives there.
They can use its files, browser, and sessions. If you sign in on a website, any of your bots can use that session until it expires. If a bot needs help, it will ask you.
Hand off
When a bot hits a wall only I can clear (a login, SSO, 2FA, a captcha, a payment) it hands me the computer. I do the hard thing, then I give the computer back.
All the sessions are saved on my machine. If I sign in to a site, any Grok Bot agent can access it for me, whenever.
For simple authentication (API keys, text entry) my bots can send a secure form to fill out. It’s been really nice for configuring MCP servers and API keys.
I can also "record" a workflow: show Grok Bot the task and it saves the steps. Then, the bot can use that workflow as a routine, on a schedule or when something happens in Slack or GitHub.
Our GTM teams have built some pretty crazy lead optimization and routing workflows - more on how I’m using Grok Bot later.
Memory
Grok Bot has access to long-term memory, which makes it feel like collaborating with a partner. This works in three layers: user, agent, & project memory
- User: Each bot knows the basics about you - your name, timezone, and preferences. Any bot can update these values.
- Agent: Each bot has its own memory - a profile file (think AGENTS(.)md: a top-level definition for that bot), plus a log of interactions.
- Project: Bots can join projects and keep project memory there: decisions, conventions, the stuff that belongs to the work rather than to one teammate.
But despite the memory layers, chat feels fast - I’ve been surprised at just how snappy the bot feels.
You can even use a multi-agent approach to solving problems with bots - an orchestrator bot can collaborate with other bots in a Project.
Running bots
But when do these bots run? Well… when you tell them to!
Bots are triggered when you send them a message, but they also fire on routines. I usually just ask a bot to set up a trigger. But you can also fine-tune triggers manually.
This can be anything from a new slack message, a git event, or on a schedule.
Bots can also trigger (and be triggered) by other bots (composability). They can even work in group threads, or “Projects.”
There's a little moment of joy the first time one asks another for help.
And similarly the first time a bot “thumbs up” one of your messages.
For coding work, bots can fire up Cursor Cloud Agents - since Grok agents are tightly coupled with Cursor, the experience is first-party.
The GitHub MCP is also straightforward to configure, which allows my bots to run gh commands from the cli and spin up new repos.
Access
Grok Bot uses the same plugins, connectors, and skills as Cursor. That means I get connectors for all the basics out of the box: Notion, Slack, GitHub, and my team's MCP servers.
This made it very easy to set up all of my context sources - most of which were already configured through Cursor. It also made it easy to hit the ground running with some really great Cursor + Grok Bots.
Safety
This was the biggest mental hurdle: would Grok Bot do what I asked it - and how do I know? The answer is mostly trust.
If you log into Amazon with an agent that has access to a computer, technically it can buy whatever it wants - the same way a human could.
The agent is kept in check by permissions, a reviewer, and allow/block lists.
You write the rules in Settings > General > Agent, in natural language. Under the hood, a separate review agent checks proposed actions and can allow, block, or escalate to you.
Allow and block lists steer that reviewer - you’re trusting the model to follow what you wrote, but all of this takes place in an isolated environment.
Most of us are used to defining agent rules in code or JSON. With Grok Bot, the rules are basically a prompt.
This feels sketchy at first, but we’re in the “trusting agents era” - we need to condition ourselves that rules imposed on agents will hold.
In my testing, I’ve yet to observe adverse behavior. Recent research suggests we’re at the point where we should trust agents more than we trust ourselves.
That’s a tad harrowing, but also pretty cool.
How I’m using Grok Bot
I’ve found a few different uses for Grok Bot across all areas of my life.
Demo Bot is my favorite. Each day, it goes through my @X bookmarks and finds a cool new technology, maybe an npm package or an agent skill.
Then it drafts a prompt using my writing skill and runs it by me.
If I approve the prompt, it uses my project-planning skill to fire up a new Cursor Cloud agent in a tech-demos repo.
In 15 minutes, I have a working prototype available in my Cursor app. I bind the port and play with it.
The point of this bot is to remove the friction from trying new things. Every agent validates work via screen recordings, so I can be sure it is doing what it should.
Grok Bot also helps me keep up with Cursor, find content ideas, and explore new technology.
My Content Bot scans engineering and product Slack channels hourly for small ships. It alerts me in a new chat with suggested social language. It also has access to the Typefully MCP, so I can quickly turn that language into a draft.
I want to be the first to know about the little things that make Cursor great and I want it to be easier to share those things with you all.
My Product Bot is a slower, more detailed version of Content Bot.
Its job is to keep me up to speed on all things Cursor product. It monitors the bigger announcement channels once per day and posts a general update, so I do not have to dig through Slack.
Having some fun
I don’t mind grocery shopping, but I disdain UIs for grocery delivery.
My grocery delivery agent looks across Instacart and Amazon delivery to compare products, quantities, and prices. It manages two carts and compares delivery costs. Then it prompts me to place an order each Friday for Saturday delivery.
I was genuinely surprised at how well it navigated these interfaces.
Much of my experience with personal workflows is navigating high-cognitive-load UIs and comparing inconsistent experiences across sites.
Grok Bot eliminates that work entirely, which brings me to my next agent.
The Cursor team uses DoorDash a lot, and there are frequently group orders across many channels.
This is overwhelming because I am both inept at DoorDash and easily prone to getting nerd sniped by dinner optimization. Do I get a rice bowl with chicken, or a plate of chicken with a side of rice? How do I join a group order? What should I get?
My DoorDash agent watches Slack for drd.sh mentions, then alerts me when there is an order.
I can ask it to dig into a menu that I like. It gives me options, then links me exactly where I need to join the order and add my food.
Building trust
Admittedly, it took me a while to trust Grok Bot.
After a few years in AI, most “do it all” tools do, well, relatively little. Furthermore, I put a lot of trust in a system I didn't quite understand at first.
But any novel product needs to earn its trust - in the past few weeks Grok Bot has done just that.
We’re at the point where we can put more trust in these systems - and trusting Grok Bot has proven some really amazing things.
It feels like the future is finally here. Not a future of replacing human work, but a future of simplifying interfaces and eliminating tedium. A future where everything is accessible to those who ask.
https://x.com/elonmusk/status/1906096888560165107
Everything is computer, and so is Grok Bot. It turns out, that’s all I really wanted.
Follow along at @Bot and give it a shot :)
