After some initial nonsense where it wouldn’t accept any of my cards (declined multiple times, on multiple cards, and then a few days later it just worked with no explanation), I’ve finally signed up for Grok Bot and waded into the world of personal agents.

For some background on me, I’ve never used Cursor or Grok, but I do like Twitter. I’m drawn to it over things like Hermes and OpenClaw because I don’t like to faff around with self hosting, configuration, or settings. Grok Bot looked like the most polished option out there.

My plan was to point it at work first, since that’s where I do most of my computer stuff, but perhaps I was wrong to focus there.

Use case⌗

I’ll admit up front that I don’t really have a use case for these things. I have a pretty simple life, by design. I run my own software development agency. I have a small number of clients at any one time. I work on client projects day to day, and in between I build my own products like Rivets and Jellybook.

Most of that lives in Slack, Linear and my inbox, so that seemed like the obvious place to start.

Initial setup⌗

Setup was painless. It guided me through creating multiple bots and connecting the main tools. I ended up with:

  • Bagman, my “chief of staff”. This is the one I talk to, and it relays things to the others
  • Frontier, for my agency work
  • Eversource, for my primary client
  • Family, for home stuff
  • Jellybook, for marketing and leads

The idea is that each bot looks after its own corner, and Bagman sits on top, so I can tell it what I want and let it work out who should handle it.

One thing I really like is that it’s both a desktop app and an iOS app, so I can use it from anywhere. The iOS app in particular is really good and polished, which isn’t a given. It’s also really cool and useful being able to see the bot’s screen while it works.

Getting more involved⌗

So next I thought.. okay, what else can we do here? It can’t sync to my Whoop and get health/fitness info - it offered to use a community-provided Whoop MCP, but I’d have to register an app in their developer dashboard, which doesn’t interest me. No go there. It can’t control my Hue lights, so no go there.. I had to keep thinking.

Ah! I’ve recently sent an email to the CEO of ITFC, asking if there’s any way he can help me secure some tickets for a game while I’m in the UK soon. I asked Bagman to help me with that:

I recently sent an email to Mark Ashton at ITFC asking for his help getting tickets to a game during my upcoming trip to the UK. Get an agent to read that and understand it, and monitor my inbox for any response from him or ITFC, and alert me as soon as anything comes up. If I don’t hear anything, I’ll want to send a follow up, so remind me if I’ve heard nothing in a few days

It delegated, read the email, and set up an hourly check and a reminder for me in a week if I haven’t heard anything. That’s exactly what I’d have hoped it would do. (As an update: Mark did reply, and they got us tickets! Grok Bot didn’t really help with this but I was very happy when I saw its update that they had replied!)

Next, I asked my Eversource bot to send me a daily digest of anything I might have missed across Slack and Linear at 9:20am before I start, so I can read it while I make a coffee. It very helpfully pointed out that I work different hours two days a week, and asked if I’d like a second digest before I start at those times too - yes, sounds great.

The next day⌗

The very first morning, I got.. zero notifications. It didn’t run the digest for me. When I asked why:

It did fire at about 9:22am — the run failed rather than skipping. Digging into why, and I’ll pull today’s digest for you now.

Connectors are fine — looks like a platform runtime failure on the scheduled run, not a missed trigger. Building today’s digest from Slack and Linear now.

This is typical AI nonsense - just ignore the massive, fatal flaw and just move on. When asked to make sure it happens, it’s effectively saying it’s inherently unreliable and might not work again:

Grok Bot explaining that the scheduled digest failed and it can’t guarantee the next one will run

This.. is not great. The whole point of a morning digest is that I don’t have to think about it. If I have to check whether it ran, I may as well just check Slack and Linear myself.

To be fair, it’s run every day since. The problem now is timing: it seems like it starts at 9:20am, but takes 5-10 minutes to actually pull everything together, so it lands after I’ve already sat down and started reading. The obvious fix is to schedule it earlier, but it’s the kind of thing I’d have expected it to figure out itself.

Jellybook⌗

I saw a post from Cristina Cordova about using Grok Bot with X - I don’t think our competitors do anything with X, so this is potentially an interesting area for me to watch.

I sent the following prompt on a freshly created Bot to get us set up:

Your job is to help me with my Jellybook project. Jellybook (https://myjellybook.com) is a collector app for Jellycats. I want you to help me with social media and marketing mostly, but also assist with tracking feature requests and changes in Linear - I already have Frontier connected to Linear and there’s a Jellybook project in there that you can read and write to when I ask.

Firstly, I want you to start monitoring X for people looking to track their Jellycat collection. Our nearest competitor is Jelly Journal - they’re a basic web app, though they do have a decent sized community. We work on web and, most importantly, offer a completely native, beautiful, polished iOS app

It told me it would monitor Reddit, but couldn’t interact with X without credentials. So, I signed up for Jellybook’s own X account and put the details in the shared 1Password vault. From there, it’s signed in and ran a bunch of keyword searches, and will do them periodically and prompt me about them.

It didn’t find anything useful initially, so I got it to widen its search somewhat:

Can we expand somewhat so we get some more matches? I don’t want to blast spam at people but I do want to find people who might be interested even if they’re not expressly asking for it. Anything about having a large or growing collection or an addiction to Jellycats would still be useful leads!

This time, we got some real leads across both X and Reddit, and made a couple of posts off the back of them. Not a flood, but it’s the kind of slow background work I’d never prioritise myself. It was really cool that it was able to run a series of searches for me in a real browser (I watched the screen the first time), and then help me draft a couple of short replies, and then post a couple of intro tweets about the app too.

Closing thoughts⌗

Overall, my impressions were right - Grok Bot is pretty polished, and it does work pretty well. I have no doubt I’m not using it to its fullest capabilities, but that’s because at work, for me at least, it feels like a solution looking for a problem.

It’s unexpected that it doesn’t have native, even privileged access to X data - you’d think that would be a significant differentiator that they would lean on. I have no doubt that Muse gets special access to everything Meta can think of!

I’m going to keep it for another month and see how it goes. I’m not blown away, but I think that’s because I’ve been pointing it at the wrong part of my life. My work already runs smoothly, and I have tools I trust for it. My personal life is where I’ve barely tried, and that’s probably where I need to embrace it more.

The first thing I want to try there: we do our grocery shopping across two different shops, so I’m planning to give it access to both, have it take a photo of my fiancée’s shopping list, and have it reconcile that into a proper order, translating ‘cheese’ to the same block of cheese we buy every time. That would be super useful and would save her a tonne of time. It also feels like exactly the kind of thing that’ll be fragile and unusable, so we’ll see.

If anyone reading has any suggestions on how I’m failing to use this properly, please get in touch!