Don't Pick the Wrong One (Grok Bot vs Muse vs Dots)

Johnny Nel | AI for Founders· 15 min· 2,791 words· 13 min read· English ·Watch on YouTube

This is the full transcript of Don't Pick the Wrong One (Grok Bot vs Muse vs Dots), published on YouTube by Johnny Nel | AI for Founders. Every paragraph carries the moment it was spoken, so you can click any line to jump straight to that point in the video, search the whole thing for a word, or copy it out.

0:00Grockbot, Meta's Muse, and Open AAI's Dots all landed inside seven weeks. And if you choose one of the headlines, you'll most probably choose wrong. I run my business on a team of 26 agents, and two of these three are on my phone right now. So, I'm taking each one apart. What's inside, what it costs, and what it won't do, and then you pick for yourself. Nobody's paying me for this, and every price you see is dated because some have already moved since launch. By the end, you'll know which agent suits your work. You'll see all three side by

0:33side with their real prices and limits. And your first move is to check where your work already lives because that alone usually narrows it to one agent [music] before you spend anything. Hey, I'm Johnny Nel. I spent years building businesses the old hard way, managing teams, scaling operations, and doing everything manually. Today everything I build from automations and no code infrastructures to fully AI SAS products [music] has AI helping in one way or

1:00another. Let me quickly show you this. In South Africa we call this a bri. I build with AI tools that in my opinion anyone can learn to use. My mission is simple to help creators, entrepreneurs and business owners get clarity with AI. Let's build smarter together. So guys, before we open any of them up, the way I read it, these three were built for three different jobs. And which one fits you comes down to where your work already lives. Grockbot is a team of named agents you assign work to. Meta's Muse is one agent that looks after your life and now reaches into small business tools. And DOTS is one agent with

1:37OpenAI's flagship model that lives inside ChatGBT. Let's clear away what all three have in common first, because then you only have to think about the parts that are different. Each one works on a computer in the cloud with its own browser, which basically means it isn't borrowing your laptop. It keeps working on your task after you close the app, and it connects to the apps you already use. All three remember you between conversations, and all three can be set to stop and ask you before they send something or spend your money. You might be wondering, Johnny, if the base is the same, why not just grab the cheapest one? The shared parts

2:16can't tell you which agent suits your work, so the price tag is the wrong place to start. What separates these three is how each one hands work around and where it lives. Now, you'd expect somebody to have tested them head-to-head by now, and as far as I can find, nobody has done that independently yet. A comparison from Kiny AI put it plainly saying the public evidence does not establish an overall performance winner. And when Data Camp looked on launch day, they couldn't find a single independent review of DOTS. So that's the shared base, guys. And because nobody can hand you a winner, you choose by which one fits where your work

2:53already lives. Meta's Muse is the easiest of the three to picture. So let's open that one up first. It launched on the 8th of September and you get one Muse that you name and design yourself. You talk to it in the Muse app or straight inside WhatsApp, the same way you'd message a person. I've got Muse on my phone and the thing I've used it for is personal assistant work. Under the hood, it runs on Meta's Muse Spark model with its own secure computer. And even though it feels like one agent to you, Meta's research team says it launches swarms of sub agents to get your work done. Everything it learns

3:29about you goes into a memory you can open and edit, and you can tell it to forget something specific. In my opinion, being able to read and correct what it remembers about you is the most useful control in the whole app. On the money side, you can start for free with a weekly limit. Then it's $20 a month

3:46for the power plan or 100 for maximum. The catch is that Meta measures your usage in Muse tokens and doesn't say how many tokens a typical task uses. So, you can't work out in advance what your $20 covers. Where it shines is the personal assistant work. One reviewer at AI Agents Library ran 10 real business tasks through it and they scored meeting prep 10 out of 10 and a weekly brief nine. On the 29th of September, Meta launched Muse for small business. So, it now connects to Shopify, QuickBooks, High Level, etc. It does have limits and a big one is Amazon. Since the 20th of September, Amazon has been blocking Muse

4:27and Amazon says that's because Meta didn't notify them and the agent doesn't identify itself. That same AI agents library reviewer saw tasks stop dead where Muse asked for a password and a reviewer at AI Marketing Focus found its brand voice drafts came back generic. And you might be thinking, Johnny, why would I hand Meta my inbox? Meta's answer is a separate agent called Sentinel, which approves anything that reaches the internet. So, Muse itself only ever sees standin credentials, never your real passwords or payment details, and every action it takes lands in an audit trail you can read. Bear in mind, that's Meta describing its own design. The Verge has reported claims

5:08that Muse could see message contents without the permission switched on. So, I'd wait on connecting the inbox that matters most to you. If your day runs through WhatsApp and Instagram, Muse is the easy start. And you can try it for nothing before you decide. So, that's Muse. One agent that knows your life and

5:26is starting to help with your business. Grockbot works the other way around because instead of one agent you talk to, SpaceX gives you several named agents that you assign work to. It came out on the 11th of August and every bot you make is a named teammate that keeps its own memory of your work. All of your bots share one cloud computer and its files. SpaceXI says their own people run multiple bots in parallel with one managing the others. And in their words, a chief of staff sits on top with a specialist for each lane. Grockbot is the other one on my phone. And the thing

6:02I've used it for is the orchestration where several agents run and keep going. It's essentially one agent on top handing the work down to the specialists. The way I think about it is like producing a track where the chief of staff is the producer and each specialist is one instrument on its own channel. In practice, that means only the bots a job needs are running, and the chief of staff decides which one goes next. You can teach a bot by asking it to follow along while you do a task

6:30once, and it saves that as a routine. That routine can run on a schedule or the moment a Slack message lands, even with your laptop closed. And on the 28th of September, they added Team Bots, which is one shared bot for your whole team with its own Slack handle. For now, that's a beta on the Teams and enterprise plans only. There's no Grockbot subscription of its own because it comes inside other plans and the cheapest way in is Cursor Pro at $20 a month. At launch, the cheapest way in for one person was 200 a month. So, dude, that price came down a long way on

7:04the 26th of August. SpaceX's docs now spell out where your login live. And according to Kodera's guide, your usage comes as a weekly allowance that cursor doesn't publish the size of. Once you go past that allowance, you're on pay as you go. And Kodacera says a bot that's in the middle of a task can keep going past the monthly cap on that pay as you go spending. That's definitely something to take into consideration before you set 10 of them loose. The limit I'd flag for you is security because all your bots share that one computer with the

7:36same cookies and files and login. SpaceXI's own docs say it in so many words. Do not use separate bots as a security boundary. So, if one bot can reach your bank login, you should assume they all can. Now, I know some of you are thinking, Johnny, why not give every bot its own account? You could, but then they lose the shared files and browser sessions they pass work through. So, the safer move is the one SpaceX AI's docs recommend, which is keeping anything that sends or spends behind your approval. If you've got several lanes of work that need to hand off to each other, Grockbot is built for that. And

8:13that's Grockbot. The one of the three where the handoffs between agents are the whole design. And guys, a team like that only works when every agent already knows its job, how you sound, and what it must never send. Writing all of that from scratch is the slow part. And I've already been through it. So, inside the AI Founders Vault, you'll find the same 26 ready-made agents I run stacked into six lanes with their skills already written. Each of those agents runs on Grockbot or Meta's Muse or Dots and on

8:46an open-source harness such as Hermes. You'll find a community of founders building their own and help when you set up yours. The link is down in the description below. Okay, let's pick the comparison back up with Dots, the newest of the three. OpenAI launched Dots on the 29th of September at dev day. So, it's 2 days old as I'm making this. And what I'm telling you here comes from OpenAI's own pages and the launch coverage. You start with one primary dot that you name and make your own. And it runs on GPT6 Astra, which is OpenAI's flagship model. It starts from your chat GPT memory and its hands are codecs and

9:23the tools you've connected. And according to the Verge, it can connect to more than 4,000 supported apps. So if you already spend your day inside ChatGpt, your DOT arrives knowing a fair bit about you. It works on its own cloud computer and it can reach into your own machine, but that stays switched off until you turn it on. The controls are

9:43the most detailed of the three on paper. Custom rules let you decide which actions it takes alone and which ones it has to ask you about first, and a password change always comes back to you. There's no price for DOTS on its own because your first DOT comes with ChatGpt Pro, which starts at $100 a month. On the business side, you need a workspace of at least two seats. So, the cheapest way in is one premium seat at 125 a month plus one standard seat. The

10:11$20 plus plan doesn't get a DOT at all. So, this is the most expensive way in of the three. Data Camp reports that the plan allowance for its heavier work isn't counted for your first month and OpenAI hasn't published what happens after that. So, I'd wait for those terms before you plan around it. Setup only works on a desktop for now and it can't call you. One detail from Data Camp's writeup really surprised me, which is that deleting a DOT's saved memories means deleting the dot itself. So, is Dots just OpenAI's version of Muse? On the surface, yes, because you get a named agent with an avatar that you

10:47message like a person, and the press made exactly that comparison on launch day. underneath. I personally believe it works more like a single Grockbot with OpenAI's model inside because it's built for work in ChatGpt and Codeex instead of for your personal life. And OpenAI's own launch post points the same way. It says that over time they envision teams

11:09of dots working together on your behalf. And that's a promise on a road map. So today you get one dot. And I can hear some of you saying, "Johnny, it runs OpenAI's best model. So doesn't that settle it?" Not yet, because the benchmark numbers behind that claim are OpenAI's own, and a vendor's own numbers can't tell you how it'll do on your work. So, if your work already runs through ChatGpt or Codeex and you want one very capable agent, now DOTS is yours. So, let's put the three side by side and let you pick. Because once you strip out the launch noise, a few questions do most of the work. Start by

11:48asking yourself whether you've got one stream of work or several lanes that need to hand off to each other because handoffs are where Grockbot pulls away from the other two. Then look at where your work already lives. If your customers and your day sit in WhatsApp and Instagram, Muse is the natural fit. For anyone who lives inside ChatGpt and Codeex all day, that's Dots. And if your team already pays for Cursor, Grockbot comes inside the plan you're on. It's a [clears throat] bit like digging a channel on a farm because you dig it where the water already runs instead of fighting the slope. An agent that sits

12:23where your work already lives is useful from its very first task. And then look at what you're willing to spend before you've seen it work because Muse is the only one with a free tier. And Dots asks for $100 before your first task. Some people end up running two, which is what I do, Muse and Grockbot, because different work lives in different places. My own team runs on Grockbot, so weigh what I've said about it with that in mind. And I've tried to give each of the three its real best case. You might be asking, Johnny, what about Google's Gemini Spark? It's a real always on

12:58agent built around Google Workspace. And that's a separate comparison, so I've kept it out of this one. And guys, none of this is a prescription for what you should do because your context is your context and no outcomes are guaranteed here. So that's the path and for most of you where your work already lives will settle it within about 48 hours at the end of September. SpaceX AI shipped team bots. Meta pushed Muse into small business and OpenAI launched dots with Teams on its road map. And just for fun, TechCrunch reported that XAI owns the web address dot.com and it sends you

13:37straight to the Grock download page. Nobody outside XAI knows whether that was a dig at OpenAI or a typogor forbot.com. So, I wouldn't read anything into it. Now, back to those 48 hours because they tell you the feature gaps I've walked you through will most probably shrink every month. Where your work already lives stays the same while the features catch up. And that's the filter I'd lean on if I were choosing today. Big takeaways for me was that dots and muse only look alike on the surface. The moment you look at how each one hands work around and where it lives, the choice gets a lot easier than this

14:14launch season made it look for our challenge. That's the lens I'm using every time I hand a new job to an agent. I can't say too much as of yet, but once OpenAI publishes what DOT costs after month one, I'll be coming back to this whole comparison. So, which of the three would you trust with your inbox first and why? Drop it in the comments because I'll be reading through them. And [clears throat] guys, if you want the 26 agents I run already set up with their jobs and their skills, that's all inside the AI founders vault with the community. They run on Grockbot or Meta's Muse or Dots and on an

14:49open-source harness such as Hermes. You'll find the agent fit sheet from this video in the vault with every price dated and the link is down in the description below. If you feel called to it, subscribe to follow the journey. See you in the next.

Where these words come from. This is the caption track YouTube holds for this video, written automatically by YouTube rather than by the creator. We read it, tidied the line breaks and laid it out so it can be read. The plain text version is at https://viewrankai.com/tools/youtube-transcript/CLiH1xFMqWI.txt.

All rights in this video belong to Johnny Nel | AI for Founders. Watch it on YouTube. If this is your video and you would rather this page did not exist, tell us and we will remove it.