Building a System of AI Agents with Grok Bot

How I AI as a PM· 11 min· 1,980 words· 9 min read· English ·Watch on YouTube

This is the full transcript of Building a System of AI Agents with Grok Bot, published on YouTube by How I AI as a PM. 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.

0:00Hey everyone. Uh today I want to talk about one of the most phenomenal products I've used in recent times called group bots. And I want to talk about how how I have created like a personal system of agents with growth bots. Um this is hands down one of the best experience I've had with any product uh since clot code CLI when it came and everything just was just working. So I'll quickly talk about what is growthbot and uh then I'll talk about what how I created like a system of agents with groupbot and why it is so helpful. Um let's start with what is groupbot. So groupbot is basically one

0:43of the products which is built around this always on agents system. Uh it started with openclaw and hermes. They were these open source harness. Um, and you could host them on the cloud and you can build a lot of agents and everything on top of it and you can just talk to them via WhatsApp or any any channel, Telegram, any channel. Um, that unlocked something really really powerful and sort of the world understood that if you really provide an LLM with a lot of infrastructure and computer in the cloud, it can do phenomenal things. And uh where I feel uh they lacked is that they were still built for this hacker

1:20persona or technical persona and they sort of got stuck there and I'm sure they're trying to expand but then Grogbot came and and it just uh blew the experience out of the park and I feel this is the product that is built for everyone uh not just technical people. Uh so let's just quickly go to uh Brokebot Mac app itself. And let's talk about what is a group bot. So here's a nice video obviously created by groupbot with just a prompt. Uh so your group bot it has your bots, your agents, computer connectors, routines, works anywhere, you stay in control and teamwork. So

2:03let's let's understand this one by one. So let's understand in details what is group bot. So group bot is basically the whole product and I can interface it with the grow group grow group bot app on Mac, iPhone. It's simple chat app where I can just give instructions. Um group bot has all the bots and basically I can create any bot. uh bot is nothing but a special purpose agent you can say that and I have like chief of staff whose job is to interface with me and then delegate task deep research for going deep into certain topic that I want and just generating a report that I

2:39would like to read uh AI learner which is where I talk about harness and evals and and it helps me learn about AI. The best part is the infra. So groupbot has its own computer and all these agents have access to this computer and they are basically using the browser they are running code they are creating files they are creating diagrams they are creating videos whatever is needed they are creating on this computer which is always on available in the cloud now obviously u any system of agents basically need a lot of connectors because I have my own Gmail my own GitHub uh Twitter uh Versel and so on So

3:18I've connected everything possible to group bot. Um and uh about schedule jobs again these routines and everything you can think of it's the technical word but the idea is I just tell it that every Sunday do XYZ things just one line and it'll create a routine task. Um and the best thing which actually blew my mind is I could literally message one agent to another. So I could say to my chief of staff just pass on this message to deep research and it actually passes a chat message. So basically these bots are working like a team and that is where a system of agents comes into

3:54picture instead of these individual agents and that's where the real productivity unlock happens and then obviously you stay in control any kind of right actions uh uh it drafts before sending it does not send to email it drafts and keep it there. Um it approves before risky actions. So the idea is that I have this system of agents where I interface with one. It knows everything. It has all the in it has connection to all the tools and everything just works by pure one line two-line instructions. Uh and there's not a single task that I have tried and it has not worked the first time. That

4:32is a very very different kind of experience that I have had and this is certainly not my experience with claude because even with claude claude co-work cla chat cloud code cloud code desktop app there's a tons of effort that is needed to create this system same with her same with with open cloud but grow bot just worked out of box and I think that's the northstar experience every AI product needs to have where it just works that's the northstar So now let's go to the second topic u why a system of agents and not individual agents and then I'll show you an example of growth bot in action. Um

5:13first of all why system of agents and not individual agents. I think u I don't know if you've read this book called getting things done and it's about like uh when you get a task uh uh where do you file the task uh how do you create a task filing system for yourself uh then how do you pick up the task which ones you do which ones you delegate and so on so it's it's about productivity and all it talks about is creating systems and uh I think that's where a lot of productivity will be unlocked where it's not about Like having tens or 20 or

5:48hundreds of agents working and then you going and reviewing one output, two output, three output that creates a lot of review hassle and then moving around context from one to another. That will never give productivity. The real productivity unlock is the system where the system essentially orchestrates the agents the context and everything around the task that you have given and you just simply give the task and it just knows everything and that's what I feel like the kind of experience I can see with robbot and that's where like true productivity will get unlocked. Um also the productivity is is not about like having uh 10 different small small tasks

6:28being automated by uh real productivity will come when the higher order jobs or like end toend work that I do throughout my day. The more and more end toend work that I do gets automated the better. So for example in in let's say uh I want to create blog with AI which I will just read the draft. I have tried a lot of systems where it will try to uh uh basically write for me and then it will uh I will go to clot code and ask it to like update the website uh code add add the blog and then publish on versal and then I'll review there. So I've tried

7:06these multiple steps but with growth bots basically it's about simply just brainstorming on a topic and u and the outline and the PR versus PR preview of that blog is already ready for me. So that is like an end toend job being completely automated for my website blog and the idea is that I just uh discuss the outline and then I review the final work and just merge the PR. So uh that's that's about like end toend work that's one end to end work being automated largely and the more end to end work you automate the more productivity you will get and essentially humans sort of get

7:48abstracted out and focus on more bigger and bigger things. the more you extract from the execution, the more you can think big and and that's how like the companies work essentially right like you you give um um um you delegate to someone and then you think big uh and more high your order I think that's where I feel this uh I I read somewhere like we are all managers and that's very very true because this is pure delegation pattern right so in in in office how it works is your boss has a task they don't like go and give task to like 10 different people. They give it

8:24to like a one manager or one chief of staff and that manager is basically managing like a a large number of specialist you know individual contributors and so on right and the IC's are doing basically their own specialist job coding designing product managing and whatnot. So the idea is is uh is how do you rightly delegate and review which I feel like ma those who are good at managing good managers I think would probably learn it even faster. So let me show you how the

8:57system works. Uh so this is my grog bot. I gave it some four five tasks. I told it to do deep research on AI wells platforms like raindrop.ai. Um um u clot do it is launched which is a competitive group board. I asked my last 30 days agent to just go and research on the X what what people are saying about it site audit for kush.com my website uh product marketer for some ideas and I learned them for learning about graph so I gave this five task and all of these tasks it basically my chief of staff agent figured out and it figured out to which agent to give what

9:33so basically uh it has this own bot so this is the chief of staff bot and this is the five tasks I can write a task here and it will automatically pick up uh or it will add the task here as well. If I talk to group B directly it figured out to which agent to give created the task took the notes um and gave me the results. So let's say teach me about graph frag if I just open finally it gave me this doc right and uh to be honest I read this and that's a great research I think I can't understand the difference if it use opus 5.5 or group

10:104.7 to be honest or 4.8 on it. It just worked and I actually read through it. Um pretty pretty good to understand about this topic, right? And the best part is I just have it here. I can just open um you know and just read it. That's it. Uh and all of this is already on my um um on my drive. My drive is also connected to group bot. My notion is also connected and everything works right. So that's the beauty of group bot and that's the beauty of my chief of staff agent knowing everything about all the other agents and and I just keep giving task and it just works. Uh I feel

10:49like uh sky is the limit with this kind of product certainly has set a benchmark for the right kind of experience for any AI product. I hope you have a lot of fun. are definitely worth subscribing and giving giving it a try if you are an AI enthusiast or in general you like to uh simplify your life with AI and focus on

11:09some bigger fun things. Thank you.

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/PbdbxF_0ABY.txt.

All rights in this video belong to How I AI as a PM. 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.