We Built Multiplayer AI, No New Agents Needed

0:00So, you're using folder structures and

0:01second brains and all of these things.

0:03You're using my ICM method if you've

0:05been following me for a while. And if

0:06you haven't, maybe you've been

0:07structuring prompts and data and ideas

0:09in a really good folder structure

0:10instead of making agents. Well, one

0:12problem that everyone's having is

0:14scaling and deploying it. And a lot of

0:16my members and a lot of people in my

0:17community have been wanting something

0:19where they can share, work together on

0:21documents, things like that. And that is

0:23exactly what we built and it is live

0:25today. Rather than give you the usual

0:28spiced-up perfect demo, this is alpha,

0:31but it works. It works the way I want it

0:33to and I'm super happy with our final

0:35version here for alpha 1.0 or I guess

0:38alpha 0.01.

0:40Uh and we're just going to dive into the

0:41ways I think you could use it, the ways

0:43it solves problems with deployment of

0:45folder structures, ignores a lot of the

0:47need for a whole bunch of crazy agentic

0:49work, and uh also is just a video for

0:52all of my members that are about to use

0:54it uh to kind of help them understand

0:56how, what, and why. So, let's go ahead

0:59and dive right in. Now, right off the

1:01bat, it's simple, right? Let's say

1:03you've built out some sort of folder

1:05structure that tracks things, right?

1:07It's using my ICM method or maybe you're

1:10using Kaparthy's LLM Wiki, right? You've

1:13got folders, you've got markdown files.

1:15When I ask the agent a question, as it

1:18travels through the folders, it follows

1:20through, it knows where to go, and

1:22becomes the agent I need at the time,

1:26right? So, if I'm using Claude Code or

1:28Claude Code Work or I'm using it in VS

1:30Code or I'm using Codex or Gemini CLI,

1:34when I say what workflows are still

1:36needing to be built for my data and

1:37finance teams, instead of some random

1:39agent that someone built that has all

1:40sorts of infrastructure, I'm using one

1:42model with a good harness, it reads the

1:45file map, decides what it needs to know

1:47and where it needs to go based on the

1:49question, reads all of that stuff, and

1:51immediately brings it to me, so I can

1:53quickly find and understand those files

1:55and do work inside of it, right? Way

1:58more efficient. This is why we've got

2:0043,000 people, almost 44,000 in our

2:02community. But again, this is great and

2:04it needs a folder. It needs folder

2:06structure. Mouse to that runs locally

2:09and it's hard to deploy this in certain

2:11ways. People are using GitHub, there's

2:13different tools out there, but there's

2:14no place to really organize it all

2:16together. Well, we decided and have been

2:19building that for a while to make it

2:21simple, keep it strong, but get the

2:23value that you would without having to

2:25build brand new agents and all of these

2:27other things. The other side of it is Y

2:30Combinator has given requests for

2:31startups as of 2026 fall and they want a

2:34cloud for small software and multiplayer

2:37AI. Essentially, they want purpose-built

2:39tools that can just deploy quickly,

2:41right? Obviously, we have amazing

2:42deployment layers like Azure and AWS for

2:44big software that scales, but how can we

2:47have cloud or scalability for these

2:49small deployments internally or if

2:51you're a solo operator or smaller

2:53company, how can you deploy processes

2:55with authentication, permissions, things

2:58like that and actually move it forward?

3:00Well, sometimes you don't even need

3:01software. You just need some good

3:03folders with good prompts and the AI is

3:05the runtime and you'll get the outcome

3:08of that software. So, how do you deploy

3:10that? The other side of it is

3:12multiplayer AI. Well, you've deployed

3:13these things, we have these chats, but

3:15right now it's just me and Claude or me

3:17and Codex. I don't Unless someone has my

3:19login information, it's hard for us to

3:21work and be inside of the same chat

3:24uh to continue on. Like if I kick Claude

3:26off and walk away, can someone be able

3:28to jump back in from anywhere in the

3:29world? Well, we solved that as well. So,

3:32let me give an example what my community

3:34is about to be testing and a lot of our

3:36VIP members. So, every week we do

3:38something called high tea and inside of

3:40high tea, I record, we answer questions,

3:43we build out stuff. Uh people are

3:45building their own versions of ICMs on

3:47their own process here. And essentially,

3:50what I've done is locally, I've created

3:53a main workspace, a main ICM template

3:57that actually has to do with that,

4:00right? So, if I come in and I go, this

4:03is a local folder that's usually on my

4:05computer, right? And so, if I go look at

4:07it, right? It's got all my folder

4:09structure, it has workflows, it has each

4:11of our sessions where we're talking

4:13about second brains, where I talk about

4:15teams, all of these different things.

4:18And so, if I right now want to give this

4:20to people and have them use it with

4:22their own AI, I have to send them the

4:24zip file. They have to download it. They

4:25have to make sure it works. They make

4:27sure that they have Claude or Codex or

4:29all of these things installed. But now,

4:32what they can do is they can just come

4:34in and as long as they're part of my

4:36organization, and of course, I can

4:38manage users, which I'm about to invite

4:40a whole bunch of people to this, then

4:42they can just come in and they can look

4:43at it and go, oh, you have a template

4:45here, which is, you know, basically just

4:47an uploaded folder. There's workbenches

4:49inside of it. And essentially, what this

4:51means is it's copies of that workflow,

4:54of that ICM template that are being run

4:57inside of a Docker container. So,

4:59essentially, a lot of people are

5:00building or creating and buying, uh, you

5:03know, Mac minis or creating virtual

5:05operating systems. Well, we said, why

5:07don't we just run up containers that act

5:09as these operating systems for you so

5:10you can keep everything secure, it's

5:12separate from all of your stuff, but it

5:14still operates like it's running locally

5:16on your computer,

5:17uh, with a lot less compute, right? I

5:18mean, we're still paying a bit here or

5:20there, but nothing bad. And so, you can

5:22come in and we've created a workbench. I

5:25can create multiple chats inside of

5:26this, right? As if I would sitting

5:28inside of my VS Code or inside of my

5:31Claude, I'd have multiple chats, right?

5:34The problem is how people connect to it,

5:36when they connect to it, the

5:38editability. Currently, it's editing

5:39stuff on my computer, which is

5:41problematic if other people are going to

5:43work with me. So, instead, again, this

5:45is deployed as a container. So, I

5:48created a chat that comes in and I say,

5:50"What question should my VIP members ask

5:52about how to use this?" And as you can

5:54see, when you come inside of here, we've

5:56got it just like you could look at it

5:58anyway. It renders the markdown. If you

6:00want to look at the source of the

6:01markdown, it's there. If you want to

6:03download anything, there's options to

6:05download certain stuff. You can

6:07immediately have the entire beautiful

6:10process of my ICM method, my research

6:12paper, but all deployable in cloud. And

6:15so, not only can I access this chat, and

6:18not only is this not just Claude Sonnet,

6:20this is Claude Code running with Claude

6:23Sonnet as a model. We could use any

6:24harness. We could use Hermes with Kimi

6:26K2. We could use anything. At this

6:28point, we're offering a free version of

6:30it through Anthropic, through these

6:32areas of Claude Code inside of it,

6:34because well, we like it, we trust it.

6:37And honestly, I'm able to handle a lot

6:39more token consumption there for reasons

6:42I'll explain in a moment. But, it

6:44immediately is navigating, walking

6:46through it, diving into these processes.

6:49You can see it's reading as if it is

6:51locally, going and diving through it.

6:53And now, anyone else can come in and ask

6:56questions, work with me as if I would be

6:58working on it. Or if I'm currently

7:00working on this package, like I would be

7:02in VS Code or in Claude Code, we can now

7:05work on it the exact same way with no

7:07decrease in quality or ability, right?

7:10No decrease there. And on top of that, I

7:12can start a new chat, which now all

7:15three of us can be doing different tasks

7:18on the same folder structure. And as one

7:21person is editing it, I can also be

7:23editing it. But, we can protect certain

7:25parts of it. We can edit certain parts

7:27of it at certain times. So, now context

7:30is shared as we're working. So, we get

7:32to work together. It's not deploying it

7:35and pushing it and sharing it constantly

7:37and waiting until it updates. It's all

7:39in the cloud and live. And the best part

7:41is, maybe I finish working on this

7:43computer and want to log in on my

7:44laptop, it's right there. Maybe I want

7:46to share the link with another member,

7:48it's right there. Maybe I want to start

7:49a new one with a different company, it's

7:52right there. And it's all deployable,

7:54easy to use, easy to read, it renders

7:57HTML if we really want to do that as

7:59well, and it has it all built out

8:01inside. Now, the initial ones that I'm

8:04launching to my community are going to

8:05be a vault of all of our ever sessions

8:08and them able to essentially query it

8:10and ask it questions. We can go back to

8:13all of our previous talks and the

8:14packages that we built out together, and

8:17they can even upload and talk about it

8:19in their own way. We get to see what

8:20other questions people are asking and

8:22really just explore what it means to

8:24work together in augment, cuz I don't

8:26think this type of workflow's been

8:27available for anyone yet. So, there's

8:29still a lot of questions we have to ask,

8:32right? The other one is also a workbench

8:34or an ICM to be able to audit your own,

8:37right? If you're building these locally

8:38on your computer, I have a GitHub repo

8:40that dives into it and does it for you,

8:42but what if you would ask questions

8:44about how to do this? So, I built an ICM

8:47of corpus of processes of what it means

8:50to talk and work together and build out

8:52into this. What does it actually mean?

8:54What can I do? How can I ask questions?

8:57You could upload a markdown file of your

9:00ICM that describes it, and this will

9:02audit it and you can see, and then guess

9:03what? I can see what questions you're

9:06asking, so we can work together. But,

9:08where is this actually going? Where do I

9:10see the big long-term value? Well, that

9:13goes for how I'm using it internally

9:14with my company and what it means to

9:16have a company brain, to work together

9:19at a deployable level and still have

9:21control. And in order to do that, all I

9:23have to do is go to my other

9:25organization. So, here's all my

9:27workbenches and ICMs, but if I come over

9:29to my main company area, I can actually

9:31go and see I have multiple templates

9:34here. And here is some data and

9:36templates we're actually using

9:37internally. These are second brains or

9:39workflows that we built for clients that

9:42we are working with, right? And if we

9:44were to dive in and look, there's

9:46nothing in this one. So, if I started a

9:48new workbench and said, "Let's chat

9:51about our work with Field so far."

9:56Immediately, I now need to name this

9:58workbench. What we're doing is we're

9:59copying my ICM, my folder. Now,

10:02originally, this is locally on my

10:04computer in VS Code as a folder. But

10:07now, it's deployed and it is currently

10:09deploying this into a container.

10:12Um main. We're just going to make this

10:14the main one where most of our chats

10:15are. Immediately, I now have a perfect

10:18place to go ask,

10:20"What did we do in Chicago?"

10:24Chicago?

10:26Immediately, what's actually happening

10:29is it's looking at these questions.

10:31Right now, we have a version of Claude

10:33Codec going in, reading through all of

10:35these, looking through commands, and

10:37deciding how to ask. And the best part

10:38is, I don't have to sit here and think.

10:40I can immediately just bounce out.

10:42That's going to sit there and work. I

10:44can make a new chat in the same field,

10:46immediately start working on some

10:48project, and just come back to this when

10:50it's done actually chatting. The other

10:52thing that I can do is someone else can

10:55now ask questions and dive into it,

10:57right? If we come in here and we have a

10:59test workbench, I'm asking what we're

11:02doing for this company. I say, "Hey,

11:03what's in week two?" Someone else has

11:06already been building all of these

11:07files, talking about what our workshops

11:09are looking like for this client, who

11:11they are, intake forms with them, and

11:14it's only inside of my organization and

11:16the people I allow to see it that have

11:18access to it. It immediately tells what

11:20we're doing, what kind of work we're

11:22doing, what days, because it's just

11:24pulling from our folders and files on

11:26this work. But I don't know if you

11:28noticed this earlier, there was another

11:30chat here. I didn't start this chat. One

11:32of my team members did. And he wanted to

11:35know, "Hey, what's going on in this

11:36workspace?" He's onboarding to this

11:37client. Immediately tells everything

11:39about the workspace, what we're doing,

11:41who we're working with, how we're doing

11:43it. Hey, can you find information on the

11:4514th August drop? Oh, yeah, okay. Boom,

11:47boom, boom. We've created a company

11:49brain with two buttons. We created a zip

11:52with a bunch of folders and information

11:53and markdown files, and we uploaded it

11:55here and deployed it. And then, I can

11:57jump in and say, "Oh, I'm in this chat

11:58now, by the way." Okay. Cool. Thanks for

12:00being here, Jake. Now, we both can work

12:03on a similar project. We both can send

12:05out a version of Claude code. Again, it

12:07could be any other one, but a really

12:09good model to build, add to this

12:11workspace, look, create, and work

12:14together without me having to email

12:15these back, upload it to GitHub, send it

12:18somewhere in Google Drive. No, it's all

12:20here. It's our company operating system.

12:22A true agentic OS. But, there's only one

12:27agent. We don't need hundreds of agents

12:29moving things everywhere. It's just

12:31folders and files. One key point of

12:34failure, that's separations of concerns

12:36that allow us to audit it and govern it

12:38so much better. And all I have to do to

12:41create another client, another place, is

12:43either organize, right? Create another

12:46organization, or I can manage API keys,

12:50I can manage external APIs, so things

12:52that aren't even AI. If I have

12:54connections to Google Drive or anything

12:56like that, it's as simple as throwing it

12:58in here. It never gets added to the

13:00workspace. It never gets added to the

13:02folders, so it can't ever get leaked.

13:04It's separate. It is a separate call. On

13:07top of that, I can add new model

13:09providers. If I want to add in something

13:11from Anthropic, or if I wanted to add in

13:14something from uh maybe we want to do

13:16Google Gemini or XGrock. Maybe I have

13:19local models. Being able to throw that

13:21in here is the main goal. Some of this

13:22is still being tested, full

13:23transparency. Uh but, that is the main

13:26area. And then, of course, if I want to

13:28manage all of my templates, the ICMs

13:31that are being uploaded, it's all right

13:33here. And I can log in, join in, and ask

13:36about my company, right? So, this is my

13:38management that I have for my entire

13:40Aduba, and it's never been used before.

13:42I just uploaded the zip file. I say, uh,

13:46"Where are we in our fundraising, right?

13:50Now."

13:51And again, I haven't made the workbench

13:53yet. It automatically makes the

13:54workbench. I'm just going to call this

13:56one main as well. Sh- opens up the

13:59container, opens up the actual

14:01processes, deploys it as a simple

14:03folder, and immediately makes a request

14:06to our Claud process. Now, while that's

14:09sitting there, immediately pulls it up,

14:11it's going to start working and

14:12answering my question for me. And you

14:14see, while that's finishing, I can go

14:16back to my other workbench that we had

14:19asked earlier, remember? And so, we can

14:21just go out, go back to our templates,

14:23come over to the main field team one.

14:25And look, it should be done here.

14:27Amazing, right? Let's chat about our

14:29work with field so far. What did we do

14:31in Chicago? And also, while it was

14:32thinking, I threw in just explain in

14:34bullet points. And that's the best part

14:36is while it's thinking, you can actually

14:37throw it in, right? It's this process of

14:40collaborative work together. Not just

14:42one output done, one output done. Claud,

14:44code, codex, all of these things, they

14:46have the ability to continue chatting

14:48while it's thinking. And we wanted to

14:50make sure that that was in here, too.

14:52Now, as we come through, here's the

14:53Chicago work, built in bullet points.

14:55Here's what we did, who we worked with,

14:57here's the main programs, all of that

14:59stuff. I now have a company brain that I

15:01can query quickly, easily, with very low

15:04deployment needs, but all of the good

15:06stuff of enterprise deployment. I can

15:08share this with multiple teams, we can

15:09chat together, and more importantly,

15:11this isn't just pulling from everything,

15:13it can edit and access this. It can

15:16change each of these processes and

15:18ideas. It can change the fonts or the

15:21skills or the ideas while we're working.

15:24It is not just a skill that it is

15:27editing of folders and files. It is a

15:29breathable workable workspace that

15:32multiple people can edit together and

15:34keeps everything up to par. And if I

15:37come back to the conversation where I

15:38asked about our fundraising goals, I

15:40noticed that in the chat above, it gave

15:42me very specific details cuz it has

15:44access to all of that. It's all in

15:45siloed data and we have it elsewhere,

15:48but I just needed it anonymized so share

15:50it on the screen here because some

15:51things we don't want to share. And so I

15:53simply said, "Oh, hey, give me a readout

15:54that's slightly anonymized."

15:55Immediately, that's exactly what it did.

15:58It knows my processes, who we've talked

16:00to, where we're talking about due

16:01diligence and processes, different

16:04pipelines and ideas, what we've closed,

16:06and it even gives me a quick little look

16:08of where it's been looking. If it was to

16:10add anything or edit anything, I can

16:13easily wander in and everything is

16:15human-readable.

16:17There's no crazy agentic infrastructure.

16:19There's not multiple agents doing crazy

16:21things. It's all right there, simple to

16:24build in, simple to iterate on. And that

16:26is the power of a lot of this. Now, it

16:28is early on. I want to make this very

16:31specific cuz right now we're having

16:33people come in and test inside of

16:35organizations we create. But in the

16:37future, many of you will be creating

16:39your own organizations. You might have

16:40multiple for different clients or maybe

16:43you have your own company's

16:44organizations, and inside of each one,

16:47you have workbenches for each client,

16:49right? It's a very dependent thing. But

16:51just by naturally uploading these

16:53folders, these files, these areas,

16:55putting it here, you're creating the

16:57company brain as you do your workflows,

17:00as you learn how to use this. You can

17:02easily delete and move and edit it. You

17:05don't have to go and talk to some vendor

17:07to change their agentic workflow and

17:09then decide where their LLM wiki second

17:11brain should design it in. You're in

17:13control. You know where the folders are

17:16and how they are. You can decide who has

17:17access to what and where who can and

17:20can't change certain things, what your

17:22users are doing, where their roles are,

17:25inviting them at scale and bulk, or

17:27inviting them back. In my idea of the

17:29future, this is a great platform to

17:32handle a lot of the problems we've been

17:33seeing when working with companies that

17:35were trying to get ROI in AI. Now, I

17:39could probably going to do a lot more

17:40videos to kind of exemplify this, but I

17:42just wanted to do a quick launch video,

17:44show what's out there, give that idea.

17:46Again, this is only launching to our uh

17:50main uh alpha testers first, people that

17:53are getting approved through VIP for two

17:55big reasons. One, I'm covering all the

17:57AI costs right now. You don't even need

17:59to have a Claude Code Max account to get

18:02the ability or capability that would

18:03have been with Claude Code or Codex. We

18:05are covering those costs. Because of

18:07that, uh it's going to cost us a little

18:09bit of cash more people that use this.

18:11Luckily, uh Anthropic as well as others,

18:13we're on the scaling tier, which is the

18:15highest tier that you can have with next

18:17to custom. So, we get over 10 million

18:19tokens per minute. Uh we get over

18:21$200,000 in max spend if we really

18:23wanted to do that. But, this is going to

18:25be a lot of money that I'm spending

18:27testing this for you all, solving these

18:29problems for you. So, I need people to

18:31pay, or else we're going to just not be

18:33able to scale with this. As we learn how

18:35much it costs as we go through it, we're

18:37going to allow people through. The other

18:39side of it is, when we see people

18:41working well with it and they have a

18:43good use case, we're going to allow them

18:45to put their own API key in, or even in

18:48the future, just put your subscription

18:50that you're already paying for with

18:51whatever service provider it would be

18:54inside of it. And that's a real big goal

18:56there. Uh but that's just super

18:57important. We have a lot of credential

18:59usage and dashboards of how many credits

19:02people have, people requesting more

19:04credits. So, if you want to use your own

19:05AI API key, a company doesn't have any

19:08AI accounts at all, you can just walk

19:10in, build their stuff, have them log in,

19:12and it's super low friction. Uh we also

19:15get to look at users' dashboards, and

19:17we're creating different versions of

19:18this, which of course will be updated

19:20with time. This is early alpha.

19:23Uh but that's it. Right now, I think

19:25over the next 2 weeks, it's just going

19:26to be a lot of testing, a lot of work,

19:28seeing how people work together. I'm

19:30really, really interested to see how we

19:33kind of do this multiplayer AI work in

19:36chats together, ask questions, build

19:38together. I'm excited to see where

19:40that's going to go. And it's finally at

19:41a point where it's actually working in

19:43high quality. It knows how to navigate

19:45folders, build things, create HTMLs,

19:48install packages, uh use Python, use

19:51Playwright and Chromium, all the good

19:53stuff that you would get out of these

19:54coding agents, but in a in a multiplayer

19:57environment, so to speak. Uh so I'm

19:59excited to see what we can do. Uh if

20:01there's anyone who wants to work heavier

20:03with this, I have the email uh

20:05info@aduba.ai

20:07um

20:07or .aio, sorry.

20:10And of course, in our VIP community and

20:12our Discord, we'll be sharing a lot more

20:14about that. We will be doing a larger

20:16public release uh little by little and

20:18getting people involved. You can sign up

20:21right now, even if you're not part of

20:22our community, you just won't have

20:23access to anything. Uh and so we can get

20:26you into the system and start getting

20:27you there. Uh but as always, uh you

20:29know, we're in alpha right now, and we

20:31need people to test it. So, hopefully

20:33this is interesting to all of you.

20:34Hopefully you see the crazy use cases

20:36that you could rip through on this and

20:38uh want to explore. And uh in the

20:40meantime, before all that, uh happy

20:43learning, and I can't wait to see what

20:45we build together. Thanks, everyone.