Big Idea

The argument in one line.

ChatGPT's Dot is only as useful as the context you feed it before turning it on, and that context should live in portable markdown files you own, not locked inside any single AI app.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You have ChatGPT Business/Pro access to Dot and want to know what setup actually matters before you click through the onboarding.
You're building an AI operating system around your business and want a concrete knowledge/state/memory framework to organize context.
You're comparing always-on agent products (ChatGPT Dot, Grok, Claude) and want an honest read on what's half-finished right now.
You run scheduled or webhook-triggered AI tasks and want to understand the tradeoff between cloud-native scheduling and context-dependent skills.

SKIP IF…

You don't have Pro/Business-tier ChatGPT access, since Dot isn't rolled out below that tier.
You're looking for a finished product review rather than a live setup walkthrough with rough edges shown on screen.

TL;DR

The full version, fast.

ChatGPT's new always-on agent, Dot, has two computers behind it: a cloud machine that runs 24/7 and your local machine. Before turning it on, the video argues you need to feed it context in three buckets, knowledge you already know, state that lives in a database or pipeline, and memory the AI learns over time, stored as portable markdown files rather than locked into one app. An onboarding skill harvests that context from your website and inbox, you copy it to the Dot's cloud workspace, and from there you can wire up Gmail triggers, scheduled tasks, Slack notifications, and the new Spaces canvas feature. The verdict: it works, but it's rushed and inferior to Grok's agent experience right now, voice mode is slower than just typing, and anything that can run as a simple scheduled skill doesn't need a constantly-on bot babysitting it.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 01:31

01 · ChatGPT Dots: access and plugin setup

Dot is explained as a Pro/Business-100+ feature not yet rolled out everywhere. Before clicking it on, the creator connects every daily-use SaaS plugin so the agent has tools to work with.

01:31 – 04:38

02 · Business context and the AIOS onboarding plugin

A sketchnote breaks context into knowledge, state, and memory. An onboarding skill is introduced that interviews the user and scans their website and Gmail to pre-fill a portable markdown context folder.

04:38 – 07:45

03 · Create your Dot and set up its cloud workspace

The creator disables the default model-training toggle, grants calendar and local-computer access, names the Dot 'Argos,' and tours its always-on cloud computer and shared workspace folders.

07:45 – 11:33

04 · Testing voice mode and checking the context transfer

A phone call to Argos confirms chat tasks keep running during a voice call. The context folder is copied from the local machine to the cloud workspace, then voice mode's actual usefulness is debated.

11:33 – 16:05

05 · Gmail triggers, scheduled tasks and Slack

Argos is set up to review new Gmail against a trigger, with a human-approval gate before anything runs. A Slack connection is added so Argos can answer questions directly inside a workspace channel.

16:05 – 18:16

06 · Running skills in the cloud: files and dependencies

Native cloud-scheduled skills only ship the skill package itself, so any skill relying on local context or files breaks unless that context is placed in the cloud workspace or reachable via Notion, Dropbox, or Supabase.

18:16 – 21:11

07 · ChatGPT Space: Pages, skills and project trackers

Spaces merges canvas, images and site tools into one HTML workspace. A requested project tracker board only partially materializes, and a sketchnote-style duck image is generated to test native image generation.

21:11 – 22:21

08 · My verdict: what works and what still needs work

The creator calls Dot still in beta: slow, mostly accurate, rushed relative to competitors, and not worth going all-in on for anything a plain scheduled skill could handle just as well.

Atomic Insights

Lines worth screenshotting.

Dot is only available on ChatGPT's Pro/Business-100-and-up plan, and isn't rolled out to every user or region yet.
An always-on AI agent is only as smart as the systems you've connected to it before you ever click the activation button.
Context splits into three distinct buckets: knowledge (what you already know), state (what's true right now, usually in a database), and memory (what the AI learns from working with you).
Storing context as portable markdown files instead of inside one app's proprietary memory means you can plug the same context into ChatGPT, Grok, or any other tool.
An onboarding skill can read your website and inbox to pre-fill your business context before you ever talk to the agent directly.
Dot runs on two separate machines: an always-on cloud computer and your local machine, and you choose which one it can touch.
By default ChatGPT opts you into improving the model for everyone with your data, and that toggle needs to be switched off manually.
Voice mode and chat tasks run independently, so talking to the agent by phone doesn't interrupt a task it's already running in chat.
Voice-mode demos where an agent interrupts your workout to read a lead report waste more time than a 10-second chat summary would.
Voice mode makes the most sense for augmenting your own thinking, like rehearsing a sales pitch, not for routine status updates.
Scheduled tasks and webhook-triggered background jobs cover most of what people are trying to do with always-on agents, with none of the overhead.
Cloud-scheduled skills only take the skill package itself, so any skill that depends on your local context folder or files will fail unless that context also lives in the cloud.
Decoupling context into Supabase, Dropbox, GitHub, or Notion rather than a single cloud VM keeps you from getting locked into one AI vendor's infrastructure.
Notion is a better fit for team collaboration on shared context because it handles concurrent edits and conflicts better than ad hoc markdown files.
ChatGPT's Spaces feature merges what used to be separate image and canvas tools into one HTML-based collaborative workspace.
As of this recording, Dot is still rough: it is slow, it sometimes fails on multi-step instructions until you simplify the request, and it doesn't reliably follow through on visual build requests like a full project-tracker board.
Native image generation inside ChatGPT removes the need to pay for a separate image-generation subscription for routine visual assets.

Takeaway

Context first, agent second, or it just guesses.

WHAT TO LEARN

An always-on AI agent is only as useful as the knowledge, state, and memory you hand it before switching it on, and keeping that context in portable files beats locking it inside one app.

01ChatGPT Dots: access and plugin setup

Dot is gated to Pro/Business-100+ plans and isn't available to every user or region yet, so confirm access before planning a workflow around it.
Connect every SaaS tool you actually use for daily business before activating the agent. It can only act on tools it has access to.

02Business context and the AIOS onboarding plugin

Split context into knowledge (what you already know), state (what's true right now, usually in a database), and memory (what the agent learns from working with you).
An onboarding skill can read your website and inbox to assemble most of your business context automatically, then interview you for anything it can't find.
Build this context as a portable folder of markdown files, since it's the piece that should move freely between whatever AI app you're using.

03Create your Dot and set up its cloud workspace

Check the data-sharing default immediately. Several agent setups default to using your activity to improve the model for everyone.
An always-on agent runs on two separate machines, a persistent cloud computer and your local machine, and you choose which one it's allowed to touch.
Copying a context folder into the agent's cloud workspace is what makes it 'always on' with full context, not just always running.

04Testing voice mode and checking the context transfer

Voice and chat run independently, so a phone call to the agent won't interrupt a task it's already executing in chat.
Voice mode fits best for thinking out loud or rehearsing something, like a sales pitch, not for receiving routine status updates that text would deliver faster.

05Gmail triggers, scheduled tasks and Slack

Build in a human-approval gate before letting a scheduled or triggered agent task take any real action, not just a notification.
Scheduled tasks and webhook triggers cover most of what people reach for an always-on agent to do, with far less overhead.

06Running skills in the cloud: files and dependencies

A cloud-scheduled skill only ships the skill package itself. If it depends on local context or files, it will fail unless that dependency also lives in the cloud.
Decoupling context storage into a database or file service you own (Supabase, Dropbox, GitHub, Notion) avoids getting locked into one vendor's infrastructure.

07ChatGPT Space: Pages, skills and project trackers

Treat new collaborative-canvas features as unfinished. Multi-step build requests (like a full project tracker board) may only partially complete and need re-prompting.
Native image generation inside the chat tool removes the need for a separate paid image-generation subscription for routine visual assets.

08My verdict: what works and what still needs work

Don't migrate your entire workflow onto a brand-new always-on agent feature. Treat it as additive, not a replacement, while it's still being ironed out.
If a job can run as a scheduled skill with no agent babysitting it, that's usually the faster and more reliable path.

Glossary

Terms worth knowing.

Dot: OpenAI's new always-on ChatGPT agent, available on Pro/Business-100+ plans, that runs persistently on its own cloud computer and can also access your local machine.
AIOS: The creator's own shorthand for an 'AI operating system,' a personal framework for organizing an AI agent's context into knowledge, state, and memory.
Context: The relevant slice of information, goals, and instructions handed to an AI agent for a specific job, assembled from knowledge, state, and memory.
Spaces: A new ChatGPT feature that merges canvas, image, and site-building tools into one collaborative HTML workspace you can build in alongside the agent.
Onboarding skill: A reusable ChatGPT skill package that interviews you and scans your website and inbox to pre-fill business context before an agent starts working.

Resources

Things they pointed at.

00:00productAI Native (Skool community) ↗

02:30toolAIOS onboarding skill (Google Drive) ↗

00:00linkBuild your business data and context layer (companion video) ↗

00:00linkHow to set up skills for your business (companion video) ↗

Quotables

Lines you could clip.

21:21

“I definitely don't think that this is going to be the thing to use for absolutely everything. Don't migrate everything and go all in on this just yet.”

blunt, standalone verdict that works as a cold open or pull quote→ TikTok hook↗ Tweet quote

09:29

“Why do you sound like you've been shooting fentanyl?”

unexpected, funny aside during the voice-mode demo that humanizes the review→ IG reel cold open↗ Tweet quote

12:36

“There is absolutely no reason to get these bots and agents involved constantly when you can have a skill running on a schedule, job done.”

contrarian, specific claim that cuts against the hype of always-on agents→ newsletter pull-quote↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

analogystory

OpenAI's new DOT promises to keep working while you're away and provide you a magical always -on agent that you can access from any device. In this video, we're going to break that down and see what you actually get with it, how to set it up properly based on your business foundations, and by the end, you'll understand how it all fits together, what you can hand over, and where it still needs babysitting.

Let's get into it. Okay, so turning this thing on is pretty straightforward. You'll see up in the top left -hand side here, you will have a DOT.

This is only available on the Pro 100 and UP plan, so right now it isn't rolled out to every single user, nor is it available in every single region. isn't that a great way to start a video but the point is here if you do have access to this the first thing that we want to do is not just come here and click through a bunch of buttons because this thing is only going to be as smart as all of the systems that you have connected to it and it's better to get all of those things set up up front give it all of the context that it needs so that once you hit this little button over here it can get started as fast as possible so the first thing that i want you guys to do is head on over to the plugins tab over here and just add all of the plugins that you use for your daily business all the sas products whatever tools that you use to get your work done, make sure that you have connected and installed all of those.

It's as easy as hitting this plus button over here. Some things to note with ChatGPT, unlike Claude at the time of me recording this video, you can add multiple accounts in here. So for me, I've got both of my email accounts attached to this and you can do the same thing for pretty much most of these plugins in here.

It's really useful to do that because that gives ChatGPT and Dot the ability to check both accounts and separate them. Maybe you want to have some live OS thing going on where it will obviously use your personal inbox and then all of your business stuff in a separate workspace. So once we've got all of our plugins connected, the next thing that we're going to want to do is harvest some context.

You might be wondering what the term context means. Well, it means a ton of different things, despite what everyone else just lumps everything into memory. For me, we have knowledge, and this is very important to understand.

This is the things that you know up front, the stuff that might live in your head, the things that you know about your business or your clients and things like that. We want to pull all of that information into our little context bucket over here. The second part of this is state.

State is how things move between systems. If you have a lead gen system, it could go from in progress to close to one. There's a whole bunch of things here, and this usually lives inside a database.

it could be SQLite, it could be Superbase. Point is, this is very different structure to the basic knowledge that we might store as a markdown file or in Notion or something like that. And then finally, we have memory.

And this is what the AI learns by working with you. And when we look at ChatGPT and all of the new bot apps that have come out, They're really, really building on that learning from you memory system that's embedded inside here.

It is storing most of that stuff inside Markdown. So if you did want to harvest them, you would be able to do that later on. But it's very important to understand the distinction between these because they all belong in different systems and they all serve different purposes.

I've covered that in other videos that I will link down below. Point is, we need to make sure that we can get as much of this information up front as possible so that we can store it in context files or at least give ChatGPT the ability to know where to find state or harvest some memories and things. things like that.

We can do that with an onboarding skill. And so the skill in question is going to be this AIOS onboarding. You'll be able to grab it down in the description below.

Once you have that, all you need to do is head on over to plugins, hit add, and then upload a plugin archive that will upload this thing. And it will do exactly what we're about to run through. Something else that you might want to do before we kick this skill off is just create a project over here or a folder space.

For me, I'm using atomic ops. This is the folder structure that I use between Claude and every single AIOS that I have between the various AI apps that just keeps everything centered. centralized.

We'll take a look at that in just a second. But we can kick this thing off. I'm just going to get it to make some dummy data because I've obviously already got all of my context and things.

For you, you would run through the skill and if you want to see what it does... all it's doing over here is it's asking you if you've already got a website so that it can go and check your website and pull down information based on your business your clients things that you do it will ask you some questions about your business who your clients might be things like that anything that i can't find from a website it will then interview you to get that information out of there it will then also connect to your gmail and other sources that's why we connected them first because it can find your writing style from there and various other traits about you that it can then install inside our little markdown files so that when we connect our DOT, it has everything that it needs to be far more useful to you.

And ultimately, once it's done that, it will ask you what the first thing is that you want help with. Obviously, for this part, we're not using this because we're going to be using DOT. But what you will be left with is a little context folder like this.

Yours would obviously look a lot smaller than this because you don't have a lot of things. But you understand the point by now. It's got my voice, my business, my website, my GTM profile.

And then I've obviously done a lot of context over the last year. So that's all stashed inside here. But this is going to be the most important part of anything that you do because it's portable.

So we can plug this into Grokbot. You can use this with ChatGPT, anything. It doesn't matter.

That's the whole point of creating these things as Markdown instead of putting it in some kind of elaborate system. There are caveats to this that we'll get into in a bit, but now let's kick off our little dot over here. You'll see here that it's running on Astra 6 .1.

The second thing here, they're just telling you that the more tools you have, the better this thing is going to be. And the third thing is that it respects your choices. Now, this is important.

You need to manage your data controls. If you head across here, by default, it's automatically going to be improving the model for everyone. So you want to flip that toggle off immediately.

And then depending on the type of work that you do or what you're going to be using this for, obviously use your discretion on this, but... of course you want network access without this it's not really going to be as intelligent as it could be next up it'll ask you to make your dot more helpful of course we want the calendar connected so i'm just going to continue with google i'll add my calendar and i'm going to hit install plugin and there we go now it's busy setting up our dot and then we can customize it and we're going to get into some of the deeper layers of how this thing actually works so there are two parts to how this thing works the first one being is that it has its own computer in the cloud it's always on that's how it gets it's always on functionality and we will certainly be taking advantage of that The second thing here is the local computer.

This is obviously giving it control over your computer so that it can find local files and things like that. I'm going to give it access to both. The first thing you can do is customize your bot.

So if you want to give it a fancy name, if you want to give it an icon, you can go ahead and do that. I'm just going to do this and I'll call it, I don't know, let's just call it Argus. There we go.

A nice machine name. you can see immediately it starts thinking over here and if we just click on that you will see that we have access to call this thing which is essentially just voice mode with this thing you can also connect it to slack which if you do have snack capability i definitely recommend getting it involved because it's a really smart way to work with this little agent you'll then see over here that we have argus's computer so if we now click on this thing we have access to a fully fledged linux instance over here so we can take control of this thing you can use it like a normal computer and that's where it's very important for us because obviously we need this thing to be able to have the things that we wanted to have always available to us in the cloud, because there's a difference between our local computer and this thing running in the cloud, obviously.

So for those of you looking for that always on solution, we need to take advantage of the always on solution. And that means that we need to head on over to our workspaces over here. You can just hit file system and then workspaces.

And here you'll see three folders. The one that we care about is shared. inside shared is where we're going to create a bunch of folder structure so that this thing always has the context and the skills that it needs in order to do the jobs that we want to do.

Well it's a good thing that we created all that stuff up front because now we can literally just get the agent to copy our context folder up in here and then we'll always have portable context inside our always on system. We can do the same thing with our skills, just chuck them in here, and that way it will always have the skills with everything that the skill needs to run in our always -on environment.

And you see when we come back here, it's offered some helpful things. I've blurred them on the screen because they're really quite personal. One of them is really interesting, and it actually reached out to Fathom and grabbed a bunch of call transcripts and things that people spoke about on their call that they either promised me or I promised them.

So that's really cool. This little thing is always working in the background, but for now we need it to help us to get our things into its cloud computer. So all we have to do is say...

from my atomic ops folder can you grab the context folder in there and copy it onto your cloud computer put it in the workspaces shared folder for me and here we go i'll find the context folder inside atomic ops on your mac and copy it to the shared workspace folder on my cloud computer and at this point you might be asking yourself what the hell is this call button for like i said it's just for voice but if i call this does that mean it will break this task let's find out Good morning.

Hey, how's it going? If I talk to you here, does that mean that it's going to break any tasks that I put into the chat or do you separate them and run them independently? Let me take a look for you.

Hey, yep, you're all good. Tasks you put into chat can keep running while we talk here. And if you want to change or stop something, just say so and I'll take care of that.

Fantastic. Why do you sound like you've been shooting fentanyl? Fair point.

That's on me. I'm here, I'm fine, and I'm ready. Go on.

So with voice mode, how would you think that I should use voice mode with my AI operating system, especially with you, The Dot? That's a great question. Let me just check the best angle for you.

I'd use voice as the place where you think out loud and give direction. And chat is where you inspect the results. With your AI operating system, that could mean voice check -ins.

or talking through a messy decision while you're walking, then having me turn it into a task and keep you updated. For you specifically, content ideas feel like a good start. Just talk me through your rough take and we'll find the angle before you sit down to record.

okay cool thanks bye so the problem that i have with this voice mode and all of these videos out there with people sitting in a gym like getting some reps in bro and then they get an important phone call from jarvis saying oh hello sir i just wanted to let you know that you have a lead we sifted through 20 leads like all that stuff is because in the time that it takes that minute segment to transition you could have had a brief send to you where you can skim through something in like 10 seconds compared to that voice phone call so using something like this only makes sense where you're potentially role playing You're trying to get better at selling to people.

You're trying to get more comfortable pitching your offer to people. I think situations like that where you're trying to augment your own intelligence makes sense here. Everything else for this, give or take a few things, is probably going to be better set up on a schedule or set up as some background task that you never even have to deal with, probably on a webhook.

Both of which can be catered for by Argos. We'll take a look in a second. And you'll see over here that it's busy copying Atomic Ops context, which what it does is opens up inside another chat.

You'll be able to see any of your recent stuff in that. little tab over there but it's telling me that it's packaged up everything from that folder which is the right location from what i can see there probably a lot of junk that it didn't need and you'll see that we have one remaining step over here the parent must materialize the archive on the cloud computer so if we head back to argos now it's still thinking so it's probably working we'll come back here in a second and if it hasn't done anything i'm just blatantly going to go and ask it and so after me prodding this thing a couple of times it now says that the job has been done so if we head on over to our workspace folder over here we go to shared We can see that we got the context folder here with everything inside it from my actual AI operating system, which means now from this thing's perspective, It is always on with all of my context.

Something else to note here, there are different ways to achieve always on context. Of course, you can use Superbase, you can use Dropbox, you could just use GitHub. If you're working as a team, you can totally use Notion.

I actually recommend that for teams because it's much better at collaboration without worrying about overrides and conflicts and things like that. So this isn't the only way to solve the always on thing, especially if you're working in different AI applications, it makes more sense to have a central cloud repo somewhere with all of your stuff.

But I will cover that in a separate video apart from the ones. that I've already done. Let's phone this thing again.

Yes, back very soon. I need your help with setting a few things up. Now that we've got the context on the cloud computer, I need to set up some schedules, I think, and perhaps some webhooks.

Can you take a look at all of the connections that I have connected to this and figure out what would be a good use case for some webhooks so that you can react responsibly for me?

Let's start with Gmail.

Let's do personal Gmail first. Do I have to keep this call connected while you're doing that or can I hang up?

You can hang up. I'll confirm once it's active. Chat to you later.

so this is really one of the reasons why i don't like voice mode it just everything seems to take so much longer and you have to sit there and just wait when you can just get all the stuff set up you can ask it to build a plan cool that plan looks good while it's doing that plan you can go do something else to me i still think it's a much faster way to work so unless you're trying to vibe on something just stick with the chat at least for setting things up okay and it started feeding back some information it failed the first time it was able to read the one box but then it got confused between trying to separate both of them and setting up a web hook but then as soon as i said to both of them it worked And this leads me into the part where I talk about how I feel about this.

So right now, I think it's definitely still in some form of beta. While it does work and it does the things that we need it to do, I think it's currently inferior as a fully functional app compared to something like Grokbot. I like the fact that Grokbot has more agents.

It's also got its own dot -like main agent now that they just released today at the time of me recording this. But the overall user experience from Grokbot, despite its own quirks, is still a lot better than this. i feel like they rushed this just to get something to show devdan considering that the guy who made open claw has been working with them i imagine on something like this you would expect over the last nine or ten months that they would have gotten something a lot better by now saying that it's obviously going to get a lot better over time with all of the same functionality Claude, we know, is getting the exact same feature.

And I think it's all going to come down to the future being everyone just copying everyone. And then it's going to be a rush on performance to the price point and all of that stuff, making it more efficient. But I know so many people are worried about being trapped into a specific app, which is why I'm teaching you this way of having these markdown files or separate systems so that your context is entirely portable.

And that way you have a universal system because skills and plugins, as you've seen, are already universal. Anyway, back to this video at hand. What you can see over here is we have our previous task, which was the one that we just ran through for setting up context.

But then also this one where we asked it to set up something based on a trigger. So review important emails that's doing exactly what it's going to do. It's written a task in here.

You can, of course, edit this if you want to. So I could go and change this however I want. but I'm happy with what it's done.

The trigger is a new email. And if we come down to advanced, you can see here, we don't have the ability to change where it runs. If you set up normal schedule tasks in here, you can choose for them to either run on your local computer or in the cloud.

With this, you have absolutely no choice because it's an always -on agent. Interestingly, you can select the model that this thing runs on. I'm going to probably use Solve for this.

I don't know why 6 .1 isn't here yet. It's quite interesting how that works. You can use it locally, but you can't use it in the cloud.

So they're probably still ironing out some kinks. I'm probably going to put it on high effort but then we just hit save and that's it waiting for our approval. This thing isn't going to do anything really important without getting your sign -off first and that's exactly how it should be.

So now if we wanted to, we can go and view all of the schedules that this thing might have set up for us. You can see here in the scheduled tasks, we have Argos review important Gmail and everything that we just looked at on that other screen is now here. We can view it, we can edit it if we want to.

So just take note that this is where your scheduled tasks will live regardless. Something else that's pretty handy, like I said, is Slack. So all you have to do is hit the Slack button.

You need to make sure that your connector is set up up front. And then you can say that you can just add it to your workspace. So we can add this to Slack and you can see it's already given it a name.

After it goes through the permission screen, you'll see that we now have Mansell Argos over here. I have my own Slack profile now, and I'm here whenever you need a hand. What can I help you with today?

Have any important emails landed from the schedule that we have set up? And also, can you tell me information about my business that you have in your context? So if we flip back here, you can see that it's in thinking mode.

We don't get any indication that it's doing anything in Slack. We just kind of have to hope. But I'm guessing by the fact that we have this little green icon over here, that means that it's online.

and roughly two and a half minutes later you can see that we have a reply in a thread so it's keeping things tidy it tells me exactly what i wanted we have no new emails and then also from the copied business context all of this was correct as well it talks about my ai business my ai community and various other things in here so it's able to pull out all the context that it needs from the cloud computer That means that we know our context is working, we know it's always available, and we know we can use it in any system that we need to.

Okay, and then earlier on, I mentioned you could also put skills inside that Cloud VM. Now, why would you want to do that if we've already got native scheduling? So if we do go over to our schedule over here, you can see that we can add a new task.

We can give it some instructions if we want to, like saying, hey, go and run my content pilot or whatever skill that we wanted to. You can then select it to run in the cloud and it will do that just fine. It will take the skill .md, it will take some scripts, it will take any references that you have alongside your skill package up into the cloud to run it.

So that works perfectly fine. The problem that can come into play though is if you have some skills that ultimately need some of that context that you have that normally you store locally on your context folder. or perhaps any other files and folders or databases that you might need to access.

That's where things can start to become problematic for it if you're just running it in the cloud native scheduling, because it's not going to pull any of those dependencies up there. It's only going to take the skill package. So we have some experimental options here, and that's why I say any dependencies that you have, you want them in some kind of cloud place out there, because then with these cloud scheduled skills, If your context lived in Notion, it could just go into Notion via MCP and grab that stuff whenever it needs to, or potentially Dropbox if some of the context lived inside there.

If you had a database that it needed to access, it could just access that from Superbase. So for me, that's why it's very good to have these cloud -native solutions because then you're untethered from everything. But for those of you who don't have that, and you do want to have this local context always available, that's where it makes sense to put a skill file up here alongside the context that you have.

And that way, any skills that you do have that rely on some of the context or some of the things that rely on things that we need inside a local folder, just make this your cloud local folder and put the skill here. And then you can get Argos to go ahead and schedule it and it will do it just fine. But you can obviously go and get Argos to just send that across for you, just like we did with the context folder.

But what I would say here, is tested first. I wouldn't want to rely on the Cloud VM for any of these apps, because again, then you're locked into their cloud -specific space.

If you can decouple yourself and put yourself on Superbase, that's probably going to be a really good way forward. Dropbox, even more so, because you own Dropbox. Those are much better approaches for these types of solutions.

And I will put out a dedicated video that will show you how to do just that for all of your context and all of your information. But for now, I want to move across to the other thing that was released, and that is Spaces. So if we head over here to Space.

So Spaces are exactly what they sound like. They are spaces for you to do things in and it's kind of just merged where your images and your sites and things like that used to live.

You can also see that it's automatically connected to my Dropbox because it can pick that up by MCP. It's kind of like they're trying to build some form of notion inside ChatGPT and I think that's cool. It still uses HTML so we can do all kinds of fun things here.

If we have a look at the guide that they ship with this, it gives you a few use cases. You can see we can have graphs inside here. You can have moving animations that play and you can change the speed.

You could even put in a mini piano if you wanted to. You get the point. The idea here is that you would be working alongside the agent to visualize any of the work that you're doing or to even build things like project trackers and task lists and things like that.

So earlier on, I went in here and I just asked Dot. You can just do at and then you'll see. You've got ChatGPT and you've got Argos.

And inside here, you can ask for any tasks that you want. So for me, I asked it to scan through all my systems connected through MCP and then build out a project tracker for any deadlines that I have to achieve and basically build me a to -do list. At the bottom over here, we could do something like calling ChatGPT and then I could say, make me a sketchnote of a duck.

And then we'll go ahead and use that. So you can use your skills inside here as well to do things. you know, talking about the sketch note.

That's what I use to make these slides in my videos. I could use that for anything that I want to visualize for any of the information that I've researched, which is part of what the point of this is. For instance, if I'm trying to ideate on a video, if I don't want to stash that stuff in Markdown and perhaps share it with my editor so that he can read through it beforehand, it might be good to create a space for us to work in over here.

And inside that space, we could have the entire project tracked, but we could also have all of the information related to that. You would then obviously just have to hit the share tab so that they can see this as well. The thing that I really love about ChatGPT at the moment is that it has native image generation.

Obviously, it's had that for ages, but I can use that inside VS Code. I can use that pretty much anywhere, which means instead of having to pay for a Key or Higgs field subscription for images, I can just generate them all natively. So my skills work a lot faster for both this generation and for any of the media that I need for any of the videos that I'm editing as well.

And there we go. Two minutes later and this thing has made us a little duck and underlined it using my sketchnote style where it's labeled the different parts of the duck. Fascinating.

And here you go. You can see this thing came back to us. It completely changed the way that this page looked.

It didn't actually build us a board. So I'm going to have to prompt it to that. But it left our picture of a duck.

So that's kind of it. But that's where I think this thing falls short. You have to be very specific about it.

It's not going to be this magic pill that does everything. you so what you can then do is go at and i'm just going to use chat gpt because i think it's faster make this a project tracker with tick boxes with chat gpt you can also select the model and i think that's really useful because you might not always want to use astra for this kind of thing and of course you can also enable fast mode over here which i definitely want to do because i'm tired of recording now And so while this thing is filling out, I'll give my final review.

I definitely don't think that this is going to be the thing to use for absolutely everything. Don't migrate everything and go all in on this just yet. It is still being ironed out.

It is quite slow. It is mostly very accurate and it's definitely going to be something great to use once it's fully flourished. But there's going to be competition from Claude and everyone else in this entire space.

So for me, I would always have this decoupled infrastructure like I keep talking about so that I can always come in and out of these tools as I need to. And more importantly, remember guys, that a lot of the work that you're doing can already just be a scheduled task. There is absolutely no reason to get these bots and agents involved constantly when you can have a skill running on a schedule, job done.

and this thing is now done you can see i put in one whole tick box so there's still some quirks to be worked out here notion is definitely superior in terms of creating these things and funnily enough using chat gpt with notions mcp to go and build out entire operating system workspaces is a lot better than it editing its own native spaces which i find quite funny but i guess they're still ironing this stuff out nevertheless i will keep you guys updated thanks very much for watching this video stay tuned for that decoupled context video and i'll see you guys in the next one

The Hook

The bait, then the rug-pull.

OpenAI's always-on Dot agent promises to keep working while you're away, from any device. Before clicking it on, the video walks through the one setup step most people skip: giving it the business context it needs to actually be useful.

Frameworks

Named ideas worth stealing.

01:31model

Knowledge, State and Memory

Knowledge — what you already know up front
State — what's true right now, usually in a database
Memory — what the agent learns by working with you

The creator's framework for the three ingredients that combine into an agent's working context before any task starts.

Steal forStructuring a context/knowledge-base folder for any AI agent or assistant, not just ChatGPT Dot.

CTA Breakdown

How they asked for the click.

VERBAL ASK

00:00newsletter

“Build + sell AI systems that actually work - skool.com/ainative”

Pinned at the top of the description rather than spoken on camera as a mid- or end-roll pitch; the spoken sign-off instead teases a follow-up video on decoupled context.

MENTIONED ON CAMERA

00:00productAI Native (Skool community) ↗

02:30toolAIOS onboarding skill (Google Drive) ↗

00:00linkBuild your business data and context layer (companion video) ↗

00:00linkHow to set up skills for your business (companion video) ↗

FROM THE DESCRIPTION

PRIMARY CTAWhere the creator wants you to go next.