ChatGPT Now Watches You... Claude Adds Watermarks (and more)

0:00 Another big week in AI with a bunch of new releases. Grockbot, which is a new AI agent from SpaceX. Chat should now watches your computer and everything you do across your apps. New Gemini 3.7. And

0:11 Claude adds watermarks to identify any AI generated text. This is a new series I'm doing here on YouTube and I'm going to be releasing these episodes every Monday. I know AI can be extremely

0:22 overwhelming with all these different updates, so I created this so you don't have to waste your time. In these episodes, I'm going to break down exactly what you need to know in the

0:30 world of AI from the previous week, so that way you don't miss out on anything important. And the best part about this is I'm not just going to sit and hype up these different tools. I'm going to give

0:38 you my honest opinion so you could apply them to your AI tech stack. Now, without further ado, let's dive right into the episode. Now, first of all, let's talk about Grockbot. And this is what it

0:48 looks like on the desktop app. But you could also download the mobile app on your phone, which you could then use to speak with your different AI agents as if you're texting them like a friend.

0:56 The best way to explain this platform is it kind of feels like Slack for AI agents, but just extremely easy and simple to use. Now, let me give you a quick tour of what my Grockbot looks

1:07 like. So, you could see the different agents that I've configured and then we'll test it out so you can understand kind of what it's capable of. So, first of all, I have pinned up here on the top

1:16 a health agent and an inbox triage agent. My health agent is just very simple. I literally just text back and forth with everything I'm eating throughout the day in order to help me

1:24 hit my macros. And it's literally just like texting a personal assistant. And it's able to track all of this for me. And the best part is I'm actually chatting with it on my phone while I'm

1:32 working out in order to give it updates. This is a very basic use case. So let's dive into some of the more complex agents that I've built. All right, so here is my invoice agent that I set up

1:41 actually a couple days ago. This has been really fun to play around with and it's been genuinely so helpful. I came up here and I basically said, I need you to generate three invoices. And these

1:50 are just fake invoices for the sake of this video cuz I don't want to like reveal any confidential information. But it went and generated these invoices in this specific style, which is an actual

1:59 skill that I created. So whenever I say generate an invoice, it's going to do it in this particular style. I said, "This is perfect. Thank you. Now add all these to ClickUp to track status along with

2:10 draft emails for all three of these invoices." And if you don't know what ClickUp is, it's a project management tool. And I plugged this directly into Grockbot so it can go and perform tasks

2:19 and pull information from this application. This is where me and my entire team stay organized. So, as you can see, it went into my invoices and payments board right here in my

2:29 workspace. And then it created these three tasks which are basically breaking down the invoices that need to be submitted. I could open this up and I could see all of the relevant

2:38 information with the price, the invoice number, due date, etc. And based on whatever status the invoice is in, it's automatically going to drag it into each stage. If I submit the invoice, it's

2:48 going to be dragged into this stage. If it's paid, it will then automatically add it there. And Grockbot is the one that's manually going and pulling my email inbox in order to see where this

2:58 is at in the invoicing process. And then on top of that, it went and drafted these three invoices right here just a couple of minutes ago. So, let me pull this up. You can see that it went and

3:07 drafted this invoice with this very basic, you know, text here. And then on top of that, it also attached the PDF, which is basically that invoice that I just showed you before that we generated

3:17 with Grockbot. This is all great. This is all cool. But one of the best parts about this is you could have specific group chats with your different AI agents. So, we are inside of this one

3:26 group chat, which is my content agent and my chief of staff, and they are actually communicating back and forth with each other in order to get different tasks done for me because each

3:35 agent specializes in different tasks. Now, there's a few other things I want to talk about Grockbot before moving on to the next release of the week, and that is that each Grockbot gets its own

3:45 virtual computer, as you could see right here. So, not only can you plug it in to the different applications you're using like Gmail or ClickUp or basically whatever you use, if it's not able to go

3:54 and do something inside of those apps via a connector, it can actually pull up a virtual computer in order to go and do it for you. And again, this is a virtual computer. So, we're still able to work

4:05 inside of Crockbot without it like taking up our screen and using our physical screen on our computer. Next up, we can create routines. So, if I click on this little computer button up

4:14 here, you can not only see this virtual computer that I just talked about, you could also see routines that we've set up. And this is basically a scheduled task we could have run on a specific

4:24 trigger or a timer. So, it could run every day. So, for example, here is the morning inbox check where my inbox agent will go and pull from my email account. And not only is it pulling from one

4:33 email account, but it's pulling from multiple of them, which is actually a pretty cool addition. This is running every day at 8 a.m., but what I could do is I could actually add a specific

4:43 trigger. And what I mean by this is we could say, "Whenever a new Slack message comes in to a specific channel, I want you to automatically go and do some specific task for me." And this is kind

4:53 of one of the first times we've seen this in one of these AI agent platforms. Before we'd have to just set a timer where it runs manually at a certain time every day, but this can now run when

5:02 something happens across one of your applications without you needing to go and manually check if a message was sent in Slack, for example. It automatically will just go and ping this. Now, let me

5:11 show you how simple it is in order to create one of these agents because this platform makes it extremely extremely simple. That's been one of the best parts about it. I'm going to come up to

5:20 new chat. I'm going to click create a new bot. I'm simply just going to say, "Hey, I want to create an email newsletter bot that connects to my Beehive account via the Zap Your MCP."

5:30 This way, we can get all the statistics for my email newsletter as well as help me draft campaigns. So, I'm going to send that off. And basically, for this specific bot, there's something that I

5:41 want to break down. So, if I come over to plugins right here, this is how we can connect our different applications. There's tons of different apps we can connect right here natively inside of

5:50 the platform. But if I look up and search for Beehive, there's no way for us to actually manually connect this in our account. Well, we could use MCP servers just like the Zapier MCP in

6:02 order to do this. If you don't know what Zapier is, it's an automation platform that connects to 9,000 plus different applications and we can connect that directly into our Grockbot. So, we could

6:12 basically work across any of the apps we use day-to-day. Now, let me show you how I set this up. So, if you want to try this out, there is a link in the description to sign up. All you have to

6:20 do is create a new MCP server. And as you can see inside of my MCP server, we have a bunch of different applications configured right here. One of which is Beehive, which again, we can't connect

6:30 to directly inside of Grockbot. And if I click on this, there's all these different actions we can perform across this application now, which is going to be really valuable. So, we could just do

6:38 a bunch of stuff across Beehive. To add an app, simply just click add app, and you could search up whatever you're trying to connect to. You're going to click on it. It's going to prompt you to

6:46 do a couple of things in order to connect your account. And then all you're gonna have to do is come into your plug-in marketplace. Just search up Zapier. There's going to be this little

6:54 add button right here. Click on that. It's then going to take you back to Zapier where you just have to authorize your account and then it should be set up and good to go. Now, looks like our

7:03 agent is being created and it says Beehive is already on Zapier through my Zap year account. So, it's going and pulling the information across my newsletter. So, let's just wait and see

7:11 the output here. And here we go. We have some statistics on our newsletter through Beehive. And again, I wouldn't have been able to do this if I didn't use that zap your MCP server that

7:20 connected it to Beehive. So, you can connect all of your different applications, create different bots inside of here that can communicate with each other, and it's genuinely been a

7:28 pretty powerful tool. I think it's really simple to use for non- techies to begin creating your own agents for specific use cases. Now, there's one massive caveat with Grockbot, which

7:38 really sucks right now, and that is pricing. If you want to use this, you're going to have to be on the Cursor Ultra plan, which is $200 per month, or you're going to have to have a Cursor Premium

7:48 Teams account, which is $120 per month. However, they have come out and said that this is a beta product, so they're only releasing it to people on the higher plan, so that way they can make

7:58 sure the platform is running smoothly before releasing it to the general public. I will say that it's probably not going to be this expensive for long. So, if you want to test it out and

8:06 you're worried about the $200 per month pricing, maybe just wait a couple weeks and I'm sure they're going to roll it out to the cheaper plans. Next up, Anthropic released this big announcement

8:15 right here basically saying that Claude now adds text watermarks to any of the generated text inside of Claude. They released this really long article here. So what I did is I just broke down the

8:25 key talking points that we need to go over in order to answer probably all of your questions about this and why this is important for us as not only claud users but if we use any model like

8:35 Gemini chatbt and so on. So first of all anthropic says why are we watermarking cla's outputs? We're implementing our watermarking to comply with the EU AI act. Anthropic along with several other

8:47 major AI model providers signed the EU code of practice on transparency of AI generated content in early July 2026. Now they went and addressed some of the main questions that we have. So let's go

8:58 through these right now. They said we use a method of watermarking that does not have any practical impact on quality or content of Claude's outputs. The difference between watermarked and

9:08 unwatermarked text will not be distinguishable to readers. nothing is added to the text and there are no hidden characters as well as this doesn't mean that it's going to be

9:17 charging us extra tokens or be more expensive. This is a big one right here. It says watermarking carries no identifying information that and can't be traced to a specific person,

9:26 organization or a chat. And then lastly, watermarking won't be specific to Claude. Basically, all the other model providers are going to have to do the exact same thing. So any model that came

9:35 after August 2nd, we're now going to have this intact. Now, your next question is why are they doing this? And I wonder if this is to combat all the different AI slop that is out there. I

9:45 mean, if you take a look at LinkedIn, it feels like every single thing you see is generated by AI. So, I'm wondering if this has to do with social media platforms in order to automatically flag

9:55 that something was generated with AI and that might in the future influence how the algorithms work of these specific social media platforms. Right here, it says, "What does a watermark actually

10:05 prove?" A watermark can only determine that Claude was likely involved with the content at some point. It cannot distinguish that Claude wrote this from Claude heavily edited this. And then

10:15 lastly, it says that we will soon be offering a watermark detection API. We're in the process of working out the details of its implementation. So this is pretty interesting cuz we're going to

10:24 be able to identify what's generated with Claude or AI or not. And again, this might affect how algorithms work because everything feels like it's flooded with AI slob right now. This is

10:33 an interesting release. I'd love to hear your thoughts in the comments below, as well as if you want to read through the entire article, I'm going to link it in the description so you can go through

10:41 and see what you think about it. Next up, we have ChatBT computer history. ChatBT can basically see everything that we're working on on our computer now across our desktop, across our

10:50 applications, you name it. And this is probably the release of the week that I think not enough people are actually talking about cuz this is really interesting when you fully understand

10:59 how this works because it kind of does feel like a glimpse into the future of how we're going to use AI. With that being said, I think there are going to be a lot of talks around privacy on this

11:08 one. So, let's dive in to show exactly what this is and then maybe talk about some of the privacy concerns here. So, first of all, OpenAI just dropped this tweet a couple of days ago saying,

11:17 "Chatbt can now remember your activity across apps and websites on your computer. With computer history and the desktop app, future interactions feel more personalized and require less

11:27 explanation. And then before I show you how to set this up and how it works in real time, I wanted to show you this screenshot here cuz it kind of gives you a TLDDR of how this works. So, what it

11:36 does is if you have this turned on, it can see everything that is on your computer screen. Right here, it shows that it prepared a launch update. It shows all of the different apps that you

11:45 use and it even said you reviewed the launch plan, checked the implementation, and gathered feedback before updating this documentation. And the coolest part about this right here is it even went

11:56 and suggested a skill in order to recreate this workflow on autopilot. So, you don't always need to do this yourself. And then you can see right here at 9:00 a.m. shows that it

12:06 organized work for the day. And then it suggested a specific automation so you don't need to do it yourself anymore. Now, first of all, let's talk about how to turn this on inside of chatbt cuz

12:15 it's not on by default, which is probably a good thing. If you want to use it, you can turn it on, but it's not going to be on automatically. So, what we're going to do is come down to the

12:23 bottom, click on our username, and then come over to settings, and then inside of here, we're going to come over to where it says integrations on the lefth hand side of our screen, and then right

12:32 at the very top, we have computer history. So, let's go ahead and click on this. Now, we're going to see this prompt here where it says, "Chat should be T can summarize your activity across

12:41 the apps and websites you use without ever recording your screen or your audio. Ask about what you are working on, get help without explaining everything again, or discover

12:49 opportunities to automate repeated tasks. And all we have to do is click turn on right here. And what we could do is we can customize which apps we can grant access to. Or if I want to just

12:59 say YOLO, I could allow all these different apps right here. But I'm just going to click on customize applications. Now, if I click on customize apps, you could see that we

13:06 could exclude any of the different apps that we don't want it to have access to. So, let's say for whatever reason I don't want it to have access to Google Chrome. I could select that as well as

13:14 any websites that I don't want to access. I could add that URL as well. Then just click continue. Now, right here, we're going to see the history. So, I tested this out a couple of days

13:23 ago. And it basically shows a timeline on everything that you touched on your computer and the different tasks that you ran. And then from here, I could just click ask about your history. And

13:32 in order to do this, we're going to type in at@computer history and then we could ask questions about some of the work that we were doing earlier in the day or even a couple of days ago. I'm going to

13:40 send that off and now it's going to pull from our history in order to answer this question. Now, I think the part about this that's pretty cool is sometimes like I'll be working on something or

13:49 I'll be reading through an email and I'll completely forget about it 2 days later. I could just go and ask Chacha BT and say, "Hey, what did Matt say to me on email a couple of days ago?" and it's

13:58 going to go and source that for me and pull up that information. Now, I had this on for a couple of minutes and I was going and performing pretty random tasks on my computer. And then this is

14:06 the history. So, I was able to analyze from 2:50 p.m. to 300 p.m. for about 10 minutes. I was working across a couple of these different tools. So, it said that I was doing my weekly digest video

14:18 editing. I moved from chatbt asking about computer history to reviewing a related document and asset for this video. It then showed that I worked in Screen Studio, which is my video

14:27 software. And then on top of that, it suggested a skill for us to create. This particular workflow I wouldn't necessarily turn into a skill cuz it was all over the place and I was using

14:36 multiple different apps at once. But if I wanted to, I could just click create weekly digest production skill and it would go and turn that into a markdown file. Now, one thing I do want to note

14:45 is if I click on this reveal and finder button next to this task that I was working on, this is basically how it's storing all this computer history. Now, what it does is it turns all these like

14:57 10-minute segments into markdown files. And this is how it's able to reference it. So, it's literally just turning it into a summary memory file. And again, it does this in increments of 10

15:07 minutes. So, if we're working on something for 10 minutes, it'll create a markdown file and store this locally on your computer for that task. And then we'll do that for the next 10 minutes.

15:15 So, we're basically just collecting markdown files with different timestamps along with what we did on our computer in that time. Now, let's quickly talk about how this works and the privacy

15:24 here cuz it's probably what you're wondering. You're probably thinking, "Man, OpenAI is now seeing every single thing I do on my computer and why would we ever let this happen. So, there's a

15:31 couple distinctions we need to make here so you could fully understand that. So, here is how computer history actually works. So, we do work on our computer just like usual. Then our Mac operating

15:41 system records the events and then we're going to see a timeline chatt can search which I showed you earlier on. And now all this lives on our Mac. So any of these activity events are kept for up to

15:52 48 hours. And then we have our memory files, the markdown files that I just showed you which are kept until we go and delete them. Now where does all this activity go? Is it just being stored on

16:02 our computer or is this going to the OpenAI servers so they can actually use this to train their models? Let's quickly break this down now. So what this does is it stores this on our Mac

16:10 like I just talked about and then it's going to go to the OpenAI servers processed in a temporary session not stored and not used for training. At least this is what OpenAI says. Then

16:20 it's going to go back to our Mac as a markdown file and it's going to be stored until you delete it just like I mentioned. So it sounds like they're using the OpenAI servers to process this

16:28 information, but they're not storing it and they're not using it to train their models. And again, at any point in time, you could come into computer history and turn it off and it's not going to be on

16:37 by default. So you don't need to worry about it. I do think this is going to be cool in the future when we can show chatbt and claude in these different AI applications, what we're working on

16:45 throughout the day, and then it can go and reverse engineer and create skills and automations to help us streamline our work. Now, let me know in the comments below what you guys think about

16:54 this. I'm curious to hear if you guys are going to turn this on and use this. Next up, we have Gemini 3.7 Flash, which is an upgrade from Gemini's previous 3.6 Flash. Now, let's quickly go over

17:04 benchmarks. Even though at this point in time I really don't pay attention to benchmarks much because it doesn't really tell us anything. Right here it shows that we have a pretty big increase

17:13 from 3.6 flash. It even says that this is higher than Claude Sonnet 5 by just like a single percentage right here as well as chatbt 5.6 Terra. Here are the benchmarks for long horizon software

17:25 engineering which to be honest like this is a pretty loaded word. I don't even really know what this benchmark tests for. And then here are the benchmarks for web development. And we could go so

17:34 on and so on down the line. One thing I do want to mention is as of right now, we don't have actual access inside of the Gemini application. So, if I come here and try to select a model, we have

17:42 3.6 Flash, which is the previous model. We have 3.6 Thinking. We have the 3.1 Pro model here. I'd assume that they're going to roll this out over the next couple of days, as well as I have heard

17:53 that it is powering the new Spark model. If you don't know what Gemini Spark is, this is basically Google's version of Claude Co-work and Chhatebt Work. So, if you want to test it out, you can use it

18:03 right here. Well, I'm going to be completely honest. I haven't been using Gemini much these days. I use Nano Banana image model, but that's about it. I'm mainly using things like chatbt

18:12 work, claude co-work, and now Grockbot. But if you are a Gemini user, this will help you out because you're going to get a better model that is pretty fast once it releases to the chat mode. But again,

18:22 you could try it out right now inside of Gemini Spark if you want to test it out. And there you have it, guys. That's the first AI weekly digest. I hope you got some value from this video. I'm testing

18:31 out a new style and I'm going to be releasing these every single Monday so you could stay up to date in this super fastmoving world of AI. I'm going to continue my other content styles as

18:39 well, so make sure to subscribe to this channel if you're a nontechnical person trying to stay up to date with AI. With that being said, thank you so much for staying to the end and I'll see you in

18:46 the next video. Cheers.