Enjoying this issue?
Get tomorrow's AI & engineering digest in your inbox — hand-picked, summarized, and always spam-free.
TLDR
Claude Duet is Anthropic's new proactive AI agent that autonomously monitors your apps and comes to you with tasks, running on its own cloud computer. It directly competes with Grokbot, which just added pages (artifact-like reports) and phone integrations. Duet's channel system feels like Slack, but its task and make sections are standout features, though it shares usage limits and is US-only for now.
Key points
Claude Duet proactively checks connected apps and sends morning briefs, operating autonomously on a cloud computer.
Grokbot launched 'pages,' its version of artifacts, for building reports and dashboards visually.
Critical PHP bugs (9.2/10 severity) were discovered by AI, affecting ~70% of websites, with a fix released Oct 21.
Anthropic set aside 5% of company shares into a Public Benefit Fund, potentially growing to ~20%, for public benefit.
Claude will run in auto mode by default on Pro/Max/Teams plans, blocking only risky actions for faster execution.
Tools mentioned
Techniques
- proactive AI agents
- autonomous background execution
- AI artifact generation
- AI video generation
- AI-discovered software vulnerabilities
Stop scrolling. Start reading smarter.
Receive the day's most important AI & engineering updates in one concise email. No spam.
Transcript (captions)
What an absolutely hectic week in the world of AI with some major releases coming from Grockbot and Enthropic as well as some big security issues across the board. Grockbot just got its own
version of artifacts called pages. And your bots can now use the apps on your phone like your calendar, your reminders, and your contacts. Enthropic also just dropped Claw Duet, which is
their own personal AI agent, and I've been testing it to see how it stacks up against Grockbot. There's also been a major problem in the software behind about seven out of 10 different websites
on the internet. And this might affect you. So, I'm going to explain exactly what it means for you, even if you're non-technical at all. I'm also going to dive into why Enthropic just set aside
about a fifth of the entire company for the public. If you're new to the channel, I do these weekly updates every single Friday, so you could stay up to date without being overwhelmed since
this is a full-time job keeping up with AI. So, without further ado, let's dive right into the first release. First up, Enthropic just launched Claw Duet, which is basically their version of a personal
AI agent. This is a direct competitor to Grobbot, Chatbti Dots, and Metamuse. So, the simple way to think about this is that the regular Claude waits for you to ask it something, whether it's a Claude
chat, co-work, or Claude code. While Claude Duet goes and checks your connected apps and comes to you with what actually needs your attention, it sends you a short brief every morning.
This way we could have Claude run more autonomously in the background even when we're not using it on our computer. It also has its own computer in the cloud so it could keep working on tasks even
when our phone is off. I got access from the weight list and I've been testing it out. So I'm going to show you what it actually looks like inside of the app and then give you my honest thoughts on
this. Now I'm blurring out some information here that's confidential because there's not really an easy way for me to demo this without blurring some of these different things. So this
is the main chat inside of Claude Duet where it brings me what actually needs my attention. And over here I've got something called channels. So for example, this one is everything YouTube
in my short form channel. I asked it to set up my keyword DM for today's Instagram reel and it went and built it for me then checked with me before it went live. You can also see exactly what
it has access to like my files, my mail, and my calendar. Now making a new channel is really simple. So, I'm just going to call this one finance and give it a little dollar sign to signal that
this is everything finance related. And now it asks if I want it to suggest different tasks it can do for me inside of here. And it's going to do this by looking through its previous context and
conversation threads with me inside of Claude, as well as look at all my different applications and what's going on inside of those connected apps. Next up, we have something called tasks,
which is honestly one of my favorite parts of Claude Duet. It goes through my tools and brings me a short list of things that it could take off my plate. So, I can see exactly what's next at any
given time. And the best part about this is this runs autonomously. I didn't have to go and ask for it to go and try to find tasks it could take off my plate. It's going and being proactive in order
to help me as much as it possibly can. And this is the make section. So, everything that our Claw Duet builds and creates for us, it's going to live inside of here like documents, pages,
dashboards, and all the different artifacts that we have it generate. So, for example, this one pulls the stats for my email newsletter. So, I could see how many people opened it and how many
actually clicked. And if I want something new, I just hit the plus button and pick what I want, like a claw design, slides, a doc, or even a dashboard like I just showed you. So,
I'm just going to ask it to create a dashboard to visualize my YouTube shorts. And the nice thing about this is all of this runs on its own computer in the cloud. So, I could close the app and
it keeps working in the background. And you can see right here, Claude is already working on it. And here's a dashboard it already built for me on my last three shorts. It tells me that the
real stop paying for Salesforce is a clear winner with 7,300 views. And it even breaks down why, like how many people stayed to watch and how much YouTube pushed it in the shorts feed.
Now, one key thing I do want to point out here is that Claude Duet shares the same usage limits as the rest of your Claude plan. And Enthropic says its background check-ins count towards those
limits. So, so if you're already running into your limits, this is going to eat them up a little bit more. Right now, it's only for Pro and Max users in the US, which sucks for anybody that's in
Europe, and you have to join the weight list, but it doesn't cost anything extra on top of the plan you're using. Also, this comes as a standalone app on our phone that we can access. And they're
going to roll this out over time to the computer as well. So, now the big question is how this compares to Grockbot because I've been using Grockbot every single day for the last 2
months to help run my business, and I have been loving it. What I really like about Grockbot is that I could have multiple AI agents that actually talk to each other. So, for example, my main
agent works with my invoice agent and my YouTube agent, and that way it has all the context between these agents. I could also create specialist agents that have a very specific task for it to go
do, so it can be an expert in that given task. Cloud Duet has something kind of similar with channels, which honestly, after using it feels a lot like Slack, where each part of your life or your
business gets its own separate space. After playing around with this for a while, I'm not sure how I feel about channels yet, and I personally think I like the way that Grockbot does it with
separate agents. That feels a lot cleaner and easier to use. Now, one thing I really do like about Claude Duet right now is the make section and the task section, which I showed you earlier
on. I think Claude has always been really good at building things like dashboards and presentations with artifacts, and I could see exactly what it's working on and what's up next. This
is something that I feel like Grobqua doesn't have in their interface. The other big difference I want to point out here is the model because Elon has said that Grobba is going to use the best
model for any given task. While Claude Duet is built on Opus 5.5, so with Duet, you're pretty much betting on Claude always having the best models. And as we know, the the company with the best
models can change literally every single week, it seems. Now, I will say it's not a bad bet if you do want to bet on Anthropic because I use Claude every single day for my more complicated work
inside of Claude Code and it's extremely powerful. On top of that, Opus 5.5 is pretty efficient when it comes to token usage, so you're not spending an arm and a leg like you were inside of Codeex. As
of right now, I'm sticking with Grobot as my main agent, but I'm going to keep testing Duet and I'll give my honest thoughts and feedback on this channel. And if you want to see the full video I
made testing this out, so I'll link that in the description. All right, moving on. And and speaking of Grockbot, there's a few updates that I want to talk about right now that are pretty
exciting. They just launched a new feature called pages which lets your bot build you a report, a memo or a dashboard. This is basically Grock's version of claw artifacts or chat sites.
And I honestly think that Grockbot has been missing something like this for a while now. It's really nice to be able to have your bot build something visually so you could actually see your
data instead of just reading through a bunch of different text. To test this out, I'm just going to ask my bot to build me a page breaking down how my last few YouTube videos performed along
with some ideas for what I should make next. So, personally, a way that I'm going to use this is having Rockbot send me a brief every single morning on everything that happened across my
finances, my email inbox, and the rest of my business as a single page that I could look at, which is basically just going to be my morning brief. As of right now, it's rolling out on the
latest version of the app. So, if you don't see it yet, go ahead and update your Grockbot. And other big Grockbot update this week is that your bot can now use the apps on your phone, like
your calendar, your reminders, and your contacts. So basically once you give the Grockbot app permission on your phone, your bot can see and add to the calendar reminders and contacts that are already
in your phone. Just like your other apps, it's going to ask for permission one app at a time so you could decide exactly what it can touch. Now in XAI's own release picture, they also showed
iMes and send your text, but I updated my app and I'm not seeing this directly inside of my interface. So I'm wondering whether or not they're going to roll that out. To give you an example of what
this could do, I could just tell my bot to put a call on my calendar and remind me an hour before and it takes care of all of that from my iPhone. There was a bunch of other Grobot updates this week.
So, I'm going to run through the ones that actually matter right now very quickly. XAI says your bot is now about three times faster on average when it's doing tasks on its computer. If you can
connect your X account, it can read your X chat messages and you can even tag atbot in a post on X in order to get it working. Also on the iPhone, you can now set Grockbot as your action button. So
you just press it once and you can begin talking to your bot this way. You can now send files up to 250 megbytes and videos up to 1 GB with up to 20 files included in one message. Forms also show
up already filled in with what your bot knows, so you could just check them and hit submit. Lastly, we got some new connectors that are important if you're into investing, like Robin Hood,
Coinbase, and we also got Trello and Beehive. Next up, Cling just released Cling 4.0 know, and I used it to put myself in five completely different places without filming anything new.
Let's take a look at what I mean. And honestly, AI is moving so fast right now that staying on top of all of this could be someone's entire job. That's exactly why I put out these episodes every
single Friday so you could get everything that actually matters to you in one quick video without spending the whole week digging through everything yourself. So, that was all one clip of
me from last week's AI news episode, and Clling turned it into multiple different scenes by swapping the setting. If you've never heard of Clling, it's one of the most popular AI video generators
out there. And the last version could only make videos up to 15 seconds. And this one can now go up to 30 seconds in one single shot. You can give it up to 15 reference images, videos, or even a
voice over to keep your characters looking and sounding the same. You can also pin up to 10 different moments in the video so it hits exactly the shots you want. It also has stereo sound in
more languages and you can edit a video after it's made, like swapping the character or background without having to start over. Now, I'm going to show you real quickly how to use this inside
of Higsfield. So, inside of Higfield, I'm just going to come over and click Cling 4.0. I'm going to go over to the edit tab and upload a clip from one of my previous videos like I just showed
you. Then, I just describe where I want to be. So, for example, this prompt says a sideline reporter at a football game. And I'm just going to click on generate. We could also choose how many of these
videos we want to generate, but I'm just going to stick with one to save me all my credits. And the best part about all of this is you could plug Higsfield straight into your AI agents like Claude
and ChatVt. So your agent can make these cling 4.0 videos right from Claude or ChatVt. All right. So in order to set up Higsfield, what I'm going to do is I'm going to come over to connectors here.
Um inside of my Claude account, I'm just going to look up Higsfield. Then I'm going to click on connect. It's then going to take me to um this page here. I'm going to click on finish connecting.
And then I'm just going to need to sign into my account here. And there we go. We now have Higsfield connected. And we could actually start generating videos with Cling. So I'm just going to come
back here and say, "Hey, I want you to generate a test video with Cling. I want it to be a 15-second 16x9 video of somebody riding a motorcycle through the streets of Japan. I wanted to be
hyperrealistic." All right. So just like that, a couple of minutes later, we now have this video that was generated with Cling 4.0 directly inside of Claude without us
even having to leave the platform and go into Hicksfield. So, I could take a look at this right now. Very simple. It's just a guy riding a motorcycle through the cities of Japan. Um, looks pretty
good. But that's how we could actually access Hicksfield directly inside of any of our AI agents and start taking advantage of the new Clling 4.0 model that just came out. All right, so here's
the first one I made side by side with the original so you could see exactly what it changed. Another week in more AI updates and a few of these releases might change how you use AI every single
day. Now, I also wanted to see if it's actually better than the old version, which is Cling 3.0. So, I gave Cling 3.0 and Cling 4.0 the exact same prompt in order to test the output. All right,
let's go ahead and take a look at the output here. On the bottom, we got Cling 4.0. Top we have Cling 3.0. Now, I think Cling 4.0 looks a bit better. I really like the details here.
Have it zooming in on our rider. So you can kind of see the difference here. And again, this was only a 15-second video generation, but Clink 4.0 can now go up to 30 seconds as opposed to just 15
second clips. Next up, chatbt now lets you lock your private chats behind a pin or your face ID. So basically, if you've ever had a chat about your health, your money, or something personal, you can
move it out of your regular chat history and search into its own locked space. On the computer, you unlock it with a pin. in on your phone. You could use a PIN or Face ID or even your fingerprint and
you'll find all the locked chats inside of projects. New messages in a locked chat also won't update the memory chatbt uses in your regular chats, so it can stay completely separate. It's kind of
like incognito browser but for chatbt. Next up, we have some pretty serious security issues that we've had from the last couple of days. And a lot of people have been talking about this and it can
be pretty overwhelming if you don't fully understand what this means for you as a non techie. So, I'm going to break this down very simply so anybody can understand it. If you've never heard of
PHP, it's basically the engine running behind the scenes of about seven out of 10 websites, including most WordPress websites. They found two bugs that were rated 9.2 two out of 10 for how serious
they are, which puts them in the critical range. The simple way to think about this is it's like a lock on a front door that pops open when a stranger tries to jingle it the right
way. They don't need a key or a password, and once they're inside, they could take control of the entire computer that runs the website. And the crazy part is that AI actually helped
find these bugs. Security researchers used AI models from Google and OpenAI along with their own tools. And the bugs had been sitting inside of PHP for about a decade. The fix came out on October
21st and some security companies say people are already trying to use these bugs. Nobody has confirmed an actual break-in yet, but this is a kind of thing that moves very fast. So, the
question is, what does that actually mean for you and I? If you just use websites, you don't need to download or install anything because the problem is on the websit's computers, not your
phone or laptop. The risk to you is more indirect, like a hacked site leaking your account details. So, this is a really good week to make sure you're using different passwords and that you
have two factor login turned on. And if you own a website like a WordPress website, you're going to want to check with your hosting company in order to make sure your PHP version got updated.
And updating WordPress by itself does not fix this because the problem is in the software underneath it. I genuinely think this is probably just the start of some pretty serious security issues
exactly like this with AI because the same tools that find these bugs for the good guys can find them for the bad people that are trying to break in as well. It's more important now than ever
to stay safe. Make sure you have your password secured and be diligent about what you're doing on the internet. Next up, Enthropic just set aside about 20% of the entire company for the public and
they said they might even look at sending some of that money straight to people. The post is called investing in the public benefit. And basically, they're creating something called the
Enthropic Public Benefit Fund. And as you could see right here inside of the post, they're going to put 5% of the company's shares into this fund as long as their stockholders approve it. And
what I think is really interesting is that it's built to grow. Every time Enthropic's value goes up by 50%, the board can add another 2% of the company into the fund. And just to give you an
idea of how fast that could happen, Anthropic was valued at $380 billion back in February. And by the end of May, it was valued at $965 billion, which is actually higher than OpenI. And now it's
reportedly worth $1.5 trillion. Now, as for what the money actually gets used for, they gave two examples right here. The first is big donations to nonprofits. And the second is what they
call direct distributions to specific populations, which basically means sending money to certain groups of people. And if you're anything like me, your mind probably started going in
circles thinking, how do they determine which groups of people are going to get access to this money? And the example they gave us is people whose jobs are affected by AI. Anthropic already put
$200 million towards studying what AI is doing to jobs. And in this post, they said that bolder approaches may be needed. And on top of this, the founders already pledged to give away 80% of
their own shares, and about half the people at Enthropic donate shares, too. So, when we do the math here and add all of this up, roughly 20% of Enthropic is now set aside for the public. Now,
before anyone gets too excited, there are some really big things we still don't know about this fund, and I think it's important to be honest about them. First of all, we don't know how this
money is going to be handed out. The post says specifically populations, which means certain groups of people, not everyone. So, this is not universal basic income, and Anthropic never said
it was. This is just the first time they've mentioned anything like this. And to me, it looks like they might be setting themselves up to pay certain groups first. We also don't know who's
going to run the fund or who decides who gets the money. Now, that's a pretty big question because it means a company would be deciding who gets paid instead of the people or the government. I want
to break down some of the math, too. The new fund alone is 5% of the company which right now is about $75 billion. And if you split that across every person in America, it would come out to
$220 each a onetime payment. So this fund on its own is not going to replace anyone's paycheck right now. In order to get anywhere close to that, these companies have to keep growing and more
of them have to do the same thing. Universal basic income is basically a safety net. So just enough to cover rent and groceries and the basics. Whereas universal high income is the idea that
AI and robotics produce so much so cheaply that everyone can actually live very well. So the big takeaway from this article in my opinion is that the people building AI are already getting ready
for a world where a lot of jobs change. And that's exactly why I think right now it's more important than ever to learn how to use AI and start using it inside of your job or your business so you
could be on the cutting edge of this. And of course, I don't know for sure if this leads to universal high income, but I genuinely think it's one of the first real signs of it instead of it just
being something we talk about. All right, so before we wrap up, there are three quick claw updates that I want to talk about from this week. And I have to start with one that didn't happen. I
genuinely thought Fable 5.5 was going to come out, but I was wrong. Partner companies reportedly already have it, but Enthropic still hasn't announced it, and the latest reporting say it's coming
in the next few weeks. So, when it drops, I'm going to have a full breakdown video for you. And then starting next week, Claude is going to start running in auto mode by default on
the Pro, Max, and Teams plans across the web, desktop app, and your phone. So, basically, instead of asking you to approve every single step when it's doing something for you, an automated
check blocks anything risky and lets the rest go through. This is already how it works inside of Claude Code. And I honestly think it's going to make Claude feel a lot faster day-to-day instead of
having to approve things every single time. And if you'd rather approve everything yourself, you can switch back to manual at any time. And this last thing I want to talk about is a deadline
because the old Claude Design website at claude.ai/design is shutting down on December 14th. So if you have a design system saved over there, you're going to want to move it
into the regular Claude before then. Anyways guys, that's everything for this week that I wanted to cover. If you got some value from this, make sure to like and subscribe because I put these out
every single Friday. And let me know in the comments if you're gonna switch from Grockbot to Claude Duet or which AI agent you prefer. Thanks for saying the end and I'll see you in the next one.
Cheers.