Transcription
Cloud Code 2.0 just dropped, and you can now automate anything. And in this video, I'm going to show you the upgrades that will actually improve your vibe coding and how to use the brand new feature that will actually unlock new use cases that you'll actually use so you don't waste your time and become insanely productive. And if you don't know who I am, my name is Jack Roberts. I built and sold my last tech startup with over 60,000 customers, and I run several AI automation businesses. So, if you haven't already, grab that coffee and let's jump straight in.
So, somebody dropped off a Starbucks gold card at Anthropic HQ because they are shipping like parrots. Now, it feels like every other day Claude Code has released some crazy update, but the reality is that most of them don't really move the needle, if we're fair. But this one feels and looks like a 2.0 upgrade, and it does unlock some cool stuff that we couldn't do before. So, essentially, they've merged together chat, code, and co-work into one app, and they've shipped a new cloud agent that can work for you 24/7. And I'm going to show you actual use cases because I always ask the question, great, but is it going to change anything that I'm doing? In this case, it actually does. And I'll show you that in a cool use case. So, think about it like this. Now, it's essentially three products in one app, the less disconnected, and I'll show you exactly what that looks like.
So, let's take a look at what we've got in the app. Best shown by pulling it up. So, you'll notice now a little bit of the UI change, which I'll show you so you can get the most out of it. Claude, code, co-work, and chat are basically things you can toggle here. You can literally toggle here. Claude co-work basically the chat that exists right there. So it's really easy to go back and forth. And then obviously in here yourself, you'll see now this little bit of an overview when you can come down here. This is where you select the level of permissions. This right here, if you click this, will tell you how you want it to respond. Normal thinking, verb summary. I find that normal typically what's best for me. You've got this add to add files. This lets you chat and yap into it. This is your dictation settings. Then here, for example, you can dictate what file you would love to work from. And then most importantly, don't forget this. Come down and select the one that you want to open. 4.61 million or max. You can decide everything there.
But probably the coolest thing is when you give it a task like this, for example. "Hey there, what I'd like you to do is create for me a very simple page. Looks gorgeous design, and I want a form that just asks three questions, which is name, email, how much money you have, and then why do you want to work with me? Make it gorgeous." All right, very simple question. But what's really cool here is now views, because essentially what's happening here is Claude is merging into almost like an integrated development environment. The philosophy is they never want you to leave the ecosystem. So you see you've got many options here to look at different views. I can click on the terminal. So this will pivot terminal. I can now yap with my terminal directly. I can look at tasks that it creates if it will basically create tasks. So I can view that the plan, and you can basically move these around, and it's just a nicer viewport to do that. The coolest one, though, I think, is going to be this one here, which is preview. Effectively, when it creates something like the app it's building, we can actually take a look at that together.
Now, to say you're watching 16 videos, here's a very brief TLDDR of what you've got into it. Then I'll show you the really cool use cases. So, first of all, yellow dot system, which is very cool. You can dispatch, for example, five agents and get pinged only when they need you. So they just run for you autonomously in the background. Second of all, we can have parallel project sessions, which is so freaking cool because basically you have a sidebar group by the project. And as you can see, I've got here, for example, right? I've got desktop, I've got school scraper, I've got others. And if I'm really honest, this was a UI improvement we should have had a long time ago. I love the fact that we've got it now. It just makes organizing doing work just like they found with co-work that was so effective where we have these individual projects, for example, it just makes it like a million times easier. And even you can see the UI here has taken a step up. And so with the workspace, like I showed you with tasks and basically preview, you can drag and drop those around in a modular way. Then you got hooks, you got your connectors, and you've got voice, like I showed you. You can press and hold to talk or use a voice dictation app if you like. And then we have YOLO mode and a couple of other cool updates.
Now, this usually leads us on to the biggest update that exists, which is called routines. I think it's probably one of the worst names that we've ever had in a project in living memory, but it's a very cool feature and it replaces a lot of what I've been using N10 and Make for. It isn't a wholesale copy and paste, but you'll see what I mean, and it's a very different point of view on the whole system. Let me explain what I mean by first defining what it is. So, a routine is a packaged agent. Think of it just like an AI agent that runs in the cloud. In other words, whether your laptop is closed or open makes zero difference. These agents will run for us whenever we want to. We can basically trigger it crucially either when we say so. I could say, "Hey, run at 3 a.m. every day, do whatever I want," or, and this is the big feature everyone's missing out on, we can trigger it via webhook, which we'll come to shortly. So we're going to put all our connectors, all our world into it, and with our laptop closed, it is actually at the moment the biggest shift in how Cloud Code is working, but we don't care about that. What we care about is what does this unlock for my business, which I'm going to show you in a second.
Now, what's cool is we can trigger it in three different ways. Number one, we can just say, "Hey, I want you to run at X time, hourly, nightly, weekdays," you know, it's like a cron-style recurrence cadence. It's time zone aware, which is great for me because I live in Dubai. I'm actually off to LA tomorrow. So, I'm going to lose 12 hours, guys. I'm going to need as much of this coffee as I can get my freaking hands on. Check this out. I got this actually from LA, weirdly enough. But the cool one is Webhooks, right? So, Webhooks is the real unlock. And I'll show you a use case like if you have a customer form and someone fills that out, that can now trigger these really cool agent automations. So freaking cool. So, Claude is almost becoming like naked.com. This is the first step in that direction. So, it's really cool to learn. And then we've got GitHub, so you can subscribe to group events, PR open issues, labeled, etc.
So, here's a key thing to bear in mind before we build. First one is this: each run gets its own sandbox. Routines execute Anthropic managed containers. The fresh clone of your default branch work happens. Environment is destroyed. Every run is stateless. State only persists through what the agent ships out: a PR, a commit, stack message, or an email draft.
So, let's take a look at what this can actually do. First of all, let's just set up a random one just to show you how it works. On the left-hand side, you'll see you have these three things here. You have customize, which basically lets you connect your apps, create new skills, browse plugins. Come here to connect your world. Obviously, you can add everything in, which we're going to need. We can create new skills, whatever we physically want to. It is very, very cool. But if I come back off this now and come back over, click this button. We're going to click on for this one, routines. Cool. Once we're on routines, you can see you've got all, and then you've got calendar here, which is pretty cool. So, you can see everything that's happening on a daily basis. Think of these as your time schedule for your agents. We click on new routine and we click on remote. Now, give this a name. This should be something that you remember. And guys, we always add emojis because we are not savages, right? So I'm going to add a brain emoji for this one because it's cool. So, for example, in this one, I've called it "Daily Intelligence." On a daily basis, go ahead and scrape the latest news and automations. Create a short, beautiful page on Notion. Let me know the five biggest Cloud Code updates and everything else that is worth noting. Max eight points. Now, you need to select a repository. In reality, it doesn't really matter. I would call that a repo-specific stuff. So if you got like an app, like maybe you got a codebase like AIwithJack.com with our roadmap apps and stuff, or maybe it's generic, it doesn't matter if it's a generic thing where it's doing research for you, just select the repo, and in reality, it makes no difference what you're doing.
Okay. Then once you've done that, you can put here "run on recurring cron schedule," which is essentially what you want it to do. So just say when do you want it to happen? Let's say daily, uh, 9:00 a.m. sounds great. Maybe you want to get a little bit earlier. It's completely up to you, but let's just say 8:00 a.m. and add that as a trigger. And then you've got all of your connectors, which will load in by default, which is really cool. When that's all done, just come down and click on create. Okay. Then you'll see this now appears as "Daily Intelligence." Cool. And now you can sit back, and that will run for you every day. Now, the cool thing is, whether your laptop is closed, you're deep on the ground, this will run for you, and it doesn't require your laptop open. So if I click on this, for example, now, I can just click on "run now" as a for instance, and it will do it for you automatically. Now, I'd say do bear in mind, though, that it may not run perfectly straight away. So, what you want to do is actually test this by running it manually yourself. And then we're going to show you how actually you can unlock probably the best use case with this, which I think isn't talked about anywhere near enough, but this is great just to understand and get you familiar with how this works.
And as you can see, guys, that's done. And you can actually inspect your runs. So, do you remember in N10 or Make.com, we'd have your executions? They're just called runs here. They are, as I say, stateless. So, when you click on this, for example, it essentially just opens up a chat window. So, do you remember in N10 where you'd click on something or Make.com, you'd actually be able to see what kind of happened? This basically is just, it is essentially triggering a chat window. That's the best way to think about it, because conceptually, it's very different than your classic N10 to Make automations. And look, it's essentially when you create this task, all it's doing is preloading in the chamber a prompt for a chat window. Think of it like that. Think of it, all it is doing for you is entering this into a chat window. That's basically what's happening. Okay. And now it says done. So I click on this, and then we open it up, and we see what we've got. An automation briefing, April 15th, 2026. This is all now in our Notion document. So, just gives you an idea of just how that works. And now every single day, this will run for me automatically. And you can see my routines here on the left-hand side. I've got it all set there. And if I want to pin a routine in particular, I can just literally drag that up to the top. But typically speaking, they don't let you do it with routines. So you need to come to desktop and you can copy something and just bring it up and pin it. So this one I would recommend that you have, you know, for like most important stuff.
Now, in honesty, when you're thinking about this new feature, generally speaking, the best use case, honestly, out of all of it, that is very cool that it can do that, because there's things that I do every day. But the best thing that it does is its webhook feature. And what this effectively means is we can get an external source. Could be a contact form on a website, could be the fact that you just had a meeting with your mother, sister's, ex-wife, best friend's dog, and you got the transcript and you want to do something. We can use that information as a trigger, as an external trigger to create an automation. And the coolest part is that that's actually done in the cloud, which essentially means we don't need the laptop to run. So, for example, I had this create for us an example form, right? So you can see I've got this, "Let's build something extraordinary," really cool name, your email, um, date range, and then some, "Why do you want to work with us?" So let's say, for example, that you and I run a roofing company together, right? And we want to get leads. And one of the things that we'd love to do is have an automation in which once we get an application for something, we want to send an email, or maybe we want to draft an email for us to welcome them to confirm, or maybe we want to add them to our CRM. We can now do that within Claude. How freaking cool is that?
So let me show you what I mean. Okay, so we've got this here. All you're going to do first of all, once you've got a design you're happy with and we can integrate that into our website, we're going to back up over to Claude. I want you to come up to routines and again, we come down to new routine. I'm going to click on remote again. So let's give it a happy face and call this one "New Lead." Now, say, for example, you can give it a prompt. Now, here's a real hack that you're not going to find in a lot of places. So, I'm going to say to it here, "Every time you get a new lead, do research on their website, create a draft email welcoming them, welcoming them to their email address, bespoke, make it short and in my style, and then create a page in Notion with the lead details, make it a table." I'm going to say create a page in Notion just because we don't have one yet. I'm just showing you for the purposes of this demonstration so you get an idea.
Now, crucially, come down and click on API, then click on add trigger. All right, really, really freaking cool. Now, here's what this looks like. If you click on call for AI API, you can see essentially let you fire this routine from anywhere. That's fine. And you're going to get a call request, which is cool. And you can see all the connectors you've got. Generally speaking, if you think it doesn't need to use that at all, you can just remove the connectors that are not going to be relevant for it. Just to minimize in that very small chance that it may actually go ahead and do it, you know, use it when it really, really doesn't need to do that. So once we're happy with that, you're going to come down and click on create.
Now, once you have that information, you can build that into your app using Cloud Code. And once you've done that, effectively just can tell, "Hey, Farface endpoint," you'll need a backend server to do that, of course. Then come down and give it some information. So I can say like, "Jack, my email is Jack@howmuchmoney do I have? Oh my gosh, let's say we have $100,000." Then I could say, for example, "Hey, my website is glido.com. And just say, I would love to make more money via my, I don't know, voice startup, something like that. Uh, I want to scale into Africa." Okay, it doesn't make any sense, but that's besides the point. You get the idea. Then I can send application off, which is cool. Application has received. We've got that beautiful premium font. Now we're going to shoot back over to Claude. Then I click on routines on the left-hand side. As you can see now, we've got all of our things here. So if I click on "New Lead," for example, what do we see? Look at this. Runs via API. So think about this for a second. You have something on your business now. It could be a lead qualification form, and you can connect this to anything that you want to, and it will for us automatically.
And just like that, guys, it is now completed, and you can see the routines that are running on the left-hand side. Think of this as a visual way to view everything that's happening inside your environment. Okay, so it's found out what Glider is, which is a voice dictation app. It's uh, scraped things. It's done basically the research, drafted email, and a Notion page. So let's check out. It's done all this work. So I click on Notion, for example, open this up. And bear in mind, look, Jack, Jack@glider.com, Glider what they do. Glider wants this. This is what they want. You can customize this in a million different ways. That could be your CRM. It could be GoHighLevel. It could be anything you want to. Then, if I come over to drafts and look, the very top email I've got here is, "Hey Jack." So I click on this, for example. What do I see? "Excited to have you apply. Voice in Africa is a seriously underrated play. Huge mobile-first markets, blah blah blah." You can script all of this that you want to.
One hack I want to be clear on, though, that I haven't seen mentioned anywhere, is that when you're building out these sessions, guys, and routines, is I found the first time I did it, I was like, "Hey, create this thing for me." And then when I clicked into what it actually done, it basically said to me like, "Yeah, great idea, Jack. Tell me more about that. What would you like to have?" And it was having a conversation with a brick wall, basically. So what you need to do is be very explicit on its actions and essentially just tell it, "Most times, do not reply. All you will do, you'll be giving an input. You must execute on A, B, C, D, E, F, G," and let the agent do it. So effectively, you're creating an agent with big ears, listens to some kind of input on a webhook, and then goes and executes. But it does bring us on to the biggest limitation of this, which I'm going to help you avoid. And to help you understand that, it's worth calling out these different use cases.
So, you've got the morning meeting brief. Again, you could say, "Every day at 7 a.m., Claude reads Google Calendar, cross-references Gmail and Drive, posts a pre-meeting brief to Slack, flags conflicts," etc. So you got calendar, Gmail, Drive, Claude, everything comes to you every day. You don't need to worry about whether or not your laptop's open. Then you got the inbound lead reporter. Again, just like we showed here. That could be Typeform coming through, doing research, and it could add it to Slack. Drop your Slack message maybe when you get a beautiful sale so we can celebrate together. The other cool one, I think, is a 24/7 support triage bot. So think about this. You get a ticket somewhere. Um, it fires a webhook. Claude reads a ticket, searches your docs, and passes tickets for precedent. So how we handled this previously: draft reply, categorize it, then routes it to the right human, and a junior support agent that literally never sleeps. Trained on everything that you've ever done in your life. It's a use case that people aren't talking about, but it's really, really powerful, and you can see just like we blew those together.
So, what is it that routines can't do, Jack? Because it's sounding a little too good to be true? Well, it's great to understand its limitations of where it falls down. So if you do skip a section, you'll miss and waste a lot of time. First thing to bear in mind, it doesn't have memory between runs, right? So effectively, when we have it trigger and solve a customer's problem, for example, it itself won't remember it. So if it's important to you that it remembers it, what I'd like you to do is just store that knowledge in a database somewhere. So add a Notion table, add Superbase. Just say, "Once you've done this, add a file here that explains everything you've done." So you've got a master track. Second of all, no local file access if you're using it in the cloud. Okay? If you're using it locally, it can route around and touch all the things on your computer whichever way you want to. But again, your laptop must be open. The hack around that, by the way, is just put the stuff that it needs online. Like this idea of local is great, fine. But if you don't, if you don't really care if it's in GitHub or not, you can just put it there and be fine. No local browser cookies. So you can't ride your Chrome session. No computer use. Everything goes through API. That is the trade-off between local and uh, cloud. So you'll know based on the thing you're doing. If you've got a computer that's running 24/7 at home, it's always on, fine. But if not, cloud is just going to be extremely reliable.
Then you've got no mid-run approvals. So ambiguity can equal failure. You have to be so clear on what the instructions are. And that's one of the drawbacks it's got is like it is what we would call non-deterministic. In other words, you run it 100 times, you will get 100 different responses. And that's where it's different than N10 and Make.com, and I'll explain that in a second. Daily run caps, uh, which I want to call out here. Okay, so let's look at what the actual cost is. So on the Pro tier, you get five a day. Uh, tokens going to subscription pool. Max plan is 15 runs a day. Your team is 25, and Enterprise is 25 plus. It's a new feature. It's experimental. I'd like to see these numbers come up over time, but there's enough for everybody to have a little play around with. But if you're definitely looking for something like email triage, obviously you're going to want to be, um, on the Enterprise to bump those numbers up.
And like I just explained, it's probabilistic, not deterministic. Great for fuzzy judgment, bad for docs every time, which is why it doesn't completely displace Make.com and N10, because Make.com and N10 are great when it's like, "If this, then this. If this, then this." And you can build all that logic out with Claude. 100% of course you can. But if you don't mind having a non-deterministic agent sitting over looking at it and saying, "Yeah, that sounds about right. I want to do that." In that case, it is fantastic. And here's a key thing that's worth bearing in mind as I come to this kind of, uh, explainer here, is that they are going to build on this over time. This is just the beginning. Honestly, it's like Claude is like a sleeping giant right now, and it's definitely going to move into that space to make these kind of automations better. It's the logical next step. So don't think of N10, Zapier, GitHub Actions. They're not competitors. They are collaborators with this tool. This is another tool to add in within Claude. Okay.
And so here's a nice graph that you can look at if you want to compare judgment that's required, whether that's high or low, and on the Y-axis, volume and determinism to kind of basically stress test which softwares you should be using for which particular kind of use case. And to summarize that in a slightly different way, think about it like this. So Claude is for routines, fuzzy inputs, natural language setups, draft summaries, judgment calls. Should it be this? Should it be this? Where you need it to make a judgment call. In other words, if you know for sure and you could program out systematically, this is what should happen in every use case, it's not appropriate for Claude routines. That's where you would bring in either a deterministic flow that you'd build in Claude, or N10, or Zapier, or Make.com. N10, node-based transformed multi-step flows, okay, self-hosted, it's quite cheap to scale. Zapier, again, simple trigger, same with GitHub Actions. In other words, routines for the brain, N10 for the plumbing, and you can Zapier for the gloom. GitHub Actions can be for the rails in this case. Now, Cloud 2.0, know is wonderful, but CO itself is becoming unstoppable in one particular domain that will apply across most areas of your business. And I almost guarantee there's going to be some use case you can apply. And you can learn what that thing is and how to apply it by watching this video right.