Claude Code Desktop App and Routines vs OpenClaw Cron Jobs
Prefer to watch? Watch on YouTube · Subscribe for new builds · Read the transcript
In this video I go over two Anthropic features: the redesigned Claude Code desktop app and Claude routines. I compare both to my own OpenClaw setup, which runs on cron jobs. If you're non-technical and want to get your first AI agent running without setting up a server or living in the terminal, this one is for you.
I recorded this in April 2026, when routines were still a research preview. These tools change fast, so check the current docs before you build anything on top of what I describe here.
What I cover in this video
- 0:00: What Claude Code in the desktop app and Claude routines are, and why they compete with OpenClaw
- 0:45: The trade-off between OpenClaw (more open, more control) and Anthropic (easier setup, but locked to their models)
- 1:30: My 17 OpenClaw cron jobs, and why my Mac mini has to stay on for them to run
- 2:16: How routines run in the cloud, so your computer can be closed
- 3:47: The desktop redesign and its focus on parallel sessions
- 5:18: Integrated terminal, file editor, diff viewer, and SSH support on Mac
- 6:03: Why repeatable work matters most, and routine limits by plan
- 9:05: Why routines don't replace your own judgment
Routines vs my OpenClaw cron jobs
I've been building my OpenClaw setup for about eight weeks. It runs 17 cron jobs, which are really just tasks on a schedule, and they handle my business and content pipeline. They keep working while I'm asleep, at the gym, or out, but only as long as my Mac mini is on. The agents need somewhere to live, and for me that's the Mac mini.
Routines remove that requirement. You save a prompt, connect it to a repo and your tools, and it runs on Anthropic's cloud. Your laptop or Mac mini can be closed, off, or asleep, and the work still gets done.
I think this matters more than it seems. Model quality was never the real bottleneck. The bottleneck was a person babysitting the machine: keeping the tab open, keeping the terminal running, checking back in. That's a small cost per task, but it adds up over a day, and routines get rid of it.
There is a trade-off. OpenClaw is more open source and gives you more control. Anthropic is easier to set up, but you're stuck with their models and pay through token usage or the API. If you already run OpenClaw on a server, this probably matters less to you. If you're just starting with agents, it removes the biggest point of friction, because you don't have to manage any infrastructure. You define the job and let it run. If you're comparing options, I keep a list of OpenClaw alternatives.
The Claude Code desktop redesign is built for parallel work
The other update is a full desktop redesign for Claude Code. I think it looks great, and I think people will overlook it. The whole app is designed around running work in parallel. A new sidebar lets you:
- Manage several sessions at the same time
- Jump between active and recent sessions
- Filter sessions or group them by project
- See what's still running and what's finished
That sounds like a small UI change, but it shows that Anthropic no longer treats one conversation as the way you work. Before, you sat with one assistant, waited for it to finish, and then started the next thing. Now you can run several streams at once. One session fixes a bug, another writes tests, a third does research, and you move between them like you're managing a small team. I also think it makes building easier for beginners, because the terminal can be intimidating.
They also added an integrated terminal, an in-app file editor, a faster diff viewer, and a drag-and-drop layout. AI coding used to mean constant context switching: chat in one place, terminal in another, diffs somewhere else. That cost adds up. With everything in one window you stay focused longer and can manage several agents without burning out. There's also SSH support on Mac, which helps if your real work lives on a remote machine.
Put the two features together and you get something close to an operating system for agent work. The redesign is the cockpit for work you're actively managing. Routines are the autopilot for work that keeps going when you aren't watching.
The real value is repeatable work
A lot of people treat AI as a one-off trick: ask a question, get an answer, close the tab. I think that's the least valuable way to use it. The value is in repeatable work, the tasks that eat your mornings and that you've done the same way many times without asking why. Routines are built for exactly that. You save the job once, it runs on a schedule, and you stop doing it by hand. That's the same idea as my cron jobs in OpenClaw.
Here are the routine limits I cover in the video:
- Pro: 5 routines a day
- Max: 15 routines a day
- Team and Enterprise: 25 routines a day
When I recorded this, routines were a research preview, so expect bugs. Still, it's clear where this is heading.
You don't need to be technical to do well here. Honestly, being technical can work against you, because it's easy to over-engineer things and break them. What you need is the ability to define work clearly. Can you split your business into repeatable tasks? Can you say "this happens every morning," "this needs approval before it goes out," "this can run on its own," and "this still needs a person to look at it"? That's management, and most founders already know how to do it.
Routines don't mean you put your business in a box and walk away. If your prompt is sloppy, the output will be sloppy. If your process is confusing, the agent will be confused too. You still need oversight and good taste. What routines change is how much one focused person can manage. They expand what you can do; they don't replace you. For more on this kind of setup, see my AI agents page.
Key takeaways
- Claude routines run saved prompts on Anthropic's cloud, so your computer doesn't have to stay on the way my Mac mini does for OpenClaw.
- OpenClaw gives you more control and is more open source. Anthropic's version is easier to set up but ties you to their models and token or API costs.
- The Claude Code desktop redesign is built for running several sessions in parallel, with a sidebar, integrated terminal, file editor, diff viewer, and SSH support on Mac.
- Routine limits were 5 a day on Pro, 15 on Max, and 25 on Team and Enterprise, all in research preview.
- The most useful skill is defining repeatable work clearly. You still need to review what your agents produce.
FAQ
What are Claude routines?
Routines let you save a prompt, connect it to a repo and your tools, and have it run on a schedule on Anthropic's cloud. Your laptop can be closed or asleep and the task still runs.
How many Claude routines can I run per day?
In the video I cover the limits from when it was a research preview: Pro users get 5 routines a day, Max gets 15, and Team and Enterprise get 25. Check the current docs, since these may have changed.
Can Claude Code routines replace OpenClaw?
They cover a lot of what my OpenClaw cron jobs do, and you don't have to keep a machine running. OpenClaw is still more open source and gives you more control, while Anthropic's version locks you into their models.
Do I need to use the terminal for Claude Code anymore?
No. The redesigned desktop app includes Claude Code, so you can download it and start building without working in the terminal. It also has an integrated terminal if you want one.
Transcript
Read the full transcript
0:00Claude desktop just got a complete overhaul the other day and now you're able to use Claude code directly inside getting you away from the terminal type of way to build and now making it easier for first-time builders to just download the app and start shipping right away. They also have another feature that is called Claude routines which directly competes with open Claude setups. In routines, it's something similar where you could set up your AI agent once and it continues to perform that task similar to how I have set up cron jobs in my open Claude setup. And in this video, we're going to be breaking those two features down on how it compares to open Claude and why you should be using this new Claude code and routines feature if you're a Claude user. So when this just came out the other day, I
0:45literally had to sit down and laugh for a bit because everything that Anthropic is shipping these days is directly competing with open Claude has going on and I love it to be honest. One is more open source, one you have more control and with Anthropic, you have an ease of use setup, but again, you're forced to use their models through token usage or API. So, I'm going to break all down who is it for, why is it going to be useful in the future, and where I see this whole trend going. I'm really thankful that they're starting to compete more with open Claude because the hardest part with open Claude is getting it set up and Anthropic is making it easier for people that are looking from the outside in that are not technical to get their first AI agents up and running. You know, I've been building my whole open Claude setup for the last eight weeks.
1:30Everything runs, it's working. We have 17 cron jobs that are handling my whole business and content pipeline and these work in the background whether I'm at the computer or not. You know, you know, while I'm asleep or at the gym or with Rochelle, these run no matter what just as long as the Mac mini is on. The whole thing runs basically on cron jobs, that's the 17 agents which are basically just tasks on schedule and I'm going to relate that to what Anthropic has been building and what they've been doing with these new routines and Claude code built into the desktop app, they're really trying to build something quite similar. Like I said, my Mac mini has to be on all the time, but what Anthropic just did is that since it's in the cloud, all of that is fixed now because my agents need somewhere to live on this
2:16Mac mini and it needs to be on for it to work. Anthropic just solved that problem. They shipped the routines feature inside of Claude code. You save a prompt, you can connect it to a repo and your tools and it runs on their cloud, which means your laptop or your Mac mini can be closed, it can be off, it can be asleep, you can be at dinner, buying groceries and the work still gets done. The one change matters more than it looks on the outside because the bottleneck was never the model quality, the bottleneck was always the human babysitting the machine. You got to keep the tab open, keep the terminal running, keep checking back in. That overhead is small per task, but it does add up across the whole day. And routines kill all that. Real quick, before we keep going, if you're watching this and you
3:01want to actually build with some of these tools, not just watch videos about them, then you're going to want to check out our community down below, Shipping School. We have a full Claude code course, a full Open Claude course and four live bootcamps every single week where we actually help you get set up from scratch. Like actually set this thing up, not just watch a tutorial and figure it out by yourself. And we also provide one-on-one coaching so you can book a call with me, we can share screens and I can help you get Claude code or Open Claude running on your machine. That's it, no fluff. I built this community because watching YouTube's only gets you so far. We launched it just 3 days ago and we have over 55 members. [music] You need people around you who are actually building, people who hold you accountable and
3:47coaches who can help you when you get stuck. I'll put the link in the description down below. Get in now before the price goes up. And if you're thinking, "Okay, but you know, I already run open claw on a server." This is probably less relevant to for you, but for most people, especially people starting to build with agents, this removes the real friction point. You don't need to manage infrastructure. You just define the job and let it run. Now, they also shipped a full desktop redesign for Claude code specifically, and I think it looks beautiful. And I want to talk about it because people are going to overlook this. The new app is built around one idea, parallel work. And they added a sidebar so you can manage multiple sessions at the same time, jump between active and recent sessions, filter them, or group them by project, and track what is running and
4:32what they already finished. And that sounds like a small UI thing, you know, I know, but it's really not. And I do believe it is making it easier for people to build stuff cuz the terminal can be intimidating, but it is Anthropic's saying basically out loud that one conversation is not the model anymore. The old way was for you to sit with one assistant and wait for it to finish, start the next thing, and then the new way is you running several streams at once. So, one session fixing a bug, one writing tests, one doing research, you're bouncing between them like you would manage a small team essentially. They also added an integrated terminal, an in-app file editor, faster diff viewer, and drag-and-drop layout. Before, AI coding
5:18meant context switching constantly. Chat here, terminal there, diff viewer somewhere else. That tax adds up. When it is all in one place, you stay in flow longer, and you can actually manage multiple agents without burning out from the context switching alone. SSH support on a Mac, too, and that's amazing, which matters if your real work lives on a remote machine. So, it is not just a local tool anymore. It can reach out. Put the desktop redesign and routines together, and you get something that looks like an actual operating system for agentic work. The redesign is the cockpit. Routines is the autopilot schedule. For one managing live work, one for work that keeps going when you are not watching, and let me talk about the bigger picture here because I think
6:03that this framing is what most people use is wrong. You know, people are acting like AI is for one one-off magic tricks. Ask it a question, get an answer, close the tab. That's not it. That's the least valuable use of this technology. The biggest value is the repeatable work, the stuff that steals your mornings, the task that quietly eats into your week, the thing you have done the same way 30,000 times and never have stopped to ask why. Routine is basically Anthropic putting that in writing. Save the job once, run it on a schedule, and stop doing it manually. That's essentially how I have my cron job set up in Open Call. Pro users get five routines a day, Max get 15 routines, and team and enterprise get 25. Right now, it is a research preview, so sure it's still the early days.
6:48There's going to be bugs, but the direction is not early. The direction is clear, and that is the part that I want you to sit with is if you are newer to building with AI, you do not have to be technical to win here. And I'm straight up, that's actually probably a disadvantage nowadays cuz you end up over-engineering things and breaking stuff. What you need is the ability to define work clearly. Can you break your business into repeatable tasks? Can you say this happens every morning? This needs approval before it goes out. This can fully run automatically. This still needs a human eye. That is management. That is a skill most founders already have, and right now that skill is worth more than it has ever been because the cost of assigning that work just dropped dramatically. The people who understand jobs are dangerous right now, not just
7:34because the programmers, the clearest thinkers, the ones who noticed the same annoying task happening for the 10th time, and ask why we're still doing this by hand. If you're struggling to keep up with content, well, I'm about to save you about 40 days [music] worth of work. I built something called Content Machine. It's 10 AI agents that run on the open claw orchestration and they handle everything. Scripts, thumbnails, X posts, blogs, outreach, clips, newsletters, all of it. So, I went from 1,000 subscribers to 4,000 subscribers on YouTube in 7 days using this exact system. Every single morning I wake up and the content's already done. I spend maybe 15, 20 minutes reviewing and approving them and I move on with my
8:19day. It works for any niche, fitness, [music] finance, real estate, marketing, whatever you are building and it is 100% completely customizable to your use case. So, you get the Mission Control dashboard, all of the cron jobs, everything I've built over the last 40 days helping me gain more and more people to subscribe and join the community. So, you plug in your own thing and it molds it to you. It learns how you talk and it writes so it doesn't sound like AI slop. $97 one time, it's not a subscription. I'll put the link down below and you'll thank me later. Now, I want to be straight about one thing. Routines does not mean that you throw your business in a box and walk away. That is fantasy. If your prompt is sloppy, the output's going to be sloppy. That is just common sense. Like, if your
9:05process is confused, the agent will be confused. You still need oversight. You still need good taste. Nothing here removes the need for judgment. What it does is raise the ceiling for what one sharp person can manage. Not a replacement, expansion. And that is exactly why I have been building towards. Waking up with work already done, content drafts waiting, research already pulled, the machine already warm before I touch it. That is what I launch and that's what validates it. And it is not just a feature drop. It is Anthropic saying out loud that the future of work is parallel agents, scheduled tasks, tighter feedback loops, and one operator steering the whole thing. If you've been building that way, this week feels like confirmation you were right. If you have not started yet, that is fine, but I would not wait long. The people building
9:51this muscle now are going to look unfairly fast 6 months from now. Not because it's magic, but because they got the reps in early. So, stop thinking like a user and start thinking like an operator, a builder. Build the jobs. Build the systems. Build the routines because the moment your work keeps moving while your laptop is shut, that is when you stop using AI like a toy and start using like actual infrastructure. And once you feel the shift, you can't go back. You know, and that's what we do inside the community and try to help people take their next step with AI. It's pretty amazing. We launched the community 27 days ago. We have 216 members that have joined. We do nine live boot camp calls a week, and it's really to help you just take your next step along this journey, whether you're building with cloud code or open claw or
10:37whatever you're looking to get into, we're going to help you take that next step. And if you haven't subscribed to the channel already, click the subscribe button, hit the notification bell cuz we got updates coming to you live every single day.
Ready to start building with AI?
Join the Shipping Skool community free. Paid members get live bootcamp calls every week and weekly-updated courses.
Join Shipping SkoolKeep reading
Hermes Desktop Tutorial: Install and Run Your First AI Agent
Hermes Desktop from Nous Research puts Hermes Agent in a chat-style app that took me about 15 minutes to install. I connected my OpenAI Codex plan with GPT-5.5 and ran three agents at once: a code repo review, a YouTube channel analysis, and an X account analysis. Then I set up a weekly cron job.
Claude Code Agent Teams: The Observer Pattern for Multi-Agent Coding
The observer pattern adds one read-only agent to your Claude Code team. It watches every parallel worker through file changes, status files, and logs, then writes a shared summary of what's done, what's in progress, and what's conflicting. Nobody has to dig through each agent's work to figure out where things stand.
Claude Opus 4.8 Turns Coding Agents Into Teams
Claude Opus 4.8 and Claude Code dynamic workflows show where AI coding is going: agent teams that plan, split work, review, and verify.
