Claude Cowork for Beginners: How to Turn Your Desktop Into an AI-Powered Workflow Machine

Tate called scheduled tasks one of the most powerful pieces of Cowork. Instead of waiting for you to ask, Claude runs a defined workflow on a recurring basis: hourly, daily, weekly, or on a custom schedule.

"You could go into Cowork and say, I want you to read my email and my calendar and give me a morning briefing every morning," Tate said. "Then it will say, okay, I need to connect to your calendar, and it will ask you to connect. And then once it's got all the connections up, it will say, okay, now I'm going to schedule this to do every morning at 7 a.m., and it will create a scheduled task for you."

There's one important caveat, and Tate was upfront about it: "Your computer has to be switched on and it has to be open." There's a setting to prevent the computer from going to sleep while tasks are running, and Tate hinted that cloud-based execution is coming, but for now, your machine needs to be awake. Several attendees noted that this is driving sales of Mac Minis as dedicated automation machines.

Live demo: building a CRM dashboard in six minutes

The centerpiece of the session was a live quarterly pipeline review, pulled directly from HubSpot. Tate ran a scheduled task that kicked off a Q1 review, and attendees watched it work in real time.

Cowork identified the HubSpot connection it needed, searched the CRM for contacts, deals, and accounts, wrote itself a to-do list of steps, and began building an interactive HTML dashboard. The whole process took roughly five to six minutes.

"Normally, you'd have to get a RevOps person to go through the data to build a dashboard like this," Tate said. "And dashboard tools in CRMs are notoriously hard to manage."

The output was impressive even to Tate. It surfaced not just win/loss data but specific risk signals buried in the pipeline data. "My champion was dark for four weeks after the org restructure and there's no economic buyer," the dashboard flagged for one deal. Another: "14 weeks stall, this is a zombie. It shouldn't even be in there." A third: "Lost to competitor in February, still showing an open forecast, a false commit by a salesperson."

"I didn't expect it to do this," Tate admitted. He'd populated the CRM with detailed activity data, including call logs, email frequency, and messaging patterns, to stress-test whether Cowork would catch the problems. "It's actually done a really good job of finding the holes there."

The expense automation that impressed the accountant

While the pipeline demo drew the live applause, the automation Tate described for his own accounting workflow might be the more relatable use case.

"I take everything that comes into my email inbox and I section it off into sections," he explained. Invoices and receipts get routed to separate folders, tagged, and processed. Attachments get downloaded and organized. At the end of each week, everything gets compiled into a spreadsheet and pushed into double-entry bookkeeping. Anything missing gets flagged in a second workflow. If Cowork can't find or download a receipt automatically, it sends Tate a message telling him he'll need to retrieve it manually.

The workflow has three layers: weekly processing, monthly reconciliation, and exception handling. Tate didn't design it alone. He recorded himself clicking through his manual process, explaining each step in a screen recording. He fed that video into Google Gemini to generate a step-by-step workflow document, then brought that into Claude Chat to design the Cowork automation. "Claude said to me, okay, you need to have three different separate ones. So what I'm saying is I didn't design the workflow. I did that with Claude."

The real validation? "Not long after that, my accountant rang and said, okay, this is great. Because she didn't have to chase me all the time for stuff that I really didn't want to do."

‍

Skills came up repeatedly during the session, and Tate was clear about what they are: a set of written instructions that tell Cowork how to execute a specific kind of task. "If there's a certain thing that you want to be done in a certain way, you would say, OK, this is the definition of how I want my document to be," he explained. "Skills are very similar in that you say, OK, I want you to go through these steps."

The practical examples made the concept click. "You can have a skill that tells it how to write in Slack and a skill to tell it how to format for WhatsApp. And you don't have to write those skills. You can have AI write those skills. You can say, research WhatsApp formatting for me and write a skill so that I just need to call that skill for writing WhatsApp messages."

Cowork also ships with pre-built plugins that bundle skills, connectors, and instructions into a single package. Tate highlighted the legal plugins, which Anthropic released alongside improvements to Claude's legal reasoning. "These skills are extremely good, and they're written by lawyers, and you can apply them to your business." He walked through the NDA triage, legal risk assessment, and contract review skills, noting that each one can be customized. "You can say, I want you to do the legal review process, but I want you to use Singapore applicable law, or I want to use English applicable law."

The sales plugins also caught attention. Tate's pipeline review and forecast skills were so well-written out of the box that he didn't modify them at all. "I read it. No, perfect. That's exactly the forecast I want."

One attendee, Simon, raised a point about managing 73 skills across Claude Code and Cowork. The current limitation is that skills aren't automatically shared between the two environments. Tate's advice: "Just ask it. Say, can you see the skill? Yes, I can see the skill. Can you use the skill?" The skill file just needs to be accessible on the file system. No manual import required if it can read the file.

Dispatch: controlling your desktop from your phone

Tate demonstrated Dispatch, the feature that connects the Claude mobile app to your desktop Cowork session. He showed both his iPhone and desktop side by side, then typed a command on his phone: "Can you open WhatsApp?"

The demo was a little bumpy, as Tate predicted. Computer use, where Claude physically navigates your screen by taking screenshots and clicking, is still in research preview. "It's a little bit slow because it's doing computer use and taking screenshots of things and then reading those screenshots," he noted. But it worked: Claude minimized other apps, opened WhatsApp, and retrieved messages from the desktop.

The practical point of Dispatch isn't just remote access. "It is to trigger them and also to see if anything, so you can manage things that are running as well," Tate explained. "If you've got things that are running in the background, you can also get to those."

The security reality: what to share and what to protect

An important exchange happened when one attendee, JL, described a moment of panic. She'd given Claude access to Chrome to manage a second Gmail account, and it unexpectedly opened confidential emails from other tabs. "I'm still quite concerned because also there's a warning that says somebody else can come and go into Chrome and do stuff," she said.

Tate was honest about the trade-off. "When it opens a web page and it goes to your email, all of those emails are going to be visible and it's going to take a screenshot. So there's no real way to stop it from doing that." His advice: use a direct connector for Gmail when possible, which avoids screen-based access entirely. For anything that requires browser use, assess the risk yourself.

"There's like personal risk, personal business, and then there's like corporate risk and corporate business," he said. "If it's corporate, I think then it's just like a hard no, unless you get a permission from somebody." He shared his own example of letting Cowork access his Adobe account through the browser to download invoices, acknowledging that the same account contains his credit card information. "You have to kind of assess your own risk."

‍

In the final segment, Tate shared his own approach to file management, which surprised some attendees given his self-described tendency toward hyper-organization.

"What I'm starting to do is think, well, I actually don't need to have all of these like nested folders," he said. "I can have one folder and I can have file names that have got long descriptive file names because I don't have to type them anymore."

He uses Obsidian as his personal dashboard, with a heavily customized setup including terminals, tag systems, and project views. But the underlying files are all Markdown, a format both humans and AI can read. "All it wanted was a flat file with front matter. It was indexing off the front matter and was able to find any file anywhere." For clients, he keeps everything on Google Drive in a locked shared drive that AI doesn't touch. For his own work product (where contracts allow), he opens it to Claude.

The bigger shift, Tate acknowledged, is that he's spending less and less time in any graphical interface at all. "I'm no longer going into UI at all. I'm really trying to avoid it. The information I need, I ask for it either through Cloud Code or through Codex or through ChatGPT or through Claude. And then the information just pops up."

Key takeaways

• Start with one workflow you already do manually, ideally something repetitive like morning briefings, inbox processing, or weekly reporting, and describe it to Cowork in natural language. Let it design the automation steps.
• Use projects to load large or frequently referenced files once, rather than attaching them to individual conversations, so you avoid burning through your token allocation.
• Explore the pre-built plugins (especially legal and sales) before writing custom skills from scratch: they're professionally written and can be customized for your jurisdiction, industry, or formatting preferences.
• Set up Dispatch on your phone and keep your desktop awake so you can trigger and monitor tasks remotely. Consider a dedicated machine like a Mac Mini if you go all-in on scheduled automations.
• Be deliberate about what you give Cowork access to. Use direct connectors instead of browser use wherever possible, and keep client or corporate data behind boundaries you control.

Gary C. Tate is Chief AI Officer at Lead with AI, where he helps executives build practical AI workflows. This article is based on his live session hosted by Lead with AI PRO. Join the membership for access to live expert sessions like this one.