0:00 You've been using Claude wrong, and while you're stuck copy-pasting in chat mode, someone else already has it done their actual work. Let me show you how. When you open Claude desktop, you see three tabs: chat, co-work, and code. Most people live in chat. You ask, it answers, you copy the text, and manually do something with it. Fine for quick questions, terrible for real work. Code [music] is for developers, writing scripts, building products. Powerful, but you need technical skills. Co-work is neither of those. It's not a chatbot, it's not a coding tool. It's an AI employee that operates your computer, reads and writes your actual files, and completes entire workflows from start to finish while you do something else. Real Excel spreadsheets with working formulas, real Word documents ready to
0:47 send, real PowerPoint decks with [music] proper formatting, not text you copy into a file. Finished files. Here's the simplest way to think about it. Chat [music] is an assistant who answers questions. Co-work is an employee who completes tasks. When should you use Co-work instead of chat? Four situations: complex [music] tasks you want handled start to finish, anything that touches your local files directly, repetitive work you do [music] every week, and anything that combines file work with online research. Hit any of these, chat is the wrong tool. Here's where most people go wrong. They skip setup and wonder why their outputs feel generic. This section is the most important in the entire video. Do this right, and Co-work behaves like a
1:33 trained employee. Skip it, and you'll get AI slop every single time. Step one, create a dedicated working folder. Don't point Claude at your entire documents folder. That's chaos. Create one folder, I call mine Co-work Station. Inside it, three subfolders: >> [music] >> context, projects, output. Context holds what Claude needs to know about you. Projects is for active work. Output is where finished deliverables land. Step two, [music] your context files. This is the real unlock. Inside the context folder, create three markdown files: aboutme.md, your role, your business, your audience, what you care about. Two paragraphs is enough. brandvoice.md, your tone, your
2:18 style, examples of how you actually write. Include words you use, include words you never use. workingpreferences.md, how you want documents structured, how Claude should handle ambiguity, should it make a call or ask you first? Claude reads these automatically at the start of every session. You never re-explain yourself, ever. Step three, global instructions. Go to settings, co-work, global instructions. Set your defaults, tone, role, output format. Think of it as the standing brief you give a new hire on day one. You write it once, it applies to everything. Five minutes of setup, weeks of friction removed. Let me show you something real, something your business would actually pay someone to
3:03 do. I have six client meeting transcripts, discovery calls, strategy sessions, follow-ups. Messy text files, crosstalk [music] unclear action items, vague deadlines like sometime next week. What I need, [music] a project tracker and a follow-up email for each client. Here's the prompt: process these six meeting transcripts. For each client, create one, an Excel project tracker with tasks, owners, deadlines, and priority levels. Two, a follow-up email under 150 words confirming next steps. Use professional but conversational tone. Save all files to the output folder. Watch what happens in the task plan on the right. Claude doesn't process these one by one, it breaks this into parallel workflows,
3:50 reading multiple files simultaneously, extracting different data types, cross-referencing information across transcripts. Three minutes later, here's the output. The Excel files have proper column structure, deadlines are extracted from context when someone said end of next week, Claude calculated the actual calendar date, conditional formatting on priority levels, a summary row at the bottom of each tracker. The follow-up emails reference specific discussion points from that client's call. They confirm action items with real deadlines. Under 150 words, ready to send. This is two to three hours of manual work done in three minutes. Now, here's what I want you to notice. I didn't write complicated prompt. I didn't use any special syntax. I
4:35 described the output I wanted like I was briefing a person, and got back finished files. That's the shift. You stop thinking what can AI generate for me, and start thinking what work can AI complete for me. Let me show you a workflow that would normally take an entire afternoon. I want to understand what's working on YouTube right now in the AI tool space, not my channel, the whole niche. Top videos from the last 30 days, what hooks they're using, what topics are getting traction, where the gaps are. This is the research I do before writing every single script. Normally, this is two hours of tab switching, open YouTube, search, scroll, open videos, check view counts, take notes, repeat. By the end, you've got a messy doc and a headache. Here's the prompt: search YouTube for
5:22 top-performing videos in the AI tools niche from the last 30 days. For each of the top 10 videos, extract the title, view count, hook style, and main topic. Identify three content gaps, topics with high search interest but low competition. Save as a research report in Word format with an executive summary. Save to output folder. Watch the task plan. Claude is opening web, running searches, pulling video data, cross-referencing across results. It's doing in parallel what would take me two hours of manual work. 12 minutes later, here's the Word document. Top 10 videos ranked by views. Hook style categorized for each one: curiosity gap, shock stat, before-after. Three content gaps identified with reasoning, executive
6:08 summary at the top that reads like a strategist wrote it. Now, watch this. I add one line to the prompt and run it again. Also, suggest which of these three gaps best fits my channel based on my aboutme.md file. Same research, but now Claude cross-references the opportunities against my audience, my style, my track record. It doesn't just find gaps, it tells me which one is mine to take. That's not a search tool, that's a research analyst who already knows your business. Here's the issue with everything I just showed you. Yes, Claude remembers your general setup, your tone, your structure and preferences, but it doesn't retain the actual work context, what you did in the previous run, what decisions were made, what you adjusted in the output, or what
6:53 good [music] looked like for this specific task. So, when you run the same type of task again, even with similar inputs, you're still effectively starting from scratch at the task level. There's no continuity, no accumulated context, and no improvement based on previous outputs. That's where most people get stuck, assuming that having a good setup is enough, and treat Co-work like a smarter version of chat with file access. But without persistent task level memory, you're still re-running the same workflow each time instead of building on top of it. Projects are already part of the setup, that's what Co-work Station is running on. Most people don't really use them as projects, they treat them like a file container, a place to drop inputs and get outputs. And because of that, every
7:39 run is still isolated. Project is not just where files live, it's where your instructions, your context, [music] and your work history come together. Let me make that explicit. I'll create a new project called client deliverables, and point it to my client folder. Then I add instructions: use brand voice from context files, emails always under 150 words, professional but conversational, Excel trackers use this column order: task owner deadline priority status. Now, I run the same transcripts, same prompt, but inside this project. The spreadsheet follows my column order automatically. The emails match my tone. The formatting is consistent. I don't need to restate any of it. And the important part is what happens next.
8:25 When I run this again with new inputs, the project already contains the previous work, the outputs, and any adjustments I made. Let me make one adjustment for test on follow-ups. So, instead of starting from scratch, it builds on top of what's already there. Over time, it becomes more aligned with how I actually work. That's the difference. Not just running tasks with context, but working inside a system that accumulates it. One project per recurring workflow: client deliverables, content production, admin and finance, weekly reporting. Each one becomes more useful the more you use it. At that point, this stops being a tool, it becomes infrastructure. Now, if you want to actually build systems like this, not just copy prompts, that's what we focus on inside AI Master. I built a platform
9:11 because I was tired of jumping between tutorials without understanding the methodology. We teach you how to think about automation, how to structure workflows. The Pro Plan includes 200 plus lessons and immediate practice in our AI studio. You've got Claude production agents, Gemini assistant VO, Claude nano banana, all cutting-edge models in one place. The community is where real [music] learning happens. 5,000 plus people sharing actual systems that work. We've got 30% off annual plan right now. When you're ready to go deeper, links in the description below. Projects give you memory, skills give you repeatability. A skill is an instruction file that tells Claude exactly how to handle a specific type of work, a detailed playbook for one
9:57 workflow. Co-work ships with built-in skills for Excel, PowerPoint, Word, [music] and PDFs. When you ask Claude to build a spreadsheet, it automatically pulls the Excel skill, and you get proper formulas, charts, and formatting instead of a data dump. But, custom skills are where this gets powerful. I'll ask Claude to write a content brief without a skill. You'll get a generic structure, predictable sections, no personality. Looks like every AI output ever seen. Now, with my custom content brief skill loaded, completely different. The structure matches what actually works for my audience. The tone is mine. The research prompts are specific. It reads like something I wrote, not something a machine generated. Build a skill once for any
10:44 workflow you repeat. Weekly reports, sales proposals, SOPs, onboarding documents. You write the playbook once, Claude follows it every time. This is where Co-work becomes genuinely dangerous for anyone not using it. Write a prompt once, set a schedule, Claude runs it automatically. No code, no APIs, no third-party tools, just a prompt and a time. Click schedule in the sidebar, new task, write your prompt, pick frequency, save. Here's what I have running. Every morning at 9:00 a.m., check Gmail for unread messages from the last 12 hours, check calendar for today's events, extract action items and deadlines, save as morningbrief.md in daily briefings folder. I wake up to a summary of what actually needs my
11:29 attention. Every Monday at 12:00 a.m., search trending topics in AI productivity from the past 7 days, identify three content opportunities based on search volume and engagement patterns, save research brief to content project, post summary to Slack. My content [music] calendar fills itself. Three things to know before you schedule everything. Tasks only run while your computer is awake, and Claude desktop is open. Each task [music] counts against your usage. Keep prompts tight, no fluff. Always test manually before scheduling, but when it's running, you're not doing that work anymore. Someone else is, and that someone works for free. What if you're not at your desk? Dispatch lets you text Claude a task from your phone. It runs the full workflow on your desktop. You come back
12:16 to finished files. Setup takes 2 minutes. Open Co-work on your phone. Enable dispatch. Done. The task hits my desktop. Claude opens the project, which already has my preferences loaded. It pulls the data, runs the comparison, generates the summary, saves it. Push notification when done. Your phone is the remote control. Your desktop [music] does the heavy lifting. I can trigger full research reports from the gym, content audits from coffee shops, client deliverables from my car before meetings. Claude [music] auto routes to the right mode. Code and tasks go to code, knowledge work goes to Co-work. You don't choose, Claude figures it out based on what you asked. Look, Claude's new Co-work feature is genuinely impressive, but it still requires your
13:03 intervention, understanding of code. [music] And that's how most AI works today. Just recently, I was trying to build a simple content calendar for myself in Replit, and it looked fine until I tried to do one simple thing. Just see clean list of my unfinished content [music] tasks, and it completely broke. So, I rebuilt the same thing in Softr. Softr is an AI app builder for real business software. Not just generating code, but actually building your full system, database, pages, user roles permissions workflows all connected. I go into the AI code builder and type this, and instantly, I get a clean, structured view of everything that's not done. And here's the part I couldn't do before. I open my users, and I can literally see everything. These are my campaign managers. They can
13:49 review and approve content. These are my content managers. They can submit and edit. change access in a click. And if I switch to mobile, the whole interface just adapts automatically. But, you can also visually edit using the no-code interface. So, you're not stuck with code and debugging, and this isn't just for content tools. You can use this to build things like an entire ERP system, client onboarding portal, document management system. Basically, anything your business needs to run. At that point, you're not just building apps, you're building real software, and link is in the description. Let me show you my complete YouTube content system running live. Step one, create YouTube content project, point it at dedicated folder, add instructions, conversational
14:36 tone, pattern interrupt [music] hooks, problem first sections, spoken paragraphs, never bullets. Step two, drop my aboutme.md and brandvoice.md into the project. Step three, [music] connect Gmail, calendar, Slack, Notion. Step four, set up two scheduled tasks. Morning briefing, daily 9:00 a.m. Check Gmail calendar Notion summarize what's on my plate, save as morningbrief.md. Content research, Monday, 12:00 p.m. Search trending AI automation topics from past 7 days, identify three video opportunities, save research brief to project, post summary to Slack. I come back to a script outline that [music] matches my voice, follows my structure, references the
15:22 morning's research. Not a chatbot, not a tool, [music] a working system. Three things to walk away with. One, setup is everything. Your folder structure and MD files are non-negotiable. 5 minutes now saves you hours every single week. Two, use projects for anything you do more than once. The memory component is what separates Co-work from every other AI tool out there. It compounds, gets better, starts to feel like a real employee. Three, scheduled tasks are the real unlock. This is how you go from using AI to having AI work for you. That's a different category entirely. And remember, knowing structure is one thing. Practicing them is another. That's what AI Master is built for. Learn, practice, repeat. Link below with
16:07 30% off. I'll see you in the next one.
