How to Use Claude Cowork – Full Workflow Automation Guide 2026
AI Master · 2026-04-16 · 16м 12с · 195 398 просмотров · YouTube ↗
Топики: launch-ai-content, ai-loop-engineering
Аудио ещё не скачано.
📝 Summary
model=openai/gpt-oss-120b · prompt=summary-v7 · 4 882→1 819 tokens · 2026-05-28 08:30:06
🎯 Главная суть
Claude Cowork — это режим, в котором ИИ работает как «сотрудник», автоматически читая, изменяя и сохраняет реальные файлы на вашем компьютере, выполняя полные рабочие процессы без вашего вмешательства.
Claude Chat vs Cowork vs Code
- Chat — отвечает на вопросы, но результаты нужно копировать вручную.
- Code — пишет скрипты и программы, требуя навыков программирования.
- Cowork — читает и пишет файлы, создаёт готовые Excel‑таблицы, Word‑документы, PowerPoint‑презентации и т.п., позволяя вам заниматься другими делами.
Когда использовать Cowork
- Сложные задачи, требующие полного выполнения от начала до конца.
- Любая работа с локальными файлами (создание, изменение, сохранение).
- Повторяющиеся операции, которые вы делаете еженедельно.
- Комбинация работы с файлами и онлайн‑исследований.
В этих случаях Chat — не тот инструмент.
Настройка рабочего пространства
- Создайте отдельную папку (например, co‑workstation).
- Внутри неё три подпапки:
- context — файлы, которые ИИ читает каждый запуск.
- projects — активные задачи.
- output — готовые результаты.
- В context разместите три markdown‑файла:
about_me.md— роль, бизнес, аудитория (2 параграфа).brandvoice.md— тон, стиль, используемые и запрещённые слова.workingpreferences.md— предпочтения по структуре документов, как обрабатывать неоднозначность (сам принимать решение или спрашивать).
Эти файлы читаются автоматически, поэтому повторно их описывать не нужно.
Глобальные инструкции
В настройках Cowork задайте Global Instructions: тон, формат вывода, типичные правила. Это постоянный «бриф» для всех будущих задач, аналогичный инструкциям новому сотруднику.
Пример 1: обработка клиентских транскриптов
Запрос: «Обработай шесть записей встреч, создай для каждого клиента Excel‑трекер (задачи, ответственные, дедлайны, приоритет) и короткое письмо‑подтверждение (<150 слов)».
Claude одновременно читает все файлы, извлекает даты («конец следующей недели» → реальная дата), формирует таблицы с условным форматированием и добавляет итоговую строку. Письма включают конкретные детали разговора и готовые к отправке. Весь процесс занял ~3 минуты вместо 2–3 часов ручного труда.
Пример 2: исследование YouTube‑ниши AI‑инструментов
Запрос: «Найти топ‑10 видео за последние 30 дней в нише AI‑tools, собрать название, количество просмотров, тип «хука» и тему; выявить три контент‑пробела с высоким спросом и низкой конкуренцией; оформить в Word‑отчёт с executive‑summary».
Claude открывает браузер, собирает данные, сравнивает их, создаёт таблицу, классифицирует хуки (curiosity gap, shock stat, before/after) и пишет аналитический отчёт. Время выполнения ≈ 12 минут вместо 2 часов. Добавив одну строку к запросу («какой из пробелов лучше подходит моему каналу, учитывая файл about_me.md»), система сразу предлагает наиболее релевантный вариант.
Ограничение — отсутствие памяти уровня задачи
Claude сохраняет общие настройки (tone, preferences), но не хранит контекст конкретного проекта между запусками. При повторном запуске той же задачи ИИ начинает «с нуля», поэтому каждый раз приходится повторно задавать детали.
Проекты и навыки как решение
- Проекты — не просто контейнеры файлов, а хранилище инструкций, контекста и истории выполненных задач. При создании проекта (например, client deliverables) указываются правила:
- использовать бренд‑голос из
brandvoice.md; - письма ≤ 150 слов, профессиональный, но разговорный тон;
- порядок колонок в Excel‑трекерах (Task, Owner, Deadline, Priority, Status).
После первого запуска проект запоминает полученные результаты; при следующем запуске новые входные данные обрабатываются с учётом уже существующей структуры и правок.
- использовать бренд‑голос из
- Навыки (skills) — файлы‑инструкции, определяющие, как выполнять конкретный тип работы (Excel, PowerPoint, Word, PDF). При запросе создания таблицы автоматически подгружается Excel‑skill, что даёт готовые формулы, графики и форматирование. Пользователь может создать собственный skill (например, «content brief») и тем самым получать документы, полностью соответствующие своему стилю и структуре.
Планирование и удалённый запуск
- В боковой панели задайте Schedule: укажите запрос, частоту (ежедневно, еженедельно) и проект.
- Пример ежедневного задания (9 ч):
- проверка Gmail за последние 12 ч;
- извлечение задач из календаря;
- сохранение сводки в
morningbrief.mdв папку daily briefings.
- Пример еженедельного задания (понедельник, 12 ч):
- поиск трендов AI‑продуктивности за последние 7 дней;
- определение трёх контент‑возможностей;
- сохранение отчёта в проект content и публикация резюме в Slack.
Задачи выполняются только когда компьютер включён и открыт Claude Cowork; каждый запуск учитывается в лимите использования.
Dispatch позволяет отправлять запросы с телефона: вы пишете короткую команду в приложении, она попадает на ваш десктоп, где уже загружены проект и предпочтения, а после завершения приходит push‑уведомление с готовыми файлами. Это делает возможным запускать сложные воркфлоу из любой точки без необходимости писать код или использовать API.
📜 Transcript
en · 2 559 слов · 41 сегментов · clean
Показать текст транскрипта
You've been using Claude wrong, and while you're stuck copy-pasting in chat mode, someone else already has it doing 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 is for developers, writing scripts, building products. powerful but you need technical skills co-workers 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 send real powerpoint decks with proper formatting not text you copy into a file Finished files. Here's the simplest way to think about it. Chat is an assistant who answers questions. Cowork is an employee who completes tasks. When should you use Cowork instead of Chat? Four situations. Complex tasks you want handled start to finish. Anything that touches your local files directly. Repetitive work you do 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 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-workstation. Inside it, three subfolders. 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 your context files this is the real unlock inside the context folder create three markdown files about me.md your role your business your audience what you care about two paragraphs is enough Brandvoice.md. Your tone, your 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 3. Global instructions. Go to settings, co-work. global instructions set your defaults tone roll 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 do. I have six client meeting transcripts. Discovery calls, strategy sessions, follow-ups, messy text files, crosstalk, unclear action items, vague deadlines like sometime next week, what I need, a project tracker, and a follow-up email for each client. Here's the prompt. Process these six meeting transcripts. For each client, create 1. An Excel project tracker with tasks, owners, deadlines, and priority levels. 2. 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. 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. 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, 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 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 tools 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 screen. 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 dock and a headache here's the prompt search YouTube for 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 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 About Me 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 show 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 good 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 Cowork 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 Coworkstation is running on. But most people don't really use them as projects. They treat them like a file container. place to drop inputs and get outputs and because of that every run is still isolated project is not just where files live it's where your instructions your context 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 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 the platform 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, Veo, Clang, Nano, Banana, all the cutting edge models in one place. The community is where real 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 workflow. Co-work ships with built-in skills for Excel, PowerPoint, Word, 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 you've 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 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 9am. 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 attention. Every Monday at 12 a.m., search trending topics and AI productivity from the past seven days, identify three content opportunities based on search volume and engagement patterns, save research brief to content project. post summary to slack my content calendar fills itself three things to know before you schedule everything tasks only run while your computer is awake and claw desktop is open each task 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 to finished files. Setup takes two minutes. Open Cowork 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 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, clawed, auto routes to the write mode, coding 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 intervention and understanding of code. 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 a clean list of my unfinished content tasks, and it completely broke. So I rebuilt the same thing in Software. Software 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 co-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 review and approve content these are my content managers they can submit and edit I can 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 a 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 at a dedicated folder, add instructions, conversational tone, pattern interrupt hooks, problem first sections, spoken paragraphs, never bullets. Step two, drop my aboutme.md and brandvoice.md into the project. Step three, connect Gmail, calendar, Slack, Notion. Step four, set up two scheduled tasks, morning briefing, daily 9am check gmail calendar notion summarize what's on my plate save as morning brief md content research monday 12 p.m search trending ai automation topics from past seven days identify three video opportunities save research brief to project post summary to slack i come back to a script outline that matches my voice follows my structure references the morning's research not a chatbot not a tool a working system free things to walk away with 1. Setup is everything. Your folder structure and MD files are non-negotiable. 5 minutes now saves you hours every single week. 2. 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. 3. 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 30 off and i'll see you in the next one
⚙️ Pipeline jobs
| Stage | Status | Att. | Updated | Error |
|---|---|---|---|---|
| download | done | 1/3 | 2026-05-28 08:29:32 | |
| transcribe | done | 1/3 | 2026-05-28 08:29:50 | |
| summarize | done | 1/3 | 2026-05-28 08:30:06 | |
| embed | done | 1/3 | 2026-06-30 06:43:09 |
📄 Описание YouTube
Показать
#sponsored Start building with Softr! https://softr.io/build-with/arthur-winer?utm_source=ai_master&utm_medium=influencer&utm_campaign=ai_co-builder&utm_content=ai_master_march_launch_yt_integration1 🧠 Join 10,000+ Claude builders inside AI Master Hub → https://aimaster.me/yt/cowork 🎬 This video was made inside AI Content Engine, get the same pipeline → https://aimaster.me/ Claude Cowork turns Claude from a chatbot into an AI employee that works with your actual files, browser, and apps. This full tutorial covers everything — setup, folder structure, context files, and the workflows that make it actually useful. I walk through real demos: processing six meeting transcripts into project trackers and follow-up emails, running full YouTube niche research that outputs a formatted Word report, building custom skills, and setting up scheduled tasks that run automatically every morning. The setup section alone will change how Cowork performs for you — most people skip it and wonder why their outputs feel generic. ⏱️ TIMESTAMPS: 00:00 - Intro 00:28 - What is Claude Cowork 01:39 - The Setup That Changes Everything 03:00 - Meeting Transcripts Demo 04:53 - Live Demo: Research to Report 06:38 - Projects & Persistent Memory 09:37 - Skills: Repeatable Playbooks 10:54 - Scheduled Tasks & Dispatch 14:26 - Full System Walkthrough #ClaudeAI #ClaudeCowork #ClaudeDesktop #ClaudeTutorial #ClaudeAgent #AnthropicClaude #ClaudeComputer