How to Automate 99% of Your Work With Claude Cowork — Full Tutorial
Bart Slodyczka's comprehensive Claude Cowork tutorial covers the full automation stack: writing skills that package recurring tasks, setting up scheduled triggers, and designing multi-step workflows that run without human intervention. The goal is to reduce the portion of your workday that requires you to be at the keyboard — not just to speed up individual tasks, but to eliminate entire categories of them.
"Full Tutorial: How To Automate 99% Of Your Work With Claude Cowork" by Bart Slodyczka — Watch on YouTube →
Key Takeaways
- Skills are the core unit of Cowork automation. Each skill is a markdown file with instructions; Cowork loads it into context and executes the defined steps. The more specific the skill, the more reliably it runs without supervision.
- Skills should be task-scoped, not tool-scoped. Instead of "use Gmail skill", write "draft follow-up emails to unresponsive leads after 3 days" — a concrete job with clear success criteria.
- Scheduled tasks let Cowork fire a skill at a set time or interval. A morning briefing skill, a daily expense-logging skill, and a weekly report skill together can save 1–2 hours per day.
- Chain skills together by having one skill's output feed into the next. Example: scrape leads → score leads → draft outreach → post to CRM. Each step is a skill; Claude Cowork handles the handoffs.
- Memory across sessions is handled by writing state to a file and reading it back at the start of the next run. Cowork's file-read/write tools are the key to persistent context.
- Test new skills manually before scheduling them. Run edge-case inputs, check outputs, and refine the skill instructions until it fails gracefully on bad input rather than producing wrong output silently.
Workflows Covered in the Tutorial
- Email triage: Read inbox, categorise by priority, draft replies for routine messages, flag urgent ones for human review.
- Content pipeline: Pull trending topics from RSS, draft social posts, schedule to Buffer or Hootsuite.
- Lead enrichment: Take a list of company names, look up employees and roles, write personalised first-touch emails.
- Daily standup: Check git history, open pull requests, and calendar, then write a 3-bullet status message to Slack.
Related on OpenClawDatabase
- Claude Cowork Skills Guide — how to write effective skill files from scratch
- Claude Cowork Setup — initial installation and API key configuration
- Claude Cowork Projects — organise related skills under a shared project context
What you can actually set up from this
Extracted from the video's own transcript — the specifics the original summary left out.
Reproducible steps
- The desktop app is the only way in
Cowork needs local file access, browser control and tool connectors, and none of that is available in the browser — so the desktop app is a hard requirement rather than a preference.
- Budget for outgrowing the entry plan
The honest advice: start on the $20 monthly plan and expect to upgrade as you get better at Cowork and start using Code alongside it. The presenter moved up a tier and considers it worth it.
- Change how you talk to it — the chatbot habit is the problem
<strong>The central reframe.</strong> Chat is for "shower thoughts" — unrelated questions where you want an answer. <strong>Cowork is a virtual assistant, and you would not hire one to check the weather.</strong> You would have them draft emails and push tasks forward. The way you phrase requests has to change accordingly.
- Know which of the three products fits the job
A clean split worth stealing: <strong>Chat</strong> for questions, <strong>Cowork</strong> for personal work on your own files, <strong>Code</strong> for business projects — the example given is building an internal invoice-parsing pipeline as a proper script in Code rather than burning tokens on it repeatedly in Cowork, for both cost and privacy reasons.
- Start work from the projects panel, not the task box
Both routes work and one is better. A project is a folder on your computer dedicated to that work, starting empty and accumulating everything you and the agent produce. <strong>Starting a genuinely new piece of work from the projects panel</strong> is the recommended path.
Gotchas
- Cowork has access to local files and can control your browser. That is the point and it is also the risk surface — scope projects to folders you are comfortable with.
- The cost warning is real: doing repetitive parsing work through an agent burns tokens every run, where a script written once does not. The presenter names this as the reason to move a finance pipeline into Code.
- Plan and interface details are from April 2026 and have moved.
- Connector access (mail, task management) means the agent reads real business data. Worth deciding deliberately rather than during onboarding.
More Claude Cowork news
The Overnight Agent Loop: How a CTO Ships 2–10 PRs a Week
Claude Managed Agents: three primitives for production agent infrastructure
Claude Cowork Is a Game Changer — If You Use It Correctly
12 Claude CoWork Skills That Save 10+ Hours a Week
Build a Live Data Dashboard in Claude Cowork in Under 3 Minutes
Claude Managed Agents Add Dreaming, Outcomes, and Multi-Agent Orchestration
← Back to News digest · See also: Claude Cowork guide