The shift from using Claude to scheduling Claude. Build once, run forever: your morning brief, your weekly audit, your competitor check. All without you lifting a finger.
What this episode covers
Routines are prompts on a schedule. You write the prompt once, set the cadence, and Claude runs it automatically: morning briefings, weekly audits, content drafts, memory cleanups.
Episodes 1–3 covered how to make Claude better at one-off tasks. This episode is about the shift to automation: the work that happens before you open the app. By the end, you'll have at least one routine running in the background that saves you time every single day.
Not all schedules are equal. The right one depends on what you need running when.
Claude Code has three ways to automate recurring work. They differ on one key question: does your laptop need to be on?
A repeating prompt that fires on an interval inside the terminal session you have open right now. Type /loop 30m /your-command and Claude reruns that command every 30 minutes. Stop the terminal, the loop stops too.
A scheduled task that survives app restarts and lives in the Routines sidebar of Claude Code Desktop. If you close and reopen the app, the task is still there. If the task fires while the app is closed, it runs the next time you open it.
/scheduleRoutines that run on Anthropic's servers, not your machine. Your laptop can be off, the app can be closed. The routine runs regardless. Available on Pro, Max, Team, and Enterprise plans. Launched April 2026.
| Mode | Laptop must be on? | App must be open? | Survives restart? | Setup difficulty |
|---|---|---|---|---|
/loop |
Yes | Yes | No | Minimal: type the command |
/schedule (Desktop) |
Yes | Yes (when it fires) | Yes | Low: Routines panel |
| Cloud Routine | No | No | Yes | Low: web UI at claude.ai |
Two paths: the desktop sidebar (quick) and Cloud Routines (more powerful). We'll walk both.
In the Claude Code Desktop sidebar, click Routines. If you don't see it, look for a clock or calendar icon. The panel shows all your scheduled tasks and lets you create new ones.
Give the routine a name (e.g., "Morning brief"), pick how often it should run (daily, weekly, specific days/times), and paste in the prompt you want Claude to run each time.
A good routine prompt is exactly what you'd type if you were starting a fresh conversation. Be specific about what you want, what format, and where to save the output (if anywhere). Example:
Check my .memory/projects/project-a.md for any open loops or action items from the last session. Summarize them as a bullet list with the top 3 items I should tackle today. Save to Inbox/daily-brief-$(date +%Y-%m-%d).md
Save the routine. On the next scheduled fire, Claude runs the prompt in an isolated session and saves its output. You'll find the result in your workspace when you next open the app.
Cloud Routines live at claude.ai/code under Routines, not in the desktop app. Setup is similar but requires connecting data sources for Claude to read (since it can't access local files on your machine).
Log into claude.ai in a browser. Select a project or create a new one. Click the Routines tab.
Because this runs in the cloud, Claude can't read your local folder. Connect the tools it needs access to: a GitHub repo (for code), a Notion workspace (for docs), or a Slack workspace (for messages). These are your MCP connectors. More on those in Episode 5.
Same as before: write a clear prompt. Choose a trigger: scheduled time, a one-time date, or an event (like "whenever a new PR is opened on this repo").
Cloud Routines run on Anthropic's infrastructure. Your results appear in the Routine's output log when you next check in. For workflows that post to Slack or write to Notion, the output is already in your tools when you wake up.
.memory/ folder, local documents, etc.). They work with connected sources: GitHub repos, Notion, Slack, Airtable (anything reachable over the internet). If your workflow depends on local files, use a Desktop scheduled task instead.
Copy any of these, swap in your project name, and you have a running routine in under 5 minutes.
Surfaces your open loops and today's context before you even open the app. You arrive at your desk with a one-page status instead of reconstructing where you left off.
Read .memory/INDEX.md and .memory/projects/project-a.md. Identify all open loops and action items. Check my calendar file at .memory/reference/calendar.md for anything scheduled today or tomorrow. Write a morning brief to Inbox/$(date +%Y-%m-%d)-morning.md: - Top 3 priorities for today - Open loops (with a one-line status on each) - Anything time-sensitive in the next 48 hours Keep it under 300 words. Plain bullets, no fluff.
Audits the last 7 days of content across your channels. You show up Monday already knowing what landed, what flopped, and what's queued.
Review the project-a content tracker in Notion (database ID: [yours]). Filter for items updated or published in the last 7 days. Build a weekly content audit: - What shipped (title + channel + status) - Engagement notes if available - What's in-progress or stuck - 3 content gaps I should fill this week based on what's missing Save to Notion page "Weekly Audits" as a new entry dated today.
Surfaces a ready-to-review LinkedIn post draft before your workday starts. You pick the best one and publish, or skip it. Never stare at a blank screen again.
Use the linkedin-content skill. Check my .memory/projects/project-a.md for recent wins, decisions, or insights worth sharing publicly. Draft one LinkedIn post (150–220 words) that leads with a specific observation, not a vague claim. Save to Drafts/linkedin-$(date +%Y-%m-%d).md and mark it DRAFT — awaiting approval. Do not publish. This is a draft only.
Memory files accumulate contradictions, stale facts, and outdated project states. This routine prunes them so your context stays sharp.
Use the anthropic-skills:consolidate-memory skill. Audit every file in .memory/ for: - Facts that contradict each other - Project states older than 30 days that haven't been verified - Entries that reference work that's clearly complete or cancelled - Duplicate information stored in multiple places Write a memory health report to Inbox/memory-audit-$(date +%Y-%m).md: - Files reviewed - Issues found (with file + line) - Changes made - Items I should manually review Make the fixes where safe. Flag uncertain items for my review.
A weekly pulse on what your top competitors are doing. You get a single-page digest every Friday so you can respond in Monday's planning without doing the research yourself.
Search the web for recent activity from [Competitor A] and [Competitor B] in the last 7 days. Look for: pricing changes, new features or announcements, new content or campaigns, job postings that signal direction. Write a competitor digest to my Notion page "Competitive Intel" as a new weekly entry: - One section per competitor - Bullet points only, no padding - Flag anything that warrants a response from us - Rate urgency: High / Medium / Low for each item
Routines are prompts on a schedule. Cowork is Claude doing multi-step work on your behalf.
Cowork is a separate mode in the Claude Desktop app (alongside Chat and Code). It's designed for knowledge workers who want Claude to complete complex, multi-step tasks without touching the terminal. Think of it as the non-technical version of everything in this guide.
Cowork has its own scheduling system: the same desktop scheduler under the hood, but with a point-and-click interface designed for non-developers. You describe the outcome, Claude figures out the steps.
use the linkedin-content skill to... in the prompt and your Routine will invoke it.The mental shift that separates people who get 10x value from Claude from those who get 2x.
Most people interact with Claude reactively. Something comes up, they open a chat, they get help. That's a 2x workflow: you're still driving. The 10x workflow is: you spend 30 minutes building a routine, and Claude does that work every single day without you touching it.
Take the Morning Brief routine above. Setup time: 15 minutes. Run time: 2 minutes per day, automated. If it replaces 10 minutes of your own mental reconstruction each morning:
10 min/day × 5 days = 50 min/week mentally reconstructing where you left off. Compounded over a year: 43 hours of context-reconstruction.
15 min one-time setup. 0 min/day after that. Amortized cost after one week: 3 min/day. After one month: essentially zero.
Multiply this by five routines, and you've reclaimed the equivalent of a part-time employee's week every month, tasks Claude does while you sleep.
The test: could you write instructions for an assistant that would let them do this without asking you a single question? If yes, that's a routine candidate. Common ones: checking a dashboard, drafting a weekly update, pulling a report, reviewing a queue.
Don't write the routine prompt cold. Do the task once in a normal Claude session, note exactly what prompt got you the best output, then copy that prompt into the routine. You'll get a good routine on the first try rather than debugging it over three days.
Automated doesn't mean unsupervised. For the first three runs, read the output and spot-check for quality. A routine that goes wrong quietly is worse than no routine at all. Once it's stable, you can stop watching. It's infrastructure.
Think of routines the way you think of your email filters, your calendar automations, or your scheduled reports. You set them up once. They run. You forget about them, until the day you turn one off and realize how much it was doing.
The most productive users of Claude Code in 2026 don't have more conversations than everyone else. They have fewer, better ones, because the routine work is already done when they sit down.
New practical guides for European SMEs adopting AI, delivered when they drop. No spam, unsubscribe anytime.
Want this done for you, not just explained? Vectimo runs the AI adoption end to end.
Work with Vectimo →