The L badges are this site's skill levels, from L1 (beginner) to L6 (advanced). Follow the modules in order; you do not need to know your level to start.
L1–2 · Free
What happens when Claude runs without you
See the Scheduled Tasks panel for the first time and understand what it actually does
'Routines' just means Claude doing work on a clock instead of waiting for you to ask. There are three ways to actually run one: Cowork Scheduled Tasks (what this track teaches), Desktop Routines, and Cloud Routines. This module shows you all three, who each is for, and why Cowork Scheduled Tasks is the right place to start. There is nothing to set up yet: this module is orientation, not configuration.
Worth knowing:If you take one thing from this module: start with Cowork, not Desktop Routines or Cloud Routines. Cowork runs on the computer you already use, reads the files and apps you already connected, and needs no developer setup, so it is the one a non-developer should learn first. Desktop Routines are the natural next step; Cloud Routines come much later.
Anyone who has seen the Scheduled Tasks button and wondered what it doesAnyone finishing the Cowork track and ready for what comes next
L3
Your first scheduled briefing
Claude reads your calendar and tells you what to prepare, before you open your laptop
Your first scheduled briefing will run without errors and still disappoint you. This module is about why that happens, and the one change to your prompt that fixes it. You point a daily briefing at Google Calendar, set it for weekdays at 8am, test it with Run Now (the test-fire button inside Scheduled Tasks), then watch a prompt that works fine in a live chat fall apart the moment it runs alone. Cowork taught you to make a scheduled task exist; this is where you learn to make one you can trust.
Worth knowing:The failure mode is almost always the same: the prompt assumes Claude can ask you clarifying questions. It cannot. This module teaches you to pre-answer every question Claude would have asked.
Anyone who recaps their calendar every morning before starting workPeople who want their first automated output by end of the module
L4
Writing prompts Claude can follow alone
Writing for a Claude that runs at 6am without you is a different skill than chatting
Interactive prompts can be vague because you're there to clarify. Scheduled prompts fail when they're vague because no one is there. This module teaches three prompt patterns specific to autonomous tasks: self-contained context, time-aware guardrails (if it's after 6pm, skip today), and explicit output format instructions. Includes templates for morning briefings, weekly reports, and recurring research.
Worth knowing:One sentence prevents 30 minutes of debugging: 'If you cannot find the relevant data, write a single line saying so and stop.' Add this to every scheduled prompt you write.
Anyone whose first scheduled task produced confusing or incomplete outputAnyone going from 'it sometimes works' to 'it reliably works'
L4
Combining apps in one task
A morning briefing that reads three sources and tells you the one thing that matters
One-app tasks are reliable but limited. The bigger win is combining sources: reading Gmail, Calendar, and Slack in one pass and producing a single, unified picture of your day. Each app you link to Claude is a connector (the bridge to an outside app), and this module walks through building a three-source briefing. It also explains how each connector Claude checks counts against your usage quota (your Claude usage limit), and how to keep the task small enough that Claude can hold it all in mind at once.
Worth knowing:The hardest part of a three-app task is not connecting the apps. It's telling Claude how to combine what it finds. 'Tell me what I should focus on today' is too vague to act on; this module gives you the wording that actually works.
Anyone with Gmail, Calendar, and Slack connected to ClaudeAnyone whose single-connector briefing is no longer enough
L4–5
Making Claude remember across runs
Claude remembers what it found last week, so this week builds on it
By default, each scheduled run is a fresh session, so Claude does not remember what it found last week. This module teaches two ways to fix that. The first is a folder Claude keeps between sessions (called Cowork Projects), so it can pick up where it left off. The second is a notes file Claude reads at the start of each run and writes back to at the end, which works for any task, not just Projects. Neither is unlimited: project memory gets summarized as it grows, and a notes file that never gets trimmed will eventually crowd out the run itself. So part of the skill is deciding what is worth keeping.
Worth knowing:After a few weeks of runs, a research task starts connecting something it noted a month ago to something new today. That only happens when each run can read what past runs wrote down.
Anyone running weekly research or competitive monitoring tasksAnyone who wants their scheduled task to get smarter over time
L4–5
When scheduled tasks break
The runs that fail quietly, and the ones that fire at the wrong time
Scheduled tasks fail in ways that are easy to miss. Your laptop is closed at 8:59am, the run is skipped, and Claude fires one catch-up when the computer next wakes, which can land at 11pm instead of morning. Other tasks get stuck in Running and block the queue behind them. A connector quietly expires and the task keeps running on dead permissions. This module shows you how to spot each one, and how to write prompts that survive a late or missed run. The tell: a skipped run leaves a notification and shows in history, but the quiet failures leave nothing, so you have to design for them.
Worth knowing:The most useful guardrail you can add to any scheduled task: 'If it is after 6pm, write one line saying this ran late and stop.' That turns a stale catch-up run into a safe, known state instead of a wrong-time surprise.
Anyone whose scheduled tasks have started producing wrong or empty outputAnyone debugging why a task ran at the wrong time
L5
Getting results pushed to your phone
Your briefing is useless in a sidebar you are not looking at
By default, a scheduled task's output lands in your Claude Desktop sidebar, fine at your computer and useless when you're not. There are two ways to get it to your phone. On a Pro or Max plan (a paid Claude subscription, separate from this course), Cowork keeps a running conversation you open on your phone to check what ran, with nothing to set up. The other way, Channels, actively pushes finished output to your phone, but it asks for real setup in return.
Worth knowing:The built-in phone conversation is a read-it-when-you-open-it surface. Channels is the only way to get output pushed to you (to Telegram, Discord, or iMessage on a Mac), and it is part of Claude Code, Anthropic's developer tool. So this is the one place the track crosses into developer territory, and wiring it up is real setup, not a toggle.
Anyone whose briefing is stuck on a desktop they walk away fromAnyone ready to try Claude Code for the first time
L5–6
Knowing the ceiling
What scheduled tasks cannot do, and what you would use instead
Cowork Scheduled Tasks are capable, but they only run locally. The computer must be awake. The app must be open. Cowork itself has no cloud fallback. This module names those constraints, explains what Cloud Routines do differently (they run on Anthropic's servers even when your laptop is closed), and helps you decide which to use. It also covers when Zapier or n8n (a similar automation tool) is the better tool, what Claude handles that Zapier cannot, and what you give up if you stay in Zapier for everything.
Worth knowing:The deciding factor is reliability. If a missed 8am run is merely annoying, keep the job in Cowork. If a missed run means a client does not get billed, that job belongs in Zapier, not Claude. You will likely end up using both, each for what it does best.
Anyone whose workflow has outgrown Cowork Scheduled TasksAnyone ready to explore developer tooling