← Back to the Using Artificial Intelligence outline The Course Maker
← Using Artificial Intelligence outline
Week 13 · Lecture outline

Week 13 — Lecture Outline · Claude Cowork III: Scheduled Tasks, Dispatch & Automation

Using Artificial Intelligence · AI 101 Fall 2026 · Prof. Quinn Fictional sample

Course: Using Artificial Intelligence (AI 101) · Silver Oak University (fictional sample) · Prof. Quinn
Objective covered: Objective 6 — Automate real tasks with scheduled tasks and dispatch, and operate cross-app workflows safely.
SLOs touched: A (build agentic workflows that produce verified results) · B (use AI safely and critically, including reviewing automated outputs)
Meeting pattern: Thanksgiving week — 2 sessions × 75 min = 150 min, meeting Mon Nov 23 and Tue Nov 24 only. Segment minutes below total ~150.

Week at a Glance
The week's big question "What does it mean to trust AI to act on a schedule — and what do you need to know before you set it and walk away?"
By the end of the week, students can… (1) create a scheduled task using /schedule or the Scheduled sidebar; (2) state the awake-and-app-open constraint accurately; (3) explain what dispatch is and how it differs from a regular Cowork session; (4) catch an AI over-promising on scheduled task behavior and correct it with the official docs; (5) evaluate an automation plan for reliability flaws and propose fixes.
Key vocabulary scheduled task · dispatch · cadence · /schedule · Scheduled sidebar · skipped run · push notification · awake-and-app-open constraint · end-to-end automation · push technology
Materials slides (Deck 13) · readings (official Anthropic docs on scheduling and dispatch) · Claude Cowork desktop (required for the Studio) · one approved assistant for the tutorial/discussion/assignment
Timing note 8 segments, ~150 min total. Session 1 (Mon Nov 23) = Segments 1–4 (~75 min). Session 2 (Tue Nov 24) = Segments 5–8 (~75 min).

Segment 1 — Hook & the Promise (8 min) · Session 1 opens

Hook. Put this scenario on the slide: "It's 7 a.m. Monday. Before you open your laptop, Claude has already read your class notes folder, summarized what you covered last week, and listed your open action items — waiting for you in your Cowork task history." Ask the room: is this possible today — and what would have to be true for it to work?

Take hands. Most students will either say "yes, because cloud" (wrong about the mechanism) or "no, impossible" (wrong about the capability). The right answer is: yes, with a critical constraint — your computer must have been awake and the Claude desktop app must have been open for that 7 a.m. run to happen.

The promise (write it on the board): "By Tuesday you'll know how to make Claude work on a schedule, exactly when it will and won't run, and how to catch the AI when it over-promises."

Why it matters: This is the shift from reactive AI (you ask, it answers) to proactive AI (it works while you don't). Every professional environment is moving toward this model — you need to understand it accurately, not romantically.

Segment 2 — What Scheduled Tasks Are and the Load-Bearing Constraint (22 min)

Plain language first. A scheduled task is a Cowork task you describe once — with a prompt and a cadence (daily, weekly, hourly, weekdays, on demand) — and Cowork runs it automatically. The output lands in your task history and you review it when you're ready. Common uses from the official docs: a daily briefing (summarize emails, Slack, or calendar from the past 24 hours); a weekly report (compile data from connected tools); recurring research (track topics or industry news); file organization (periodically sort or clean up a designated folder); team updates (generate standups from project management tools).

The load-bearing constraint (say it out loud, write it on the board):

Scheduled tasks only run while your computer is awake AND the Claude desktop app is open.

This is not an edge case or a bug — it's by design. Cowork's computation happens on your machine (with access to your local files and connected apps), not on a remote server. That's what gives it access to your files and apps. The trade-off is that the machine has to be on.

What happens if the machine is asleep at run time: the task is skipped — then runs automatically once the computer wakes up or the app opens. Students see a notification. The skipped run appears in the task's history. So you don't lose the work; you just get it later.

Source (say this out loud and put the URL on a slide): "I'm reading from the official Anthropic documentation — the article 'Schedule recurring tasks in Claude Cowork' at support.claude.com/en/articles/13854387. This is the authoritative source. If any AI ever tells you something different, check here first."

Misconception + cure (do this now):
- ❌ "I set the schedule, so it'll run at 7 a.m. whether my laptop is on or not."
✅ Cure: This is the most common error. Cowork is not a cloud cron job. Your machine must be awake and the app must be open. If you want reliable 7 a.m. briefings, keep your computer plugged in and the app running.

Interaction — quick poll: "Who has a device they leave plugged in and running overnight?" (Most people do. So the constraint is manageable — you just have to know it exists.)

Segment 3 — How to Create a Scheduled Task (22 min)

Two paths, both official (put both on a slide):

Path 1 — /schedule from any task:
1. Open Cowork, click "+ New task" or open an existing task.
2. Type /schedule in the chat input.
3. Claude launches a built-in Skill to set up the scheduled task — it may ask you multiple-choice questions about cadence, what the task should do, and other details.
4. Review Claude's proposed task name, schedule, and description.
5. Click "Schedule" to confirm.

Path 2 — From the Scheduled sidebar:
1. Click "Scheduled" in the left sidebar.
2. Click "+ New task" in the upper right.
3. Fill in: task name, description, the prompt describing what to do, how frequently (hourly / daily / weekly / weekdays / manually), and optionally the model and which folder to work in.
4. Click "Save."

Manage your tasks from the Scheduled sidebar: view all scheduled tasks, see upcoming and past runs, edit instructions or cadence, pause, resume, delete, or run on demand anytime.

Live walkthrough (do this on the projector):
Build a "daily class notes briefing" live. Open Cowork, start a new task, type a prompt like: "Each morning, read the Markdown files in my AI 101 notes folder and produce a 150-word briefing of what I covered that week plus a list of open action items. Do not invent anything — only report what's in the files." Then type /schedule, set to daily, confirm. Now look at what appears in the Scheduled sidebar.

The verify-the-AI moment (preview — full version in Segment 7): "I'm about to ask an AI to help me configure this. Watch what it says about when the task will run."

Segment 4 — Misconceptions + Quick Interaction (23 min) · Session 1 closes (~75 min)

Name the misconceptions, cure each:

  • ❌ "Scheduled tasks run even if my computer is off or asleep."
    ✅ Cure: False. The official documentation is explicit: tasks run only while the computer is awake and the app is open. Skipped runs catch up when the machine wakes.

  • ❌ "Dispatch is the same as a regular Cowork chat."
    ✅ Cure: Dispatch is a persistent cross-device thread — you assign work from your phone or desktop, walk away, and receive the finished result with a push notification. A regular Cowork chat is synchronous — you watch it happen. Dispatch is asynchronous — you check in on the result.

  • ❌ "I can let a scheduled task run money transfers, purchases, or send commitments on my behalf."
    ✅ Cure: Safe-use rule (across all agentic tools): you execute consequential financial or irreversible actions yourself. Automations that summarize, compile, or organize are appropriate. Automations that spend money, send binding commitments, or take actions you can't undo are not — you do those yourself.

  • ❌ "Automations never need review — I set it up and trust it."
    ✅ Cure: Automated outputs inherit AI's failure modes: hallucinated details, missed files, unhelpful summaries. Review the first several runs; iterate the prompt if something is off. The automation handles execution; you handle judgment.

Interaction — Classify the Automation (~10 min):
Show four proposed automations. Students classify each as: ✅ appropriate to automate / ⚠ automate with caution / ❌ do not automate this way.
1. "Summarize my Slack messages every morning." (✅ — compiles information)
2. "Send my weekly report email to my team automatically." (⚠ — review before sending; adds sending step)
3. "Transfer $200 to my landlord every month." (❌ — never automate money movement)
4. "Sort my downloads folder every Sunday." (✅ — file organization)

Segment 5 — Dispatch: Background Work Across Devices (20 min) · Session 2 opens

Hook back in: "Last session: scheduled tasks — work on a clock. Today: dispatch — work in the background, from anywhere."

Plain language first. Dispatch is Claude Cowork's feature for assigning work from your phone or your desktop and receiving the finished result when it's done — without watching every step. You have one persistent cross-device thread with Claude: it doesn't reset between tasks, and Claude retains context from prior work in that thread (it remembers your preferences, your projects, how you like things done).

How dispatch works:
- Message Claude a task from your phone on the way to work.
- Claude spins up the right session on your desktop — knowledge work goes to Cowork; dev work goes to Claude Code.
- Claude works on your computer with access to your local files, connected apps, and installed plugins.
- When done, Claude sends you the result (a spreadsheet, a memo, a comparison table) and a push notification.
- You check in when you're ready — the work is waiting.

The key difference from a normal Cowork task: in a regular task, you're watching. In dispatch, you're not. Claude messages you the outcome, not a live step-by-step view.

Availability: currently in beta for Pro and Max plans. Requires both the Claude Desktop app (on your computer) and the Claude mobile app (on your phone).

The same constraint applies: your desktop computer must be awake and the Claude Desktop app must be open for Claude to do the desktop-side work. From your phone you can assign the task anytime — but the execution happens on your desktop.

Source: official documentation at support.claude.com/en/articles/13947068.

Segment 6 — Building an End-to-End Automation (20 min)

Concrete build together: Let's design a real end-to-end automation and trace every step, thinking through what could go wrong.

Scenario: a student wants a weekly class digest — every Sunday evening, Claude reads their AI 101 notes folder, summarizes the week's concepts, and lists three things they should review before the upcoming week.

Step 1 — Identify the task: knowledge-work, repeating, time-saving, verifiable. Good candidate.

Step 2 — Set up the project: in their Cowork project for AI 101, they've already connected their notes folder (from Week 11). No new setup needed.

Step 3 — Write the prompt: "Every Sunday evening, read the Markdown files in my AI 101 notes folder that were modified this week. Produce: (a) a 150-word summary of the main concepts covered, (b) three specific things I should review before next Monday's class. Base everything only on what's in the files — do not invent or extrapolate."

Step 4 — Schedule it: type /schedule → choose weekly → Sunday → 8 p.m. And note the constraint: the student needs their computer on Sunday evenings. If they're traveling on Thanksgiving, the Nov 23 run might be skipped — it'll catch up when they open the app.

Step 5 — Review and iterate: when the first output arrives, read it carefully. Did it stay within the files? Did it invent anything? Does the "review" list reflect real gaps? If not, iterate the prompt.

The AI-critique moment — do it live: ask Claude to help you write this prompt, then ask it: "Will this task run if my computer is asleep at 8 p.m. Sunday?" Watch what it says. If it says yes, that's the over-promise to catch.

Segment 7 — Technology Workflow: Verify the AI on Automation Claims (18 min)

The full workflow — Cowork automation with a verify-the-AI step:
1. Identify the repeating task — is it knowledge work? Does it repeat? Can the output be verified?
2. Write a precise prompt — describe inputs, outputs, format, and any "only use what's in the files" constraints to prevent hallucination.
3. Use /schedule or the Scheduled sidebar — choose cadence.
4. Note the awake-and-app-open constraint — build it into your reliability planning.
5. Review the first several runs — is the output accurate? Does it hallucinate? Iterate the prompt if needed.
6. Catch AI over-promises — if any tool (including Claude) tells you the task runs while the computer is off, correct it and cite the official doc.

AI-critique moment (required weekly step — do this live):

Ask an approved assistant: "Help me set up a Claude Cowork scheduled task that runs every morning at 7 a.m. to summarize my notes. Will it run reliably even if my laptop is closed?"

Watch the response. The AI may:
- Correctly state the constraint (good — no correction needed)
- Say yes, it'll run reliably, without noting the awake/app-open requirement (over-promise — catch it)
- Say it runs on Anthropic's cloud servers (wrong mechanism — catch it)

When you catch an error, say: "Actually, the official Anthropic documentation says scheduled tasks only run while the computer is awake and the Claude desktop app is open. Please revise your answer." Then verify the corrected answer against the source.

Why this is the course's through-line: the same discipline — embed the claims in context, verify against the official doc, catch the over-promise — applies whether you're checking a factual citation or a product feature. The habit is the same; the domain changes.

Segment 8 — The Push-Technology Future, Callback & Hand-off (17 min) · Session 2 closes (~75 min)

The push-technology future:
- Today: scheduled tasks run locally (your machine must be on) — your data stays local, your files are accessible, the constraint is real.
- The trajectory: more connectors, richer outputs, more reliable scheduling as the technology matures.
- The constant: the trust boundary. Even in a world of fully autonomous agents, you approve consequential actions. No money movement, no purchases, no binding commitments — you do those.
- The skill that travels: describe tasks precisely, verify outputs, catch errors, iterate. The tools change; the discipline stays.

Callback + tease:
- Callback: Three Cowork weeks now: projects + files (11) → skills + connectors + artifacts (12) → scheduled tasks + dispatch (13). Each week added a layer of autonomy. The through-line: more autonomy = more responsibility to verify the output.
- Next week (14): the most ambitious layer — computer use (Claude controls native desktop apps), Claude in Chrome (browser automation with prompt-injection caution), Claude in Excel (the spreadsheet agent), and chaining multi-step cross-app workflows. Safe-use rules become even more critical. Come ready to think about where the approval checkpoints belong.

Hand-off:
- Lecture Tutorial 13 (AI tutor, share link) — scheduled tasks, dispatch, the constraint.
- Practice Exercises 13 (ungraded warm-up) — floor-level reps before Quiz 13.
- Quiz 13 (no AI) — covers the awake/app-open constraint, /schedule, dispatch, and automation concepts.
- Discussion 13 — "should AI act on a schedule without you watching?" + error-analysis of a flawed plan.
- Assignment 13 — design a scheduled task, explain dispatch, critique a broken automation plan.
- AI Build Studio 13 — "Automate a Real Task" — configure or design a scheduled task, catch the AI over-promise, verify against official docs. Due Sun Nov 29.

Instructor FAQ — Common Stumbles

Student says / does Quick cure
"The task will run whether my computer is on or not." Official doc: tasks run only while the computer is awake AND the Claude app is open. Skipped runs catch up when the machine wakes.
Confuses dispatch with a regular Cowork chat. Dispatch = persistent cross-device thread, asynchronous, push notification when done. Regular chat = synchronous, you watch in real time.
Wants to automate money transfers. Safe-use rule: you execute financial/irreversible actions yourself. Never automate money movement.
"I set it up, so I don't need to review the outputs." AI outputs inherit AI's failure modes. Review first several runs; iterate the prompt if something is off.
Can't access scheduled tasks (free plan). Scheduled tasks require a paid plan. Students on free plans can design the task carefully and document what it would do — the design skill is what the week teaches.
"Dispatch means the task runs on Anthropic's servers, so I don't need my computer on." No — dispatch uses your desktop's compute and your local files. Your machine still needs to be on.

Scope flag

This outline covers Objective 6 at the scheduled-tasks-and-dispatch depth: creating and managing scheduled tasks, the awake-and-app-open constraint (official doc: support.claude.com/en/articles/13854387), dispatch as a cross-device background-work feature (official doc: support.claude.com/en/articles/13947068), building a real end-to-end automation, and the "push-technology future" framing. Computer use, Claude in Chrome, and Claude in Excel are Week 14 and only previewed here. Every Cowork feature claim is verified against current official Anthropic documentation; no features, menu paths, or plan details are invented.

~ Prof. Quinn's edition · Fall 2026 · built with thecoursemaker.com