---
id: work-playbooks
lang: en
title: "Putting it together: work playbooks"
seoTitle: "AI work playbooks: chain tasks end to end"
description: "Turn single AI tasks into repeatable work playbooks: research to report, inbox to actions, data to a deck, notes to follow-through, and how to build your own."
part: advanced
partTitle: "Advanced and specialised"
partOrder: 5
order: 4
level: Advanced
durationMinutes: 13
outcome: "Chain everything you have learned into repeatable, end-to-end playbooks, steal four ready-made ones, and build your own that hold up over time."
summary: "A playbook chains single AI tasks into a repeatable, end-to-end routine, research to report or inbox to actions, so good results happen the same way every time."
draft: false
updated: "2026-07-26"
keywords:
- AI workflow
- AI playbook
- end to end AI
- AI at work examples
- combine AI tasks
- Claude Cowork workflow
- automate work with AI
furtherReading:
- title: "Claude use cases (Anthropic)"
url: "https://claude.com/resources/use-cases"
note: "Real end-to-end examples across roles and industries, filterable by feature."
- title: "Get started with Claude Cowork (Help)"
url: "https://support.claude.com/en/articles/13345190-get-started-with-claude-cowork"
note: "The mode where multi-step playbooks actually run."
- title: "Delegating and scheduling tasks in Claude Cowork (tutorial)"
url: "https://claude.com/resources/tutorials/delegating-and-scheduling-tasks-in-claude-cowork"
note: "A worked walk-through of handing a whole job to Claude."
- title: "Schedule recurring tasks in Claude Cowork (Help)"
url: "https://support.claude.com/en/articles/13854387-schedule-recurring-tasks-in-claude-cowork"
note: "Put a proven playbook on a clock so it runs itself."
quiz:
intro: "Five questions to close out the whole course. Ace them and you have officially finished Rollo Academy."
passScore: 5
questions:
- q: "What is a work playbook, in one line?"
options:
- "A single clever prompt that does everything at once"
- "A repeatable recipe that chains several steps into one finished outcome"
- "A setting you switch on to make Claude faster"
answer: 1
explain: "A playbook is a recipe: a fixed set of steps that turns a known input into a known output, the same way every time. The value is that it repeats."
- q: "In the toolkit you have built, what does a project give a playbook?"
options:
- "The method, or how a task is done"
- "The reach into outside tools"
- "The context: the files and background Claude carries into every step"
answer: 2
explain: "Projects hold context, skills hold method, connectors give reach, and a scheduled task adds the clock. A playbook snaps these together."
- q: "Before you trust a playbook to run start to finish, the smart move is to..."
options:
- "Run it once by hand and check the output of each step"
- "Write it as one giant paragraph and hope"
- "Schedule it daily immediately"
answer: 0
explain: "Test it once, watch each hand-off, fix the weak step. A playbook only earns automation after it has proven it works by hand."
- q: "Which step in a playbook most deserves a human checkpoint?"
options:
- "Gathering and summarising information"
- "Any step that sends, deletes, spends, or ships to other people"
- "Formatting the output"
answer: 1
explain: "Reading and shaping are safe to automate. An action that leaves your desk and cannot be undone deserves your eyes before it goes."
- q: "What is the natural last rung of the maturity ladder for a playbook that works?"
options:
- "Delete it so nobody else can use it"
- "Keep retyping it from scratch each time"
- "Package it as a skill or plugin so your whole team runs it the same way"
answer: 2
explain: "Do it once, template it, schedule it, then share it. A proven playbook packaged as a skill turns your good workflow into everyone's."
pass:
badge: "Course complete"
title: "You did it. That's the whole course."
body: "Full marks on the final lesson. You can now chain everything from Parts 1 to 5 into playbooks that do real work, and build your own that hold up over time. Take a bow."
retry: "Almost there, on the very last lesson. Check the marked answers and finish strong."
---
> **Level:** Advanced · **Reading time:** ~13 min
>
> **What you'll be able to do:** chain everything you have learned into repeatable, end-to-end playbooks, steal four ready-made ones, and build your own that hold up over time.
This is the last lesson. A victory lap.
Every lesson before this one taught a single move. How to write a prompt, how to feed Claude context, how to reach for the right model, how to build with [artifacts](/glossary#artifact) and [Claude Design](/glossary#claude-design), how to put a task on a schedule. Useful moves, all of them. But real work is almost never a single move. Real work is a *sequence*: gather the thing, shape the thing, turn it into the thing your boss actually asked for.
**This lesson is where the single moves become plays.** A play you can run again next week without thinking. We call it a **playbook**, and by the end of this you will have four you can steal and the method to write your own.
## What a playbook actually is
**A playbook is a repeatable recipe: a fixed set of steps that turns a known input into a known output, the same way every time.**
That is the whole idea, and the important word is *repeatable*. A one-off question is not a playbook. "Summarise this contract" is a task. "Every time a new contract lands, pull out the parties, the term, the payment terms, and the get-out clauses, then flag anything unusual against our standard" is a playbook. The first you type once. The second you run forever.
The shift in your head is from *asking Claude a question* to *running Claude through a job*. Once you see your work as a set of recipes rather than a pile of one-off requests, the whole thing gets faster, because you stop reinventing the steps every single time.
## Your toolkit, finally assembled
Here is the good news you may not have noticed: **you already own every piece of a playbook.** The five parts of this course were not a random tour. They were a kit, handed to you one tool at a time. A playbook is just that kit snapped together.
- **[Projects](/glossary#project) hold the context.** The files, the background, the standing instructions Claude carries into every step, so you are not re-explaining your world each time.
- **[Skills](/glossary#skill) hold the method.** A packaged way of doing a specific task well, the same way on every run.
- **[Connectors](/glossary#connector) give the reach.** They let a step touch your real tools: email, calendar, chat, drive, your systems.
- **Artifacts and [Claude Design](/glossary#claude-design) make the thing.** A doc, a table, a deck, a page, a prototype, produced rather than described.
- **[Scheduled tasks](/glossary#scheduled-task) add the clock.** When a playbook is good, a schedule runs it for you while you sleep.
And **[Cowork](/glossary#cowork) is the room where they all work together**, the mode where Claude runs multi-step jobs across your files and tools rather than answering one message at a time. A playbook is context plus method plus reach plus a maker plus, when you want it, a clock.
## Four playbooks you can steal
Enough theory. Here are four complete playbooks, each a real job that eats professionals' time. Read them as recipes. The prompt blocks are starting points: **run each one once by hand, watch what comes back, then sharpen it for your world.**
### Playbook 1: Research to report
*The job:* you need a briefing on a topic, a competitor, or a market, pulled from many sources into one clean read. *The old way:* a dozen browser tabs and an afternoon gone.
The chain is **gather, sort, draft, check**. Web search does the gathering, Claude does the sorting and drafting, and you own the final check because facts matter.
```
You are helping me prepare a briefing for a decision.
Topic: [the competitor / market / question].
Step 1: Search the web for current, credible sources from the last 12 months. Prefer official pages, filings, and reputable news over blogs.
Step 2: Pull out the facts that matter for my decision, and note where each one came from.
Step 3: Draft a one-page briefing: a three-line summary at the top, then What we know, What is changing, and What it means for us.
Step 4: List anything you could not verify, so I know what to double-check.
Keep it under 500 words. Plain language, no filler.
```
**The habit that makes it sing:** always keep step 4. The list of what Claude could not confirm is the difference between a briefing you trust and one that quietly invents a number. You own the facts; let Claude do the legwork.
### Playbook 2: Inbox to actions
*The job:* turn a morning's worth of unread email into a short list of what actually needs you. *The old way:* forty minutes of scrolling and a nagging fear you missed something.
This one leans on a [connector](/glossary#connector) to reach your mail, and it is a natural for a [scheduled task](/glossary#scheduled-task) once it works.
```
Read my unread email from the last 24 hours.
Sort it into three buckets: Needs a reply from me, Needs a decision, and Just so you know.
For each item in the first two buckets, give me the sender, a one-line summary, and the single action you suggest.
Ignore newsletters, receipts, and automated notifications.
If nothing needs me, say "Inbox is clear" and stop.
Keep the whole thing under 200 words.
```
Notice this playbook only *reads and sorts*. It never sends a reply on its own. That is deliberate, and we will come back to why in the pitfalls. Get this running well by hand, then schedule it for 7am and start your day at the finish line instead of the start.
### Playbook 3: Data to a slide deck
*The job:* a spreadsheet of numbers needs to become a deck someone will actually present. *The old way:* copy, paste, fight the chart, repeat.
The chain runs **read the data, find the story, build the slides**. This is where [Claude Design](/glossary#claude-design) or an [artifact](/glossary#artifact) turns analysis into something you can stand up and show.
```
Here is a spreadsheet of [what it contains].
Step 1: Read it and tell me the three or four things that actually matter, the story a busy executive needs.
Step 2: For each one, suggest the simplest chart that makes the point.
Step 3: Build a short slide deck: a title slide, one slide per point with its chart and a one-line takeaway, and a closing "so what do we do" slide.
Keep the wording tight. One idea per slide.
Then wait so I can tell you what to change before you polish it.
```
That last line matters: **build in a pause.** You want to steer the story before Claude spends effort polishing the wrong three points. Analysis first, agreement second, polish last.
### Playbook 4: Meeting notes to follow-through
*The job:* a messy page of meeting notes needs to become clear actions that actually happen. *The old way:* good intentions, and a document nobody opens again.
The chain is **read, extract, assign, route**, and it shows how a playbook can end by *doing* something, carefully.
```
Here are my raw notes from today's meeting.
Step 1: Pull out every decision we made and every action item.
Step 2: For each action, give me the owner (as named in the notes), what exactly is due, and by when.
Step 3: Write a short, friendly recap I could send to the group: decisions first, then the action table.
Step 4: Draft, but do not send, a one-line message to each owner about their item.
Show me everything before anything leaves my hands.
```
Step 4 is powerful and a little dangerous, which is exactly why it says **draft, do not send**, and **show me first**. A playbook that touches other people should always stop for your nod before it acts.
Look across the four and you see the same shape each time: a raw input on the left, a couple of gather-and-shape steps in the middle, one finished output on the right. Different jobs, one skeleton. Learn the skeleton and you can build a playbook for almost anything.
## How to build your own playbook
The four above are training wheels. The real prize is spotting the playbooks hiding in *your* week. Here is the method.
**First, spot the candidate.** The best playbooks share three marks: the job **repeats**, it has a **clear input**, and it has a **clear output**. If you have done something more than twice and could describe the steps to a new hire, it is a playbook waiting to be written. The dull, recurring chores are the gold here, precisely because you would never miss doing them.
**Second, name the steps out loud.** Before you touch Claude, say the recipe in plain words: "Take *this*, do *this*, then *this*, and give me *that*." If you cannot say the steps, Claude cannot run them. Writing the playbook is mostly writing the steps down clearly.
**Third, test it once, by hand, and watch every hand-off.** Run the whole thing as one [Cowork](/glossary#cowork) task and read the output of each step, not just the end. Where did it drift? Which instruction was vague? The seams between steps are where playbooks break, so watch the hand-offs closely and tighten the weak one.
**Fourth, add checkpoints where they belong.** Decide which steps Claude can run freely (reading, sorting, drafting) and which need your eyes before it continues (anything that sends, spends, or ships). Build the pause into the prompt, as you saw in playbooks 3 and 4.
**The rule:** a playbook is only as good as its weakest step and its riskiest one. Fix the weak step so the output is right; guard the risky step so a mistake cannot escape.
## The maturity ladder
A playbook is not something you either have or you do not. It grows up in four rungs, and most of your wins come from climbing just one or two.
**Rung one: do it once, by hand.** Run the job in a single Cowork task. This alone often saves an hour. Many playbooks live happily here forever, and that is fine.
**Rung two: template it.** When a job earns its keep, save the prompt somewhere you can grab it, or drop it into a [project](/glossary#project) with the context it needs. Now you are not rewriting the recipe each time, just pasting it and pressing go.
**Rung three: schedule it.** For anything that repeats on a clock, hand it to a [scheduled task](/glossary#scheduled-task), the whole point of lesson 5.3. The morning briefing, the Monday report, the weekly watch: describe once, and come back to finished work.
**Rung four: share it.** The highest rung. Package a proven playbook as a [skill](/glossary#skill), or bundle several into a plugin, so your whole team runs it the same way. This is how one person's clever workflow quietly becomes the way the whole company works. Your good habit, multiplied.
You do not have to climb to the top. Know that the ladder exists, and take each job as high as it deserves.
## Pitfalls to avoid
A little hard-won caution, so your playbooks help rather than haunt you.
- **Do not over-chain.** A ten-step playbook where any step can wobble is fragile. Prefer a few solid steps to a long, clever contraption. If a chain keeps breaking, split it into two shorter playbooks with a human handshake in the middle.
- **Do not skip the checkpoints.** The whole point of the pauses in playbooks 3 and 4 is that some steps cannot be undone. An email sent is sent. Keep a human on anything that leaves your desk.
- **Mind the garbage-in rule.** A playbook does not fix a bad input, it scales it. Point it at the wrong folder or a vague brief and it will produce the wrong thing faster than ever. Get the input right first.
- **Automate the chore, never the judgement.** This is the thread running through the whole course. Claude is brilliant at the gathering, sorting, and drafting that eats your hours. What the output *means*, and what you do about it, stays yours. Let the playbook lay the table. You still decide what to cook.
## You made it: the whole course, in one breath
Take a second. You started this course perhaps unsure what a [large language model](/glossary#large-language-model) even was. Look at the ground you have covered.
You learned **what AI is and how it works**, and how to keep your data safe while using it. You **got started with Claude**: the plans, the interface, the models, artifacts, and how context changes everything. You learned to **prompt** properly, and to iterate when the first answer misses. You learned to **work smart**: managing the context window, and the tools that give Claude reach. And in this final part you went **advanced**: Claude Code, Claude Design, scheduled tasks, and now playbooks that tie it all together.
**You are no longer someone who uses AI to answer questions. You are someone who puts AI to work.** That is the whole difference, and you built it one lesson at a time.
Rollo AI AcademyWant this run for your whole team?The Rollo AI Academy turns this course into hands-on training: workshops, role-based programs, and coaching built around the real work your team does every day.See the training →
## Try it
Twenty minutes well spent, and the perfect way to end the course: build one real playbook from your own work.
1. **Pick a chore you do often.** Something that repeats, with a clear input and a clear output. A weekly update, a recurring summary, a report you rebuild each month.
2. **Say the steps out loud** and write them down: take *this*, do *this*, give me *that*.
3. **Run it once** as a Cowork task, using one of the four playbooks above as a template. Read the output of each step, not just the end.
4. **Fix the weakest step** and mark where you want a checkpoint before Claude acts.
5. **Decide how high to climb the ladder.** Keep it as a saved prompt, or, if it repeats on a clock, put it on a schedule.
You now have a playbook of your own, and the method to write the next ten. That is the course. Go and put AI to work.
*(General exercise. Role and industry versions come once accounts are in.)*
## Key terms
[Playbook](/glossary#playbook), [Cowork](/glossary#cowork), [Project](/glossary#project), [Skill](/glossary#skill), [Connector](/glossary#connector), [Scheduled task](/glossary#scheduled-task)