Now put Claude to work.
Claude 101 covered what Claude is and how to use it. This one answers the question people actually ask: "okay, but what do I do with it?" - with real, working examples you can copy.
These are the three questions I get asked most. The whole guide is organized around answering them - jump to whichever is yours.
Efficiency doesn't come from typing faster questions. It comes from a mindset change: stop thinking "chatbot," start thinking "delegate." The most valuable things Claude does aren't answers - they're tasks taken off your plate. Everything here is built to move you in that direction.
01 - The efficiency mindset
Ask → Delegate → Automate L1
There are three levels of payoff, and most people never leave the first. The gains compound as you climb, and you don't need to be technical to reach the top rung, just deliberate.
Ask
You have a question or a small task; Claude answers. Useful, but you're still doing all the driving. This is where "efficiency" is smallest - you saved a search, maybe a first draft.
Delegate
You hand over a whole task, with the context and tools to finish it: "read these five documents and give me the three risks," "pull this month's numbers and draft the summary." Claude does multi-step work; you review. This is where real time comes back.
Automate
You delegate a task once, then it runs itself on a schedule - no prompt required. The report writes itself every Friday; the notes get filed every morning. This is where the work disappears entirely.
The three recipes in this guide are one of each of the higher rungs. But they all rest on the same three building blocks - so let's demystify those first.
02 - The building blocks
How a real workflow is built L2L3
Every "wow, it just did that for me" workflow is the same three pieces clicked together. Once you see the pattern, you can build your own version of anything in this guide.
A Skill
Your written instructions for a task - "here's how I do this, and what good looks like." Teach it once; reuse it forever. (Claude 101, §06.)
Connectors
The bridges to your real tools and data - your email, calendar, chat, docs, databases. They let Claude read from and act in the apps you already use. (Claude 101, §08.)
A Scheduled Task
In Cowork, you can save a task and put it on a schedule. Claude then runs the whole thing on its own - unattended, on the cadence you set.
Put together, they read like a sentence: run this skill, using these connectors, on this schedule.
A Scheduled Task runs in the cloud, unattended - nobody is watching, and nobody can answer a mid-run question. That shapes what's a good fit: tasks with a clear, repeatable recipe and a definite "done." Anything needing your judgment mid-way is better run on demand (you kick it off, then review). Both are delegation; only one runs while you sleep.
Draft, don't send. For anything that leaves your hands - an email, a posted message - have the workflow produce a draft and stop, so you approve before it goes out. And check before writing. A task that creates records should confirm the item doesn't already exist first, or you'll wake up to duplicates. Both of these are choices you write into the skill.
03 - Recipe · automate
Automate a recurring chore L3
The pattern: a boring, repeatable job that moves information from one place to another on a fixed rhythm. Pull from source → transform → file it in the destination. Perfect for a Scheduled Task, because there's a clear recipe and a clear "done."
The recipe
- Name the trigger and the two ends
What kicks it off (a schedule - "every morning"), where the raw material lives (the source), and where the finished thing goes (the destination).
- Connect both ends
Add a connector for the source and one for the destination, so Claude can read one and write the other.
- Write the transform as a skill
Spell out exactly how raw input becomes finished output - the format, the length, the rules. Include the edge cases (see the guardrails below).
- Build in the safety checks
Tell it to skip anything already handled (no duplicates) and to fix the predictable mistakes (e.g. mis-heard names).
- Run it by hand, then schedule it
Trigger it manually a few times, tune the instructions until the output is right, then set it loose as a Cowork Scheduled Task.
The meeting-notes bot
- What it does
- Every morning it finds yesterday's call recordings, writes a clean, skimmable summary of each meeting, and files it in the right place in our team wiki - before I'm even online.
- Connects
- Zoom → sourceConfluence → destination
- How it runs
- As a Cowork Scheduled Task, fully unattended. I read the finished notes with my coffee; I never touch it.
- The clever bits
- It checks the wiki first so it never posts a duplicate; it corrects speaker names the transcription mis-hears; and it routes each meeting type (stand-up, all-hands, interview) to its correct folder and format.
- Why it works unattended
- The recipe is airtight and the "done" is obvious. There's no judgment call that needs me mid-run - exactly what makes a task safe to schedule.
The same shape fits a hundred chores: a weekly digest of a busy channel, an inbox that gets triaged and labeled each morning, form responses that get sorted into a tracker, new files that get renamed and filed. If it's "every [cadence], take X from here and put a tidied version there," it's this recipe.
04 - Recipe · delegate
A report pulled from many tools L2L3
The pattern: a status update, digest, or brief whose ingredients are scattered across several systems. The value isn't gathering - it's curating: pulling from everywhere, then ruthlessly cutting to what the reader cares about, in your voice.
The recipe
- List every place the news lives
Where does the raw material for this report actually come from? Usually more places than you'd guess - a tracker, a wiki, email, chat.
- Connect them all
Add a connector for each source so one pass can read across all of them.
- Write the skill: gather, then cut
The instructions do two jobs - (a) pull from each source over a set window, and (b) curate hard: merge duplicates, roll small items into themes, keep only what matters. Give it your format and an example in your voice.
- Make it draft-only
Because this goes to a real person, have it produce a draft and stop. You do the final read and send.
- Run it on demand
This one usually wants your judgment on the final cut, so trigger it yourself (e.g. every Friday) rather than fully scheduling it.
The weekly update to my manager
- What it does
- Once a week it assembles "here's what my team shipped," drafted in my voice as a short message - 3 to 5 tight bullets, each pairing what shipped with why it matters. I review and send.
- Connects
- JiraConfluenceOutlookSlack - four sources, read in one pass.
- How it runs
- On demand. It drafts the message and stops - it never sends on my behalf. That guardrail is written right into the skill.
- The clever bits
- It curates ruthlessly: the same win shows up in the tracker and a wiki page and a chat post - it merges those into one line. A dozen small fixes become a single bullet. It keeps only what a director cares about and drops the rest.
- Why not schedule it?
- The final trim needs my judgment (I know what a peer is already reporting), so this stays a delegate-and-review task, not a set-and-forget one.
Any recurring "pull-together" fits: a Monday priorities brief from your calendar and inbox, a client status from your project tools, a competitor-news digest, a monthly metrics recap. The skill is where you bottle your taste - what to include, what to cut, how it should sound.
05 - Recipe · standardize
A standardized deliverable, every time L2
The pattern: something you produce over and over that has to follow a specific structure - a ticket, a brief, a case note, a PRD, a job post. The win is consistency without re-explaining the format each time, and without the blank-page tax.
The recipe
- Capture the standard once
Write down the exact structure, rules, and an example of a great one. This becomes the skill - the single source of truth for "how we do these."
- Have it interview you
Instead of a giant form, the skill asks only for what's missing, a couple of questions at a time, and marks genuine gaps rather than inventing content.
- Choose the output
Hand back something ready to paste, or - if a connector is set up - have it create the item directly in the destination tool.
- Share the skill with your team
Now everyone's output matches, and nobody has to memorize the template. The standard lives in one place and travels.
The ticket writer
- What it does
- Turns a rough description of a bug or feature into a properly structured ticket that matches our team's exact house style - background, acceptance criteria, ownership, the lot.
- Connects
- Jira → optional It can post the ticket straight into our tracker, or just hand me clean formatted text to paste.
- How it runs
- On demand, as a skill I invoke. It reads our ticket guidelines first, then interviews me for only the details it doesn't already have.
- The clever bits
- Consistency, for free. Every ticket now follows the same structure with testable acceptance criteria - and I never re-explain the format. It fills genuine gaps with a "TBD" marker instead of making things up.
A skill captures your voice, not only a format. The same technique that standardizes a ticket can bottle the exact tone and rules of, say, your executive update - brief, non-technical, no em-dashes, lead with who it impacts most. See Claude 101, §07 - Make Claude sound like you.
No scheduling, no unattended run, minimal setup - just a written standard and an interview. If you're new to skills, build this kind first. Pick the document you produce most often, hand Claude a great example, and say "help me turn this into a skill." You'll have a reusable template-filler in twenty minutes.
06 - Quick wins
Routine tasks: a starter menu L1
You don't need a skill or a connector to start. Here's the everyday stuff Claude handles well, whatever your job. Find your area, try one this week, and notice how much of your day is actually these.
Email & communication
- Draft a reply from a few bullet points
- Summarize a long thread into "what's being asked of me"
- Soften or sharpen the tone of a tricky message
- Turn a rambling update into three clear sentences
Meetings
- Turn messy notes into decisions + action items
- Draft an agenda from the goal of the meeting
- Write the follow-up email before you forget
- Prep questions for a 1:1 or a review
Writing & documents
- First draft of anything - proposal, post, policy
- Tighten, restructure, or cut a doc to length
- Rewrite for a specific audience or reading level
- Generate an outline you fill in yourself
Data & spreadsheets
- Summarize what a sheet of numbers is telling you
- Write or explain a formula in plain English
- Clean up messy, inconsistent data
- Spot trends or outliers worth a closer look
Research & decisions
- Compare options in a clear pros/cons table
- "Poke holes in this plan" before you commit
- Brief yourself on an unfamiliar topic, fast
- Argue the opposite side to test your thinking
Planning & project work
- Break a big goal into concrete steps
- Draft a project plan or timeline
- Turn a request into a scoped task list
- Draft status updates from your raw notes
Notice which of these you do every week the same way. That one is your first candidate to promote from a one-off chat into a reusable skill. Routine → recipe is the whole move.
07 - Your daily driver
Managing your to-do list L1L2
A to-do list isn't really a memory problem - it's a prioritization and momentum problem. That's exactly where Claude helps. Three moves cover most of it.
1. The morning triage
Paste your running list and let Claude turn a wall of tasks into a plan: what to do first, what's actually urgent vs. just loud, and what to ignore today. It's the "where do I even start" fog, lifted in one message.
2. The weekly brain-dump
Once a week, dump everything on your mind - no structure. Ask Claude to sort it into Today / This week / Later / Delegate / Drop. Getting it out of your head and into buckets is most of the relief.
3. The "unstick me" move
For the task you keep sliding to tomorrow, the problem is usually that it's too big to start. Ask for just the first ten-minute action. Claude is very good at shrinking a dreaded task down to a first step you'll actually take.
Here's my to-do list. Act as my chief of staff. 1. Group these into Today, This week, and Later. 2. For Today, put them in the order I should tackle them and say why. 3. Flag anything that's really a 2-minute task I should just knock out. 4. Pick the one I'm most likely avoiding and give me its first 10-minute step. [paste your list]
Two ways to make this stick: keep a Project (Claude 101 §05) that holds your role, goals, and what "high impact" means for you - so every triage is done with your context, not generic advice. Or, if your tasks live in a task app with a connector, let Claude read and update the list directly instead of pasting. At that point your to-do list starts managing itself.
08 - Your turn
Build your own, in 5 steps L2L3
Every recipe here started as "ugh, I do this again every week." That complaint is the raw material. Here's how to turn it into a workflow - you don't write code, you write instructions.
- Notice the repeat
Watch for anything you do on a rhythm, or the same way each time. Recurring + predictable = automatable. Keep a note for a week; you'll be surprised how many there are.
- Describe how you do it
In plain language, like you're training a new hire: the steps, the rules, what "good" looks like, and the edge cases that trip people up. This description is the skill.
- Connect the tools it touches
Identify where the inputs come from and where the output goes, and turn on those connectors.
- Run it by hand and refine
Do it with Claude a few times. Each time the output isn't quite right, add a line to the instructions. This tuning is the actual work - and it's just editing text.
- Decide: on demand or on autopilot
If it needs your judgment mid-way, keep triggering it yourself. If it's a clean, unattended-safe recipe, put it on a schedule in Cowork and let it run.
Help me turn something I do repeatedly into a reusable Claude skill. Here's the task: [describe what you do] It uses these tools: [where the info comes from / goes] Good output looks like: [paste an example, or describe it] The rules and edge cases: [anything that must always be true] Ask me any clarifying questions first, then write the skill for me.
Notice that not one of those steps was "learn to code." Building a workflow is describing your own job clearly - a thing you're already an expert at. The technical-looking part (skills, connectors, schedules) is just where you put that description. Start with the standardized-deliverable recipe; it's the gentlest first build.
09 - Keep this bit
Recap
- Delegate, don't just ask. The efficiency is in handing over whole tasks, not typing better questions. Climb the ladder: Ask → Delegate → Automate.
- Every workflow is three blocks: a Skill (the recipe) + Connectors (the hands) + a Scheduled Task (the clock, in Cowork).
- Three recipes to steal: automate a recurring chore (source → transform → destination, on a schedule) · draft a report pulled from many tools (gather then curate, draft-only) · standardize a deliverable (capture the format once, let it interview you).
- Two safety habits: draft-don't-send for anything outbound, and check-before-you-write to avoid duplicates.
- For routine work, start today with the starter menu - no setup needed. Promote your most-repeated task into a skill.
- For your to-do list, use Claude to prioritize and to unstick - triage in the morning, brain-dump weekly, ask for the first 10-minute step.
You don't have to build a robot army on day one. Pick the one task that annoys you most this week and delegate it. Then the next. Efficiency with Claude isn't a big leap - it's a habit of noticing "I shouldn't be the one doing this by hand," and doing something about it.
Need the fundamentals?
← Back to Claude 101