Before you start: what this course is
This course is independent of Anthropic. Anthropic did not write, review or endorse it, and it reproduces none of Anthropic's own course or documentation text.
That matters because Claude changes often. Feature names, plan availability and limits can move between the day a lesson is written and the day you read it.
Do this throughout the course:
- Treat every feature name, plan and limit here as a snapshot dated October 2026.
- Before you rely on one, check Anthropic's current pages: the Help Center at support.claude.com, its release notes, and the pricing page.
- If this course and Anthropic's pages disagree, Anthropic's pages win.
One example of why: on 16 September 2026 Anthropic announced that Cowork, the separate mode for multi-step work, is being merged into ordinary Claude conversations. Its Help Center now says you ask for what you need and Claude decides whether that is a quick answer or a task. The change is rolling out gradually, to Pro and Max plans first, so your screen may still show a separate "Cowork" option. Level 3 covers names and plans in detail.
Two kinds of request
You already know chat: you ask, Claude answers, you read the answer and do something with it. Task work is different. You describe a result, and Claude carries out the steps to produce it.
Think of a capable colleague:
- Chat is asking them a question across the desk. You get an answer in seconds, and what happens next is up to you.
- A task is handing them a folder and saying "turn these into the monthly report". They go away, work through several steps, and return with a finished file.
| Quick answer (chat) | Delegated task | |
|---|---|---|
| What you ask for | Information, an explanation, a draft | A finished result |
| Steps involved | One | Several, done in sequence |
| What Claude works with | What you type or paste | Your files, connected apps or browser |
| What you get back | Text in the conversation | A finished file or a completed action |
| Your part afterwards | Do the work yourself | Review the work |
Even where Claude now chooses between the two for you, the distinction still matters, for three reasons:
- Cost. Anthropic's Help Center states that multi-step tasks use more of your usage allowance than a quick question.
- Wording. A task needs a clear result to aim at. A vague request gets a vague answer back.
- Review. A task changes or creates real things, so you must check what was done.
Sort a request: answer or task?
Before you type, ask these four questions.
- Is the result an answer or a deliverable? "What does this clause mean?" is an answer. "Summarise every contract in this folder into one table" is a deliverable.
- How many steps would it take you? One step points to chat. Open, read, compare, write and save points to a task.
- Does Claude need to reach something? If the work lives in files, an app or a website rather than in what you can paste, it is a task.
- Would you otherwise copy and paste repeatedly? Repetition across many items is a strong sign of a task.
If you answer "deliverable", "several", "yes" and "yes", treat it as a task.
Examples:
- "Give me a polite way to decline this meeting." → Quick answer.
- "What is the difference between a median and a mean?" → Quick answer.
- "Read the 40 survey responses in this folder, group them by theme and produce a one-page summary document." → Task.
- "Rename these 200 scanned invoices by supplier and date, then list any with a missing date." → Task.
Find three candidates in your own work
A good first candidate for delegation passes all five checks:
- Multi-step: it takes you several distinct actions.
- Clear end point: you can describe what the finished result looks like.
- Inputs you can point to: specific files, a folder or a named source.
- Checkable: you can tell whether the result is right without redoing the whole job.
- Recoverable: a mistake can be fixed. You keep the originals, and nothing is sent, paid or deleted.
Leave these out for now:
- Anything involving payments or passwords.
- Regulated or confidential data that your organisation's policy does not allow you to share.
- Actions that cannot be undone.
Later levels explain the risks behind each of these exclusions.
Now do this:
- Look back over your last two working weeks. List five jobs that took more than 20 minutes and involved several steps.
- Run each one through the five checks above.
- Cross out any that fail a check or appear on the leave-out list.
- Keep the best three. For each, write one line in this form: "From [inputs], produce [finished result]."
Example lines:
- "From the 12 monthly sales spreadsheets in the Sales folder, produce one summary workbook with a total per region."
- "From my meeting notes document, produce a formatted action list grouped by owner."
- "From the PDFs in the Suppliers folder, produce a table of supplier name, contract end date and notice period."
Keep your three lines. From Level 4 onwards you will turn them into full task briefs.
Key takeaways
- This course is independent of Anthropic. Check every feature, name and plan on Anthropic's current pages before relying on it.
- Chat returns an answer that you act on. A task returns a finished result that you review.
- A request is a task when it needs a deliverable, several steps, and access to files, apps or the browser.
- Tasks use more of your usage allowance than quick questions.
- Good first candidates are multi-step, clearly finished, checkable and recoverable, with inputs you can point to.
- Leave out payments, credentials, restricted data and anything irreversible.