How to Start Working With Claude AI

Start working with Claude AI by choosing one small work task, explaining the outcome you need, and providing the information it must use. Review its first answer against that information before asking for a revision. You can learn this workflow without installing a coding tool or building a complicated prompt library.

A useful first session produces something you can judge, such as a client email or a meeting action list. fenbs can keep the approved decisions and next steps outside Claude when that small task becomes an ongoing project.

What should your first work task be?

1. Open Claude in the app or browser and start a conversation for one task.
2. Choose something you understand well enough to check, such as rewriting your own draft or organizing non-sensitive notes.
3. Define the audience, purpose, length, and facts that must stay unchanged.
4. Ask for a draft, review it, and request one focused revision before using the result.

For example, suppose you need to tell a client that a design review has moved. Give Claude the approved date, the reason you can share, and the action the client needs to take. Ask for a concise email rather than help with client communication, which leaves too much for the assistant to guess.

Avoid using your first experiment for a decision whose accuracy you cannot assess. A familiar task makes it easier to see where Claude adds value and where it fills a gap with a plausible invention. The introduction to what Claude AI is covers its place among AI assistants.

What context does Claude need from you?

Context is the relevant background an assistant needs to do the work. Include the source material, the intended reader, and the constraints. Explain unfamiliar names or shorthand. A few clear facts usually help more than a long account of everything that has happened in the project.

Draft an email to a client about moving our design review to Thursday afternoon. Use only these facts: the revised floor plan needs one more check; the project completion date is unchanged; the client should confirm availability. Keep it under 140 words. Use a calm, direct tone. Do not invent a time, apology reason, or promise. Ask if a missing fact prevents an accurate draft.

Anthropic’s prompting best practices↗ (opens in a new tab) recommend explicit requests, relevant context, and examples. You can apply that advice in ordinary language. If a particular format matters, show a short example and tell Claude which parts of it to imitate.

Use information you are allowed to share under your workplace’s rules. Replace customer details with placeholders when the task does not need them. Start with the minimum relevant document or excerpt, and add more only when a specific gap becomes clear.

How do you review an answer before using it?

Check the facts before the phrasing. Compare dates, names, numbers, commitments, and exclusions with your source material. Then read for the audience: would the client understand the request, and does the tone sound like your organization? A polished answer can still be wrong in a consequential way.

• Underline any statement that was not in your input. Verify it or remove it.
• Look for changed meaning, such as an unchanged completion date becoming a guaranteed delivery date.
• Open important source links rather than trusting a summary or citation label.
• Keep decisions requiring judgment with the person responsible for them.

Anthropic explains that Claude can give incorrect or misleading answers↗ (opens in a new tab). Asking it to check itself may help you spot issues, but it is not independent confirmation. Compare the result with the actual source or ask a qualified colleague when the subject needs one.

How should you ask for a better second draft?

Describe the specific defect and the change you want. Too wordy is less helpful than remove the opening paragraph and put the request to confirm availability in the first sentence. Say what should stay unchanged so a revision does not quietly introduce a new promise.

Try a sequence: first make it accurate, then make it clear, then adjust the tone. This gives you a way to judge each revision. If the same problem keeps returning, change the instruction or provide a better example instead of repeating make it better.

When you approve an important choice, save it as a decision rather than expecting every later draft to reconstruct it. fenbs keeps Decisions and rules with the project, so a connected assistant can read the approved direction before it starts a different piece of work.

When should you organize work into a project?

Create a Claude Project when several conversations need the same reference documents or instructions. Anthropic’s project setup guide↗ (opens in a new tab) explains how to add project knowledge and instructions. Put the actual working information there; do not assume that a project name alone briefs the assistant.

Keep stable background separate from temporary work. A tone guide can be a lasting instruction; a draft awaiting review is a current task. Give old material a clear status so an assistant does not mistake last month’s draft for the approved answer.

You do not need Claude Code, Cowork, or an outside connection for this first workflow. If you later connect tools and encounter a connection problem, the guide to Claude Desktop MCP troubleshooting addresses that separate issue. The Cursor starting guide is for work in that different app.

How to do this with fenbs

1. Create a free account and add one project from your real work.
2. Write the goal and add a task for the next deliverable, including what a good result must contain.
3. Record the choices a person has approved in Decisions and rules. Leave unresolved choices open.
4. Connect Claude using its integration guide and ask it to read the project context before drafting.
5. Review the work, update the task, and save Where we left off with the exact next action and anything waiting on you.

fenbs is free for one person, with no credit card. For a single disposable draft, a Claude conversation may be all you need. fenbs becomes useful when you need the work record to survive a new chat, another assistant, or a teammate taking over.

What should you save after the first session?

Save the useful prompt, the approved outcome, and one thing you learned about working with Claude. Keep a short statement of what remains, such as client email approved; waiting for confirmation before scheduling the review. That tells your next session where to begin.

Repeat the same loop on a slightly harder task. You are building judgment about instructions and review, rather than collecting elaborate prompts. The AI at work collection offers more workflows once you can reliably brief, check, and improve a small piece of work.