How to put Claude to work — what to hand it, how to ask, and how to get hours of your week back. No jargon, no hype, just what works.
A few numbers from real research — not vendor promises — on what's actually happening:
Two more things the research shows. First, this isn't a coding tool anymore — around 90% of what people delegate to Claude's agent mode is ordinary office work: documents, reports, email, analysis, planning. Second, there's a learning curve, and it compounds: people who've used AI for six months get measurably better results than people who started last week. The advantage isn't access — everyone has access. The advantage is reps. This guide is designed to get you your first reps this week.
Most people fail with AI not because of bad prompts, but because of the wrong picture in their head. Fix the picture first.
It's a brilliant new hire with zero context. Anthropic's own advice: treat Claude like a talented employee on day one. It doesn't know your business, your customers, or your standards until you tell it. The more context you give, the better it performs — same as any new hire.
Its skills are jagged, not uniform. AI is startlingly good at some hard things and surprisingly bad at some easy ones — researchers call this the "jagged frontier." You can't know which side a task falls on until you try it. So try it on everything low-stakes, and keep score.
You are the human in the loop. AI drafts, you decide. It can sound confident while being wrong, so anything with your name on it gets your eyes before it goes out. You're not being replaced as the judge — you're being freed up to only be the judge.
Delegate outcomes, not keystrokes. Don't micromanage sentence by sentence. Describe the finished thing you want — "a one-page proposal my client can say yes to" — and let it work. But only delegate outcomes you're able to check.
Nine categories of everyday work. Each one opens with real examples and a copy-paste prompt to try — tap a category, copy a prompt, paste it into Claude, fill in your details.
Here's an email I received: [paste it]. Draft a reply that politely declines the discount request but keeps the relationship warm. Under 120 words, friendly-professional. Give me two versions: one warmer, one firmer.
You're my proposal assistant. Using these call notes [paste] and my service list [paste], draft a one-page proposal with: the client's problem in their own words, my recommended solution, timeline, a price section left blank for me, and clear next steps.
Here's a copy of my sales data [attach a copy, not the original — and remove customer names if you don't need them]. Tell me: the monthly trend, my top 5 customers by revenue, my weakest month, and anything unusual I should look at. Explain it like I'm not a data person, then give me one chart worth sharing.
Here's a meeting transcript [paste]. Give me: a 5-bullet summary, decisions made, action items with owners and deadlines, and a draft recap email to attendees that's short enough people will actually read it.
Compare [Tool A], [Tool B], and [Tool C] for a 5-person [industry] business: monthly cost, key features, contract lock-in, and best fit for us and why. Table format. Flag anything you're not certain about instead of guessing.
Draft a reply to this 2-star Google review [paste]: apologetic but not groveling, offers to make it right, under 80 words, sounds like a real person and not a corporation.
You're the social media manager for [business] in [town]. From this brain-dump [paste], write 12 posts for the next month: 4 educational, 4 promotional, 4 behind-the-scenes. Match the voice of this past post: [paste one you liked].
Write a job post for a part-time [role] at [business]. Requirements: [list]. Tone: warm, zero corporate jargon. Then give me 5 interview questions that reveal reliability and 3 red flags to listen for.
Here's everything on my plate this week [dump the list, messy is fine]. Sort it into: do today, schedule, delegate, drop. Then draft a realistic Mon–Fri plan with the top 3 priorities each day, and tell me which one thing matters most.
Anthropic's golden rule: if a colleague reading your request would be confused, Claude will be too.
Give the audience, the purpose, the format, and the length. Vague in, generic out.
Claude generalizes from reasons. "This is for a nervous first-time client, so keep it reassuring" beats a list of ten formatting demands.
Paste a past email, proposal, or post you were proud of and say "match this style." One good example outperforms three paragraphs of description.
The first draft is the starting point. "Shorter." "More casual." "Give me 3 options." "You missed X." Iterating two or three times is how it's supposed to work — correction is the core skill.
"Write in flowing paragraphs" works better than "don't use bullet points." Positive instructions aim; negative ones just fence.
Add "if you're not sure, say so instead of guessing" to anything factual. It measurably reduces made-up answers — and tells you where to double-check.
One more that beats all six on big tasks: ask Claude to ask you questions first. "Before you start, ask me anything you need to know to do this well." It surfaces the context you forgot you had.
Most people stay on Level 1 forever. Each step up saves more time than the last — climb one level at a time, as your trust is earned.
Ask questions, draft text, think out loud. Where everyone starts — and where copy-paste answers live.
When a prompt works, save it. A note file of your five best prompts is a real business asset — you're building your own playbook.
A Project is a workspace where Claude permanently knows your business: upload your services, prices, tone examples, and standing instructions once. Every chat inside starts already briefed. This is the single biggest upgrade most people never make.
Claude creates actual working files: Excel spreadsheets with live formulas, Word docs, PowerPoint decks, PDFs. Stop copy-pasting into documents — ask for the document.
Claude Cowork (paid plans) works like a colleague, not a chatbot: it plans multi-step work, works with entire folders of files, and keeps going after you close the laptop. Plus scheduled tasks — a Monday-morning report or daily brief that runs itself. This is where "using AI" becomes "having staff."
You're an assistant for [business name], a [what you do] in [where]. Our customers: [who]. Our tone: [3 words — e.g. warm, plain-spoken, confident]. Our services and prices: [list]. When drafting anything, match the tone, keep it concise, and ask me clarifying questions before big tasks. If you're unsure of a fact, say so.
AI's most dangerous errors aren't the dumb ones — they're the polished, confident, plausible ones. A few habits neutralize the risk.
Don't "adopt AI." Just do one 10-minute experiment a day for seven days. By Sunday you'll know exactly where it fits in your work.
Paste it in, ask for a reply in your tone. Edit, send. Notice the time.
Any notes — a call, an idea, a plan. "Structure this into a one-pager."
Duplicate the file first (copy-first rule), then ask: "What trends do you see? What should worry me? What's my best month and why?"
Ask for decisions, action items, and the recap message.
A tense reply, a pitch, a difficult message. Ask for three versions, pick the best parts.
Use the Level 3 starter prompt above. Ten minutes now, briefed-in-advance forever.
Whichever experiment saved the most time — make it a standing habit. Then add one new use case a month.
That's the whole method: start small, start now, add one use case a month. The research says the learning curve pays off within weeks — and the people who feel most confident about AI's effect on their work are the ones who delegate the most, not the least.