My wife read my last article – the one about building AI teams and not relying on off-the-shelf tools – and her response was immediate and completely fair: “That’s great. But how do I actually start?”
She is smart, busy, and has precisely zero interest in how AI works under the hood. She does not want to learn about models or prompts or agent architectures. She wants to know whether this thing can help her draft emails that do not take thirty minutes, stay on top of the logistics that pile up every week, and stop forgetting that the school needs a reply by Thursday. She wants hours of her life back.
If that sounds like you, this is for you. No code. No technical setup. Just a practical path from “I don’t know what to type” to having a personal AI that knows who you are.
The blank page problem
Here is the first thing nobody tells you: most people who open an AI tool never send a single message.
One analysis found that 60% of users who reach an AI chat screen just stare at the empty text box and leave.[1] Not because they are not smart enough. Because a blank text box with a blinking cursor is the worst possible starting point. You are supposed to type something, but what? You do not know what it can do. You do not know what format to use. You do not know if you will look daft.
The common advice is “just talk to it like a person.” That is partially right. You can talk to Claude the way you would talk to a capable friend who has offered to help, but here is the thing: even with a capable friend, the conversation goes differently depending on how much they know about you, and the situation you are asking them about right now.
Think of the difference between calling a random helpline and calling your accountant who has known you for ten years. The helpline asks you to explain everything from scratch. Your accountant already knows your situation, your preferences, how you like things explained, what you are likely to worry about. They give you better answers because they have context, not because they are smarter.
That is what you are going to build this week: the AI equivalent of an accountant who knows you. Not by learning any technology. By having a conversation.
Start here: let Claude interview you
This is the single most useful thing I can tell you, and it solves the blank page problem completely.
Go to claude.ai – the free version works for this. Instead of trying to figure out what to type, copy and paste this:
I want you to get to know me a bit better. Can we have a conversation where you ask me relevant questions that help me get the best out of AI? Ask me about my life, my work, what I struggle with, and what I’d like help with. Then give me a summary I can save and reuse.
That is it. Claude will ask you questions. You answer them. It is a conversation, not a writing exercise.
Claude will ask about your life situation, your work, your family setup, what you find hard, what you wish you had help with, and how you like to communicate. After a few rounds, it produces a neat summary of who you are and what you need – written in a format you can save and reuse.
Why does this work so well? Because it turns a writing task into a talking task. Most people find it much easier to answer questions than to write a description of themselves from a blank page. Claude asks, you answer, it synthesises. No template to fill in. No wondering what is relevant. Just a conversation.[2]
Save that summary somewhere. You will use it tomorrow.
What the difference looks like
Let me show you why this matters with an example.
Without any context, you ask Claude:
“Help me write an email to my boss asking for a day off next Friday.”
Claude produces:
“Subject: Request for Time Off – Friday, April 18th. Dear [Manager’s Name], I hope this email finds you well. I am writing to formally request a day of annual leave on Friday, April 18th, 2026. I have ensured that my current projects are on track…”
Corporate boilerplate. You would never send it.
Now imagine Claude already knows you – knows you write warm but concise emails, sign off with “Thanks, Sarah,” and hate over-explaining:
“Subject: Day off Friday 18th
Hi [name],
Could I take next Friday (18th) as annual leave? Nothing urgent in the diary and I’ll make sure everything’s covered before I go.
Thanks, Sarah”
Same request. Completely different output. The second one sounds like an actual person because Claude has context about how that person communicates.
The difference is not prompting skill. It is context. And context is what you built in that interview conversation.
Your first week
This is a five-day progression. Each day builds on the last, and each one should take less than fifteen minutes. The whole thing works on Claude’s free tier.[3]
Day 1: The interview
You did this already. You asked Claude to interview you, you had the conversation, and you saved the summary. If you have not done it yet, go back to the prompt above and start there.
Day 2: Make it remember
Go to claude.ai and create what Claude calls a Project – a workspace where you can store instructions that Claude reads at the start of every conversation. Call it whatever makes sense to you – “My Assistant” or “Life Admin” or just your name.
Paste in the summary Claude produced from yesterday’s interview as the Project instruction. Now every conversation you start inside this Project begins with Claude already knowing who you are. You never have to re-introduce yourself.
Try it. Ask the same question you asked yesterday, but from inside the Project. Notice how the answer is more specific, more like something you would find useful.
Day 3: Teach it a recurring task
Claude has a feature called Skills – saved sets of instructions for something you do regularly. Instead of explaining what you want every time, you teach Claude once and it remembers.
In a conversation inside your Project, try something like:
I want to create a Skill for drafting emails. Here’s how I write: keep it warm but short, no more than five or six sentences. Start with “Hi [name],” not “Dear.” Get to the point in the first sentence. If I’m asking for something, make the ask clear and specific. Sign off with “Thanks, Sarah.” Never use “I hope this email finds you well” or anything that sounds like a template. Save this as a Skill.
Next time you say “help me write an email to…” Claude remembers all of that without you having to explain your style again.
The concept is stable even if the exact menu options shift over time – Anthropic is actively developing how Skills work, so the interface might look slightly different when you try it. The idea is the same: teach it once, use it repeatedly.[4]
Days 4-5: Start saving the good stuff
When Claude produces something genuinely useful – a meal plan that works, an email you were proud of, meeting prep notes that helped – save it somewhere. Apple Notes, Google Docs, a folder on your desktop. Whatever you already use.
You are not building a system. You are just keeping the useful things where you can find them again. A meal plan template that worked perfectly. A packing list for a family trip you will reuse. A summary of something complex that finally made sense.
This is the seed of something powerful, but right now it is just a folder. That is fine.
When conversations get long
There is one honest limitation you should know about.
Think of a conversation with Claude like a phone call. Claude can hold a certain amount of the conversation in its head at once. In a short conversation, it remembers everything. But as the conversation gets longer – dozens of back-and-forth messages, long documents pasted in – the earliest parts start to fade. Claude will not tell you it has forgotten something. It will just give you an answer that misses a detail you mentioned earlier.
Here is the mental model that makes this simple: a conversation is a phone call. A Project is a notebook. The phone call ends and the details fade. The notebook is always there.
So the rule is:
- Start new conversations for new topics. Do not use one marathon conversation for everything.
- Put the important stuff in your Project instructions. Who you are, how you like things, your family setup, your preferences – anything Claude needs to know every time should live there, not in a conversation that will fade.
- Save outputs that matter. If Claude produces something you want to keep, copy it to your notes folder. Do not rely on scrolling back through old conversations.
- If a conversation gets complex, summarise and restart. Ask Claude to summarise everything you have agreed so far. Copy the summary, start a new conversation, paste it in, and continue. You have just refreshed the whole thing with all the important context intact.
This sounds like a workaround, and it is. But it is also how the technology works right now, and knowing about it means you will not be caught out by it.
What you can use this for
Once you have a Project that knows you, here are things that work well. I am giving you the full prompts so you can copy and adapt them – research on how people learn new tools shows that complete examples are far more useful than instructions to “figure it out."[5]
Meal planning:
Plan our dinners for this week. Four people, two adults and two kids (7 and 10). Budget is about 60 pounds for the week. Nothing over 40 minutes on a school night. We have chicken, rice, and tinned tomatoes already. Give me meals Monday to Friday and a shopping list.
Emails you are dreading:
I need to email my son’s teacher. He’s been struggling with homework recently and I want to flag it without sounding like I’m criticising the teaching. I want to come across as collaborative, not confrontational. His name is James and his teacher is Mrs Patel.
Meeting prep:
I have a meeting tomorrow with my manager about my workload. I’m feeling overwhelmed but I don’t want to seem like I can’t cope. Help me prepare: a clear picture of what’s on my plate, what should be deprioritised, and how to frame this as problem-solving rather than complaining.
Understanding something complex:
I want to understand what a mortgage fixed rate is versus a tracker rate. Explain it like I’m intelligent but have never thought about this before. Use real numbers. Keep it under 300 words.
Family logistics:
Help me plan half-term week. Two kids, ages 7 and 10. Budget about 200 pounds for activities (not food). We live in [city]. Mix of outdoor stuff, one big day out, one quiet day at home. It’s meant to rain on Wednesday.
Job applications (upload your CV to the Project first):
I’m applying for this role [paste job description]. Review my CV against their requirements. Where am I strong, where am I weak, and what should I emphasise in a cover letter? Don’t sugarcoat it.
Notice the pattern. Every good prompt has the same ingredients: who you are (which the Project already handles), what you need, and enough specific detail for Claude to give you something you can use immediately. Not technique. Just context.
Where it gets powerful: Skills and Projects together
The examples above use Projects – Claude knows who you are, so the answers are personal. But Skills take it further. Once you have taught Claude how you do something, Projects and Skills combine: Claude knows who you are and how you want this specific task done.
Here are a few Skills that work well in practice:
Weekly planning Skill:
Create a Skill for weekly planning. Every Sunday evening, I want to go through my week: what meetings I have, what deadlines are coming, what I need to prepare, and what I keep putting off. Ask me these questions one at a time, then produce a simple plan for the week with no more than three priorities per day. Keep it realistic – I always overestimate what I can do on Monday.
Meeting prep Skill:
Create a Skill for meeting prep. When I tell you I have a meeting coming up, ask me: who is it with, what is it about, and what do I want out of it. Then give me three things to say, two questions to ask, and one thing to watch out for. Keep it on one screen – I will read this on my phone five minutes before walking in.
Meal planning Skill (Projects + Skills combined):
Create a Skill for weekly meal planning. Plan 5 weeknight dinners for our family. Nothing over 40 minutes on a school night. My daughter does not eat mushrooms. Give me a shopping list at the end, organised by supermarket section.
Because the Project already knows your family setup, budget, and food preferences, the Skill only needs to handle the task-specific instructions. You say “do my meal plan” and Claude combines both – who you are and how you want this done.
A note on Memory
Claude also has a Memory feature – things it remembers across all your conversations, not just inside a Project. If you tell Claude “I prefer bullet points to long paragraphs” or “my manager’s name is David,” it can store that and apply it everywhere.
You do not need to do anything special to use it. Just talk to Claude naturally, and if something comes up that it thinks is worth remembering, it will ask. You can also tell it directly: “Remember that I work part-time on Fridays” or “Remember that my kids are at [school name].”
Memory, Projects, and Skills work as layers. Memory is what Claude knows about you everywhere. A Project is what Claude knows about you in a specific workspace. A Skill is how Claude does a specific task. Together, they mean you spend less time explaining and more time getting useful output.
You now have a personal AI
By the end of week one, you have:
- A Claude Project that knows who you are, how you communicate, and what you need help with
- At least one Skill for a task you do every week
- A small folder of outputs that worked
- Enough experience to know what prompts produce useful results and what falls flat
In week two, just use it. Every time you think “I should really sort out [thing],” open your Project and ask Claude. You will start to notice something shift: you stop thinking of it as a search engine or a chatbot and start thinking of it as an assistant that knows your situation. Every conversation starts from a foundation, not from zero.
I built a version of this for someone starting a new job – a system to help them capture what they were learning, prepare for meetings, and keep track of the people they were meeting. The core principle was the same: lightweight, forgiving, and designed to give before it asks. If you miss a day, nothing breaks. If you use it every day, it gets more useful every week.
The paid tier (Claude Pro, $20/month) gives you more usage and priority access. The core progression I have described here – Projects, Skills, the interview, saving outputs – works on the free tier. If you find yourself hitting usage limits and wanting more, Pro is there. But start free and see what it does for you first.
One last thing
The reason most people bounce off AI is not that they are not technical enough. It is that they are handed a blank page and told to be creative. That is a terrible starting point for anyone.
What I have described here is not a technical setup. It is a relationship. You tell Claude who you are. It remembers. You teach it how you like things done. It learns. Over time, the gap between what you ask for and what you get shrinks to almost nothing.
It is not going to be perfect every time. Sometimes it will produce something that misses the mark, and you will tell it what was wrong, and it will do better next time. That is how every good working relationship works.
Start with the interview. Give it fifteen minutes today. See what comes back.
What comes next
Once you are comfortable with Projects and Skills – once you have an AI that knows who you are and how you work – there is a next step worth knowing about. Claude has a collaborative mode called Co-work, where it works alongside you in real time rather than waiting for you to send a message.
I will not go into the details here. That is a piece for another day. But if you get through the five-day progression and find yourself thinking “this is useful, what else can it do?” – Co-work is the answer to that question.
Notes
[1] “The AI Blank Prompt Problem,” Nextbuild, 2026. A UX analysis of AI interfaces found that the majority of users who reach an AI chat screen leave without sending a message. The study identified four friction points: unclear purpose, unknown capabilities, format uncertainty, and fear of failure.
[2] The interview approach draws on prompt literacy research from the University of Michigan and MIT Sloan, both of which found that effective AI interaction uses the same skills as briefing a colleague: context, goals, and constraints. The key insight: hands-on practice with immediate feedback is more effective than reading instructions (arXiv, 2025).
[3] As of early 2026, Claude’s free tier includes access to Projects, Skills, and Memory. Feature availability may shift over time – check claude.ai for current details.
[4] Claude’s Skills feature is actively evolving. The concept – saved instructions for recurring tasks – is stable, but the specific interface for creating and managing Skills may look different from what is described here.
[5] John Sweller’s worked example effect (1988, replicated extensively) demonstrates that beginners learn significantly better from studying complete examples than from trying to solve problems themselves. The mechanism is cognitive load: working through examples preserves mental capacity for learning, while figuring things out from scratch consumes it all on the problem-solving process.