I Wasted 3 Months on Claude Before I Figured Out I Was Doing It Wrong

AI TOOLS  |  REAL TALK

I wasted three months on Claude before I figured out I was doing it completely wrong

Here's the ten-step fix that took me one week and now runs my mornings on autopilot.

Based on Ruben Hassid's guide at how-to-ai.guide

Okay so I have to be real with you. For months I thought I was some kind of power user. I had Claude open basically all day. I was prompting it constantly. Asking it to write things, summarize things, brainstorm things.

Then I saw Ruben Hassid's infographic and I realized I had been using the digital equivalent of a sports car to go to the corner shop. I was not even close to what this thing could actually do.

The version of Claude I was using answered questions. Fine. Useful sometimes. But there is a whole other version that wakes up before you do, checks your Slack and your Notion, writes you a briefing doc, and has it sitting in a folder by the time you open your laptop. I did not know that version existed.

After one week of following this guide, I switched over completely. This is what that week looks like.

 

— Day One —

Step 1

Download the actual app. The website is the tourist version.

First thing Hassid says. Do not use the browser. Go get the desktop app from claude.com/download.

I pushed back on this mentally. The website felt fine. I had tabs open, I had it bookmarked, what was the difference.

The difference is Cowork. That is Claude's deep-work mode where all the serious stuff happens, and it flat out does not exist in the browser version. Mac or Windows only. No mobile, no web workaround. You either have the app or you skip this entirely.

Pro plan is twenty bucks a month. Seventeen if you pay for the year. I had been on free this whole time wondering why things felt limited. There is your answer.

Install the app, open it, find Cowork in the sidebar. That's your base from now on.

Step 2

You have been in the wrong mode. Almost everyone is.

Claude has four modes. Most people know exactly one of them. It is Claude Chat, the default, the one that looks like every other chatbot you have ever used.

Here is the brutal truth about Claude Chat. It does not remember you. Every session starts from zero. It has no connection to your files, your tools, your projects. You are a stranger every single time you open it.

Projects is the step up from that. It holds context between sessions. Good for recurring work where you need Claude to remember what the situation is without you re-explaining it every week.

Cowork is the one this whole guide is about. It connects directly to folders on your computer. It reads your actual files. It writes output directly to a folder you control. It can handle long, complex documents without losing the thread halfway through. This is the mode.

Code is for developers. If you write software, great. If not, move on.

Switch Cowork to your default right now. Notice how different day one feels compared to what you were doing before.

 

— Day Two —

Step 3

Four folders. That's it. Don't make it complicated.

This is the part where I want you to resist the urge to get creative. The folder structure is not a suggestion. It is the thing that makes everything else work. Mess it up and you are going to spend the rest of the week confused.

Make a folder somewhere on your computer. Call it Claude-Cowork. Open it and make four folders inside it. Not six. Not four plus a random extras folder. Four.

Folder one is About Me. Read-only. This is where you tell Claude who you are before it does anything. Your actual job, what projects are live right now, what matters to you, what you are trying to get done this month. Claude reads this at the start of every Cowork session. It is the reason you stop having to introduce yourself every single day.

Folder two is Projects. One subfolder per live project. Brief inside, working files inside, references inside. Nothing shared between projects.

Folder three is Templates. Save the structure of your best past work here. Not the content. The skeleton. The brief format that worked. The proposal layout that converts. Claude uses these so it is not guessing at format from scratch every time.

Folder four is Outputs. This is the only folder Claude writes to. Full stop. Everything else is read-only. Claude creates, Claude puts it here. You review from here.

When you feel the urge to make a fifth folder, ask yourself which of the four it really belongs in. It almost always fits somewhere.

Step 4

Two files. This is where Claude stops sounding like a robot.

Most people either skip this or write something like 'I am a marketing manager who likes clear writing.' That is not enough. Claude will use whatever you give it. Give it nothing useful and you get nothing useful back.

File one is about.me.md. It goes in your About Me folder. Write it like you are catching up a smart friend who just joined your team. Tell them what your job actually involves day to day. What projects are eating your time right now. What kind of person you are when it comes to how you like work presented. Real stuff. Honest stuff. Not a LinkedIn bio.

File two is anti-ai-style.md. This one changed my output more than anything else. Write down every phrase that makes your skin crawl when you see it in AI writing. 'It is worth noting.' 'Delve into.' 'Game-changer.' 'Robust solution.' 'Leverage.' 'Seamlessly.' 'At the end of the day.' Then write what you actually want instead. Short words. Direct sentences. Something that sounds like a real person wrote it on a Tuesday afternoon.

Claude reads both of these before it does anything. The output shifts fast. Within a session or two you will notice it.

Update these files every few weeks. Your projects change, your priorities shift, and Claude should shift with them.

The closer Claude knows you, the less time you spend editing what it gives back.

 

— Day Three —

Step 5

One template file saves you from rewriting the same prompt forever.

Something annoying happens when you work with Claude regularly. You figure out a really good prompt. You get great output. You close the session. Next time you need something similar, you try to recreate that prompt from memory and it is never quite right.

Stop doing that. Make a file called prompt-template.md and put it in your Templates folder. Fill it with your standard instructions. Your format preferences. Your tone notes. The path to your Outputs folder. Whatever you find yourself repeating session after session, it goes in here once.

Then set up a keyboard shortcut. On Mac you do it in System Settings under Keyboard, Text Replacements. Type /prompt and it expands into your full template. On Windows, AutoHotkey does the same thing. Three keystrokes and your entire setup is loaded in.

You will not remember what it felt like to start from scratch after a week of this.

Spend half an hour building this properly once. It pays that back in the first two days of use.

Step 6

Stop explaining everything yourself. Let Claude figure out what it needs.

The habit most people have with AI is to write a long careful prompt that tries to cover every angle, then hold their breath and see if it lands. When it does not quite work they write another long prompt. And another.

There is a smarter loop. In Cowork, you can ask Claude to ask you questions instead of the other way around. It generates a short form based on the task. You pick from options. You drag to rank what matters most. Claude builds its plan from your answers.

What I found is that this works especially well when I am not totally sure what I want. Sometimes the questions themselves are what help me figure it out. Better than staring at a blank prompt box for ten minutes.

The whole thing takes under a minute. You click, Claude plans, you approve, it executes. Faster than a kick-off meeting. Genuinely.

Next time you open a new task and feel stuck on how to explain it, ask Claude to interview you instead. See what comes out.

 

— Day Four —

Step 7

One plugin. Pick it because you actually need it, not because it looks cool.

The plugin marketplace is dangerous if you are the kind of person who installs things and never uses them. I am that person. I know myself.

Hassid says pick one plugin that solves a specific problem in your actual work. Not the most interesting one. The most relevant one. Install it, use it every day for a week before you even look at anything else.

If you are in marketing, something like marketingdraft.ai connects to Claude and gives you properly structured campaign output that would normally take a dozen back-and-forths to get. If your work is data-heavy, find a plugin that handles CSV files properly. Claude can crunch data natively but a dedicated plugin gives you cleaner, more reliable results. If you review contracts, there are plugins built specifically for that work. Clause checking, flagging weird terms, plain-English summaries of dense paragraphs.

Find your use case. One plugin. Use it until it becomes muscle memory.

Ask yourself what task you repeat most often and costs you the most time. Find the plugin that handles that specific thing.

Step 8

Your tools are sitting right there. Connect them already.

This is the step where Claude stops being a smart box you type into and starts actually knowing what is happening in your work.

Settings, Connectors, Browse, Add. The list of apps is longer than I expected. Google Drive, Gmail, Slack, Notion. Connect them and Claude can reach into all of it mid-task without you copying and pasting anything.

Before I connected Slack, I had to dig up old threads myself, paste the relevant bits in, and explain the context. Now I say 'check Slack for what we decided about the onboarding flow' and Claude goes and looks. Same with Notion docs. Same with Google Drive files.

The output quality goes up noticeably because Claude is working from your actual situation instead of working from what you remembered to tell it. Those are different things and the difference shows up in what you get.

Start with whatever tool holds your most important information. Get one connection working really well before you add the next one.

 

— Day Five —

Step 9

If you work with other people, build something you can all share.

Solo setup is useful. Shared setup is a different thing entirely.

When everyone on a team prompts Claude individually, the output sounds like it came from different people. Because it did. Different context, different assumptions, different prompt styles. Inconsistency every time.

A shared team project in Cowork fixes that. You put the brand voice doc in there. The style guide. The strategy brief. The current campaign context. Now anyone who uses Claude from that shared project gets output that draws from the same source. It sounds consistent because it is pulling from consistent material.

The main thing to get right is keeping the folder honest. Old documents that no longer reflect how things actually work should come out. Someone needs to own that maintenance. The folder should contain what is true right now.

Start with three or four documents that everyone on the team agrees actually represent how you work. Build from there slowly.

Step 10

Schedule one automated task. This is what the whole week was building toward.

Everything else was setup. This is where the setup starts giving back.

Claude has a /schedule plugin. You tell it to build a morning briefing document before you start work each day. It pulls from your connected tools, your active project folders, whatever landed in your Outputs folder overnight. It writes a short summary of what matters today and puts it in your Outputs folder.

You do not need to be there when it runs. You wake up. You open the Outputs folder. The briefing is already there. You read it in five minutes and you know exactly where the day starts. The first decision of the morning is already made.

That is what the whole thing is pointing at. Not Claude answering your questions. Claude doing work while you are asleep so that when you show up, you are not starting from nothing.

Once the morning briefing is running reliably, try adding a Friday wrap-up. Two automated tasks and suddenly both ends of your week feel different.

 

One more thing before you close this

Look, none of this is complicated. Building the folders takes an afternoon. Writing the two files takes an honest hour. Everything else is fifteen, twenty minutes of connecting things that are already there.

The reason most people do not do it is that it asks you to slow down and build before you use. And most people want the output right now. So they stay in Claude Chat. They get decent answers. They never know what they missed.

I was those people for three months. I do not think I could go back to that now even if I wanted to. Not because this setup is perfect, but because my mornings are actually different. The work starts differently. The output lands closer to what I needed without me having to chase it.

Set aside a week. Follow the ten steps in order. By day three something will click and you will stop thinking about how to use Claude and start just using it.

That is what the week is for.

 

Ruben Hassid's original guide, prompts, and templates live at how-to-ai.guide