You commented BS on the reel, or somebody sent you this. The promise was the same either way: set up an AI assistant so the boring parts of your job stop being your job.
I run my own version of this. A set of automated agents drafts my content and checks my facts while I am at my day job. You do not need any of that on day one. You need one assistant that actually knows your job, and that takes about an hour to set up.
Three quick wins are marked through this page. 0 of 3 banked so far.
Step one: set Claude up properly
Go to claude.ai and make an account. The free tier is enough. The paid plan lifts the limits once you are relying on it.
Then do the one thing most people skip. Stop using one-off chats and open a Project instead. A Project is a workspace that remembers context you load into it once, so every chat inside it starts already knowing your job. One-off chats are why most people think AI gives generic answers. It was generic because it knew nothing about you.
Step two: write the handover document
Inside your Project, open the project instructions and describe your job the way you would brief a new hire on their first day. This single document is most of the setup. Copy this and fill in the brackets:
You are my work assistant.
About me: [your role, the kind of company, who you report to]
My regular outputs: [for example: client emails, a weekly status
report, meeting summaries, project updates]
How I write: [for example: short sentences, plain words, no
exclamation marks, sign off with just my first name]
Here are two real examples of my writing:
[paste two emails or reports you sent]
Rules:
- Draft, never send. I review everything.
- If you are not sure about a fact, say so instead of guessing.
- Ask me before assuming anything about dates, names or numbers.
The two real examples matter more than the description. Claude picks up your tone from what you actually wrote, not from how you describe it.
Step three: hand over the first five boring tasks
Start with tasks where a decent draft saves you real minutes and a mistake costs you nothing, because you review before anything leaves your hands.
- Email replies. Paste the thread. "Draft my reply. Two short paragraphs, my tone, and commit to nothing beyond what I already said in the thread."
- Meeting notes. Paste your rough notes or the transcript. "Pull out every action item with an owner and a date. Then write a five-line summary for people who were not there."
- Status reports. "Here is what I did this week, as bullets. Turn it into my weekly status format." After step two, it knows the format.
- Spreadsheet cleanup. Upload the file. "Find duplicates, inconsistent naming and empty required fields. List the fixes before changing anything."
- First drafts of anything. "Give me a rough first draft of [the doc]. I will rewrite it. Your job is to kill the blank page."
A week of this and the pattern shows up. You have stopped producing drafts and started reviewing them. Reviewing is faster.
Tap each task above as you hand it over. 0 of 5 handed over.
Step four: connect your actual tools
In Claude's settings you can connect the places your work already lives, like Gmail, Google Calendar and Google Drive. Once connected, you stop pasting things in. "Find the latest thread from [client] and draft my reply" works directly, and "what does my week look like" reads your real calendar.
The rule from step two does not move. It drafts, you send. Claude can send email once Gmail is connected, and it asks for your approval first by default; my advice is to keep pressing send yourself anyway.
Step five: the level most people never reach
Everything above happens while you sit there. The next level runs without you. That is Claude Code on your computer, with tasks on a schedule. Mine run every morning before I am up. One drafts the day's content, another verifies facts against primary sources, and each one ends by writing a report I read in two minutes over coffee.
This level is a weekend project, not an hour, and you do not need it to get value from steps one to four. It is where this ends up once the drafting habit pays for itself and you start wondering what else can run without you.
The rules that keep you out of trouble
- Nothing secret goes in. No passwords, no card numbers, no client data your contract says stays internal. Check what your employer allows before connecting work accounts.
- Review before anything ships. It drafts, you decide. Every time.
- Numbers get checked. If Claude gives you a figure, ask where it came from, and verify it before it goes in front of your boss.
- You own the output. "The AI wrote it" has never once worked as an excuse.
Pick one task from step three and hand it over this week. That is the whole method: one boring job at a time.
If this worked, the next read is how the whole fleet fits together: The architecture of a multi-agent content operation.
If this helped, send me a DM and tell me which task you handed over first.
Questions people ask
Is Claude free to use?
There is a free tier and it includes Projects, so everything in steps one to three works without paying. Paid plans lift the usage limits.
Do I need to know how to code?
No. Steps one to four happen entirely in the claude.ai interface. Code only enters at step five, and that level is optional.
Can Claude send emails for me?
Once Gmail is connected it can, and by default it asks for your approval before anything goes out. My rule stays the same either way: it drafts, you send.
What should I never put into it?
Passwords, card numbers, and client data your contract says stays internal. Check what your employer allows before connecting a work account.