Making Claude
your assistant.
The four things you asked on the forum: connected chats, social posts, your inbox, and writing that sounds like you.
Claude runs a loop
On its own, Claude just predicts words — it chats. What turns that into real work in your files is a loop it runs the moment you ask.
You can keep adding context while it runs — or press Esc to stop.
Weak at steps — so we gave it recipes
Its one weak spot: following a long list of steps. It starts fine, then drifts.
Your recipe, followed the same way every time — that's the gap it fills.
A worker and its team — all reading one CLAUDE.md memory.
Same order today: memory first, then the assistants that run on it.
Two questions, in order
The one you asked most. We start here, and explain it fully.
Then we turn it into assistants that do the work — social, inbox, writing.
First the memory. Then the work.
Stop using it like a chatbot.
Set it up like an assistant.
A chatbot forgets you every time. An assistant knows your business, your voice and your goal — and does the work. Every question today is really this one.
How do we connect all of these?
Every one a separate chat — each starting from zero.
“Do my chats know each other?”
You open a new chat and it feels like a stranger. You explain your business all over again. And that chat from yesterday? This one has never heard of it.
So let's be honest about what's really going on. →
Claude can't see your other chats.
Every conversation starts fresh.
That's not a fault. Each chat is its own room with the door closed. What connects them isn't the chat — it's what you keep outside the chat.
You can't merge them — but there are two things you can do about it.
Two things you can do
The important, stable things Claude should know about you — in CLAUDE.md / memory. Set once, read in every chat.
For what you were mid-way through: let your old chat write a prompt that briefs the new one.
One runs in the background. One you trigger when you start fresh.
What Claude should always know
The stable stuff — your business, your voice, your preferences — lives in CLAUDE.md. Write it once; every chat reads it.
Your facts and rules, in plain language. You stay in control of what's in there.
Claude also keeps a general memory by itself, in the background, across your chats.
Together: the “who you are” you never have to re-explain.
Let one chat brief the next
Starting fresh on a task? Don't re-explain it. Ask your current chat to write the prompt for the next one — then paste it in.
Do this a lot? Turn it into a skill — one command that writes the handoff every time.
How do we make this pleasant?
The terminal does the job — but it's a bare place to live. There's a calmer, more visual window onto the same Claude: Cursor.
Nothing changes about what it can do — just a friendlier place to work.
Your project, chat and changes side by side — no black screen.
This is Cursor
Two things worth a look — your terminals, and the status bar.
Three assistants we're building
You've got the foundation. This week we focus on building three assistants on top of it — one for each question you asked.
One idea → on-brand posts, captions and hooks, in your voice.
Summarizes long threads and drafts replies in your tone.
Writes anything — pages, newsletters, posts — the way you'd say it.
Same memory underneath all three — that's why they sound like you.
Claude as your marketing assistant
This is where the memory pays off. Because Claude already knows your brand, it writes posts and ads that sound like you — not like generic AI.
Set up once — every post comes out on-brand.
One idea → many posts, captions and hooks, in seconds.
The better your memory, the less you edit afterwards.
One product, three posts
Same tool. The specific ask plus your brand memory = posts you barely touch.
Make an Instagram post
- Write posts in my brand voice
- Handle carousels and single posts
- I approve — nothing posts on its own
Less time in your mail
You stay in charge: you read and send. Wiring it to your real inbox is a later step — for now, paste the email in.
Triage & replies, every morning
- Check my inbox every morning
- Sort important from spam
- Draft replies I only have to send
Writing that sounds like you
Default AI writing has a smell: tidy, generic, a little soulless. The fix isn't writing a perfect prompt every time — it's teaching Claude your voice once.
Same move as your brand: capture your voice, then reuse it everywhere. →
Default AI vs. your voice
A voice skill (plus a humanizer pass) strips the AI shine. It reads like a person — you.
One voice skill, many tones
- Write in my voice, per channel
- Keep tones apart — LinkedIn, email, newsletter
- Strip the AI shine
One rhythm — three assistants
They all follow the same five steps. Claude does the middle bit; you stay in charge at the end — you approve, you send.
Runs by itself on a time — inbox in the morning, newsletter at night.
You call it when you need it — a post, or a piece of writing.
If this were your assistant…
- What would matter most to you — before you'd trust it?
- What work should it take off your plate first?
Your answers shape the assistant you build this week.
Ellen and I are building these.
Keep an eye on us next week — we're building these three assistants for real, and sharing how it goes.
- Social — on-brand posts in your voice.
- Inbox — morning triage and draft replies.
- Writing — one voice, many tones.
Friday — we share an update.
Questions along the way?
- Stuck this week? Post it on the forum.
- Friday 9:00 — forum session. We answer the week's questions together.
See you Friday