Your AI-OS Learning Path — Dieuwke × FlowPhase

Why — the reason to bother

Why do this at all?

Not because AI is trendy. Because of what your weeks are made of: meetings that need follow-up, plans that need structuring, mails that need writing — and above all, content that needs producing: presentations, brochures, posters, client PDFs, checklists. That work is necessary — but it doesn't need all of your energy. Especially not the blank-page part.

⏱️

Reason 1

Get hours back

The recurring writing-and-structuring work — notes, lists, drafts, slide texts, brochure copy — shrinks from hours to minutes. Starting on day one, not someday.

🧠

Reason 2

Think with a partner

Not just faster typing — better thinking. A sparring partner that knows your context, asks good questions, and never gets tired of your "wait, one more idea".

🌱

Reason 3

Lead the change

Van Aanpak has decided AI matters, and you're the one guiding the team. The fastest way to lead a change is to live it yourself first — hands-on, on your own work.

🚀

Reason 4

Build the unimaginable

A checklist web page, a small tool, your own workflow — things that used to need a developer become an afternoon. You'll build things you could never imagine building yourself.

In short: this isn't about technology. It's about how your work feels three months from now — calmer Mondays, faster follow-ups, and the confidence of someone who understands the thing everyone else is only talking about.

How — the idea

What is an "AI operating system"?

You've probably used ChatGPT or Copilot: you type a question, you get an answer. Useful — but every conversation starts from zero. The AI doesn't know who you are, what you're working on, or how you like things done. So you explain. Every. Single. Time. And whatever it produces, you copy-paste somewhere and it's gone.

An AI operating system turns this around. Instead of you visiting the AI in a browser, the AI moves in with you — into one folder on your Mac. That folder contains a short briefing about you (who you are, how you work, what matters), plus your notes and projects. The AI reads all of it before it lifts a finger. And everything it produces lands back in the folder as files you keep.

The result feels less like using a tool and more like working with a colleague who has actually read the file. You say "turn my meeting notes into an action list" — and it knows which meeting, which project, and that you like your lists short.

That's the whole idea. No platform, no login-jungle, no IT project. A folder, a briefing, and an AI that reads it.

	Chatbot in the browser	Your AI-OS
Knows you?	Starts from zero every time	Reads your briefing first, every session
Your context	Whatever you paste in	Your notes & projects, right there in the folder
The results	Copy-paste, then gone	Saved as files you keep and reuse
Gets better over time?	No — same conversation forever	Yes — every briefing update makes it smarter
Who owns it	The platform	You. Plain text files on your Mac

How — the building blocks

You start with three things

Two free programs and a text file — that's genuinely the entire starter stack. The rest comes later, and only if you want it.

🗂️

VS Code

The desk

A free program that shows a folder and the files inside it — like a tidy desk where everything you're working on is visible at a glance.

You never have to "code" anything. For you, it's a viewer with a chat window.

🤝

Claude Code

The colleague

An AI assistant that works inside that folder. You talk to it in normal language — Dutch or English, both fine. It reads your files, writes new ones, and asks when it's unsure.

Not a chatbot you visit — a colleague who sits at your desk.

📋

me.md & CLAUDE.md

The briefing

Two short text files that tell the colleague who you are, how you work and what matters to you. It reads them at the start of every session — so you never explain yourself twice.

This is why it feels personal. And you can edit it anytime.

That's all you need for months. And when you've grown into it, three more building blocks are already waiting — you'll meet them on the Pro level and beyond:

📖

Later · skills/

The recipes

Written instructions the colleague can reuse: "how we write follow-ups", "how we build checklists". Write a good way of working down once — use it forever.

Your templates/ folder (Mission 9) is the first step toward this.

🧑‍🤝‍🧑

Later · Agents

The specialists

Extra colleagues with a specialty — a proof-reader, a planner, a researcher. Each has its own small briefing for a recurring role, and your main colleague can call them in.

Genuinely later: one colleague is plenty for months.

🔌

Later · Connections & workflows

The plumbing

The colleague can connect to other systems — mail, calendar, your Microsoft world — and recurring routines can run on a schedule. This is where a personal AI-OS grows into real automation.

A "together with Jan" topic — governance first, always.

How — for real

This is what your very first session looks like

Your folder on the left, the file being worked on in the middle, the conversation on the right — exactly how your cockpit will look. This is Step 0 live: the starter interview runs on the right, and your briefing takes shape in the middle as you answer. (On a phone: scroll sideways.)

dieuwke-os — Visual Studio Code

▾ DIEUWKE-OS
  START-HERE.md
  CLAUDE.md ✦ new
  me.md ✦ new
  ▾ brand/ ✦
    brand-kit.md [?]
  ▸ notes/ ✦
  ▸ actions/ ✦
  ▸ projects/ ✦
  ▸ templates/ ✦
  ▸ skills/ ✦

— three months from now,
  projects, skills and templates
  will have grown in here —

# CLAUDE.md — briefing
Ontstaat nu live uit het interview →

## Hoe we werken
- Korte antwoorden, standaard Nederlands
- Kies een lijn — Dieuwke corrigeert
- Onzeker? Markeer met [?], nooit gokken

## Huisregels
- Stoplicht: 🟢 eigen notities · 🟡 teamdocumenten
eerst afstemmen · 🔴 cliëntdata nooit
- Je verstuurt nooit iets
- Nooit verwijderen zonder akkoord

## Grootste tijdvreter
- Vergaderingen nawerken, contentproductie ← uit vraag 3

✳ Claude Code — starter-interview

Dieuwke: Lees START-HERE.md en laten we mijn AI-OS bouwen.

Claude: Graag! Eerste vraag: wat is jouw rol bij Van Aanpak, in je eigen woorden?

Dieuwke: Projectleider — en ik maak de meeste presentaties en brochures.

Claude: Genoteerd. Welke terugkerende taken kosten je de meeste tijd of energie?

Dieuwke: Vergaderingen nawerken. En content beginnen vanaf een lege pagina.

Claude: ✓ Ik heb CLAUDE.md en me.md aangemaakt — links zie je de nieuwe bestanden, in het midden je briefing. Zet je hierna je logo in brand/? Dan komt alles wat ik ooit voor je opstel meteen in Van Aanpak-stijl.

How — the bigger picture

How this fits Van Aanpak

Van Aanpak's AI roadmap runs on Microsoft — SharePoint, Copilot, Power Automate. That stays exactly as planned. This here is something different: your personal layer.

Track 1 · The team

Microsoft 365 + Copilot

The shared systems, where the team's work and data live — with permissions, governance and structure.

SharePoint knowledge base
Power Automate workflows
Copilot in Outlook, Word and Teams

Track 2 · You personally

Your AI-OS

Your private thinking and working cockpit — for everything before and around the shared systems.

Drafting, planning, structuring your own work
Meeting notes → action lists in seconds
The place where you learn what AI can really do

→ And here's the beautiful part: as Adoption Champion, everything you learn on Track 2 makes you better at driving Track 1. Same skills, same instincts — you just learn them hands-on, on your own work.

How — staying safe

The traffic light

One honest thing first: what you type into the chat is processed in the cloud, like with any AI assistant. Your files stay on your Mac — but the conversation travels. That's exactly why we sort content into three colours, and why the red rule has no exceptions.

Green — go

Your own notes, planning, drafts, ideas, to-dos. Everything that is about your work and your thinking — not about other people.

Amber — ask first

Internal Van Aanpak documents and team material. Fine in principle — but align with the team once before it goes into your setup.

Red — never

Client data, care data, anything about identifiable people (AVG). That belongs in Van Aanpak's governed Microsoft environment with proper agreements — never in a personal tool.

Rule of thumb: if a sentence names or describes a person in care, it's red. If you're unsure which colour something is — that's a perfectly good question for Jan, and asking it is the professional move, not the embarrassing one.

What — look inside

Real examples: one briefing, three skills

This is what those text files actually look like — a grown-up CLAUDE.md and three "recipes" from the skills/ building block. Click to read, copy — or download and drop them straight into your own folder as a head start.

📋 CLAUDE-example.md — what the briefing looks like after a few weeks ›

CLAUDE-example.md

# CLAUDE.md — example briefing
*A realistic example of what your briefing can look like after a few weeks. Yours will be personal — Claude writes the first version for you in the starter interview and it grows from there.*

## Who I work for
Read `me.md` at the start of every session — it says who I am, how I work and what matters to me.

## How we work together
- Short answers by default; I ask when I want detail.
- Default language: Dutch. Switch when I switch.
- Be opinionated: propose, don't list options. I'll correct you.
- Mark anything uncertain with **[?]** instead of guessing.
- Never invent facts. If you don't know, say so.

## House rules (non-negotiable)
- **Traffic light:** 🟢 my own notes, planning, drafts → fine · 🟡 internal Van Aanpak documents → I align with the team first · 🔴 client or care data (AVG) → **never**; stop and explain instead of processing.
- **You never send anything.** Mails and messages are drafted — I review and send.
- **Never delete files** without my explicit OK.
- **On-brand by default:** before drafting any content, read `brand/brand-kit.md`.

## Folder map
- `notes/` — raw input: meeting notes, transcripts, ideas
- `actions/` — action lists you produce
- `projects/<name>/` — one room per project; read its `project.md` first
- `templates/` — reusable structures; use one when I name it
- `brand/` — logo + `brand-kit.md`
- `skills/` — recipes; when I say *"use the X skill"*, follow that file exactly

## Standing routines
- **Monday:** I brain-dump → you build `notes/week-plan.md` (recipe: `skills/weekly-plan.md`)
- **After meetings:** notes → action list (recipe: `skills/meeting-notes.md`)
- **Friday:** short week close — done / carried over / parked

⬇ Download

🧾 skill-meeting-notes.md — the recipe behind Mission 1 ›

skill-meeting-notes.md

# Skill: meeting notes → action list
*When I say "use the meeting-notes skill" — or simply paste raw notes after a meeting — follow this recipe.*

## Input
Raw notes or a transcript file. Messy is expected — half sentences, mixed languages, all fine.

## Steps
1. Read the whole input first. Identify three kinds of content: decisions, actions, open questions.
2. Build the output in exactly this order:
   - **Decisions** — what was agreed, one short line each
   - **Actions** — one line each: what · owner · deadline (**[?]** where unclear)
   - **Waiting on others** — what and on whom
   - **Open questions** — worth raising next time
3. Keep the language of my notes.
4. Save as `actions/YYYY-MM-DD-<topic>.md` and show me the result.

## Quality bar
- No invented deadlines or owners — when unsure, include the item with **[?]**.
- Nothing dropped: better one **[?]** too many than one action lost.
- Short lines. No prose paragraphs.

⬇ Download

🗓️ skill-weekly-plan.md — the Monday ritual as a reusable skill ›

skill-weekly-plan.md

# Skill: the Monday week plan
*When I say "Monday plan" and brain-dump what's on my mind, follow this recipe.*

## Steps
1. If `notes/week-plan.md` exists from last week: read it first. Start the new plan with **Carried over** — what didn't get done and still matters.
2. From my brain-dump, build:
   - **Top 3 priorities** — be opinionated, pick for me; I'll correct you
   - **Day by day** — a realistic suggestion, respecting fixed appointments I mention
   - **Not this week** — parked, captured, out of my head
3. Save as `notes/week-plan.md` (overwrite last week's — history lives in git of my memory, not in files).
4. End with one question: *"What would make this week a win?"*

## Friday close (same skill, other end)
When I say "Friday close": read the plan, list **done / moves to next week / can be dropped?** — three lines, no ceremony.

⬇ Download

🎨 skill-brand-content.md — on-brand drafts, every single time ›

skill-brand-content.md

# Skill: on-brand content draft
*For presentations, brochures, posters and client PDFs. When I say "use the content skill", follow this.*

## Before anything
1. Read `brand/brand-kit.md` — colours, fonts, tone words, do/don't phrases.
2. Check `templates/` — if a matching template exists, use its structure.
3. Ask me your **3 sharpest questions** (audience, goal, must-contain) — then draft.

## Per format
- **Presentation:** slide by slide — title · key message · max 4 bullets · one speaker note. First and last slide carry the core message.
- **Brochure / poster:** sections with headlines and full copy. Headlines do the selling, body does the explaining.
- **Client PDF / one-pager:** structure for skimming — a reader who only reads headlines must still get it.

## Quality bar
- Tone = brand kit, always. If my input conflicts with the brand kit, say so instead of silently choosing.
- Mark every factual claim you're not sure about with **[?]** — I verify before anything goes out.
- Save to `notes/` (or the project room if I named one). Design stays my job — you deliver the raw material at 80%.

⬇ Download

The road ahead

Where this can go: from cockpit to automation

The path above makes you faster. The horizon behind it: things start running by themselves — always with you as the editor. Three stages, honestly labelled:

Today — this path

Your AI cockpit

You work with a colleague who knows your context. Notes become actions, drafts appear in your tone, content starts at 80%. Everything happens because you asked.

Next — connected tools

Your folder talks to your tools

Mail, calendar, your Microsoft world: the colleague reads and prepares across systems. The SharePoint and Teams missions on the Pro level are your first taste of this.

Horizon — automation

Routines run on their own

With a workflow engine like n8n, recurring routines run on schedule: incoming requests pre-sorted, meeting follow-ups pre-drafted, weekly reports prepared overnight — and every result lands as a draft for your eyes, never as an auto-send.

Stage 2 in practice

What "connected tools" actually looks like

Mail: the colleague reads a whole thread and drafts the reply in your tone — you still send from Outlook
Calendar: a briefing before every external meeting — who you're meeting, what's open, what the goal is
SharePoint & Teams: team documents summarised, transcripts turned into minutes — Missions 10 & 11 are exactly this
Content: brand kit + template library applied across everything — every deck, brochure and checklist starts on-brand

Everything here still starts because you asked. The step up is reach, not autonomy.

Stage 3 in practice

What "routines run on their own" looks like

Inbox triage: incoming requests sorted overnight — each with a suggested reply waiting as a draft
Meeting pipeline: Teams transcript lands → minutes, actions and follow-up are drafted before you're back at your desk
Friday report: the week summary builds itself from your project logs — every Friday at four, ready for review
Content factory: a checklist idea noted on Monday is a draft web page by Tuesday morning

Iron rule at this stage: every automation ends in a draft for your eyes — nothing leaves the house on its own. And every workflow has a monthly cost you know upfront.

How does all of that work? — explained simply

How do you connect a tool? (the thing called "MCP") ›

Think of MCP as a universal wall socket for AI. Any tool can offer such a socket — your calendar, your mail, a database. Plug your colleague in, and it gains exactly the abilities that socket offers: "read my calendar", "search these documents". Connecting one is a one-time step of a few configuration lines — the kind of thing you do once, in twenty minutes, together with Jan. The important part: every socket comes with permissions. Read-only is the default — the colleague may look, not touch. That's why stage 2 needs no new trust, only new plugs. It's like giving a colleague the key to one specific room — never the whole building.

How do you build an agent? ›

An agent is a job description in a text file — really. You write: "You are the proof-reader. You check tone, brand kit and clarity. You never change facts. You return a marked-up version." Save it, done: your main colleague can now hand work to this specialist and bring the result back. No code, no server — one page of clear instructions. If you can write a good briefing (you did, in the very first mission), you can build an agent. It's hiring by writing a one-page job ad.

How do you build a workflow? ›

A workflow is a recipe with a trigger: "WHEN a new mail with an attachment arrives → extract the key data → write a summary → put a reply in my drafts folder." In a tool like n8n you build this by connecting boxes on a canvas — every box is one step, every line between them means "then". It looks like a flowchart because it is one. The AI sits inside single boxes ("summarise this", "draft that") — the workflow around it decides when things happen and what goes in and out. Recipe thinking, not programming.

How does it all fit together? ›

Think in layers, each standing on the one below: (1) your folder + briefing = the memory · (2) the colleague = the brain that reads it · (3) MCP plugs = senses and hands into other systems · (4) agents = extra specialist brains for recurring roles · (5) workflows = the conveyor belt that starts work without being asked. And across all five layers: the traffic light and the draft rule. The beautiful part: remove any upper layer and the ones below still work. You can stop at any stage and still have something genuinely useful — that's the opposite of an all-or-nothing platform.

Who builds stage 2 and 3 — and what does it cost? ›

Stage 2 is mostly plugging in: an afternoon together with Jan, not a project. Stage 3 is real (small) engineering — that's exactly FlowPhase's home turf. n8n itself runs free or for a few euros a month; each workflow has known running costs that are on the table before it's built. House rule: you'll never discover a bill — you'll approve one. And every workflow ships with its review step built in.

→ Stage 2 and 3 are exactly what FlowPhase builds with teams. When you're through the path, that's the natural next conversation — you'll be the best-prepared person in the room for it.

Honest answers

Questions you're probably asking

Do I need to be able to program? ›

No. You'll type exactly one technical-looking command in your life (the install line in Step 0, which you copy-paste). Everything after that is normal conversation in normal language. VS Code looks like a developer tool because it is one — but you're only using the folder view and the chat.

Where do my files and conversations actually go? ›

Your files live in the folder on your Mac — they don't go anywhere unless you put them in the chat. What you type in the conversation is processed on Anthropic's servers (the company behind Claude), like with any cloud AI assistant. That's the honest trade-off, and it's exactly why the traffic light exists: green content in, red content never.

What if I break something? ›

Realistically: you won't. Your AI-OS is a folder of text files — there's no database, no server, nothing to corrupt. If a conversation goes sideways, close it and start fresh; your files and briefing are untouched. And Claude has standing orders never to delete files without your explicit OK.

Why not just use ChatGPT or Copilot in the browser? ›

Use them too! The difference is memory and output. A browser chat starts from zero and its results vanish into copy-paste. Your AI-OS knows your context (the briefing), works directly on your files, and everything it produces is saved and reusable. After Week 4 you'll feel the difference yourself — that's a promise.

Can I work in Dutch? ›

Completely. Claude speaks fluent Dutch — notes in Dutch, answers in Dutch, drafts in Dutch. You set your preference once during the starter interview, and you can always switch mid-conversation. Mixing languages is fine too; it follows your lead.

What does all this cost? ›

VS Code: € 0. Claude Code: € 0. The starter kit and this page: € 0. The only running cost is the Claude Pro subscription, ~€ 20/month, which powers the AI itself. No setup fees, no per-user licenses, no lock-in — cancel anytime and your files simply stay yours.

How much time does this really take? ›

Setup: ~30 minutes, once. After that: one small daily rep of 5–10 minutes, embedded in work you were doing anyway — the missions are designed to save time from day one, not cost it. If a day is crazy, skip it. The path waits; there's no falling behind.

Keep handy

Cheat sheet

Phrases that work well

"Ask me questions first" — before big tasks; the colleague interviews you instead of guessing
"Be opinionated, I'll correct you" — gets you a real proposal instead of a menu of options
"Mark uncertainty with [?] instead of guessing" — turns invisible guesses into visible questions
"Shorter." — totally fine. It's a colleague, not a professor
"Save this as a file in notes/" — anything worth keeping goes into the folder
"…and remember that" — it updates its own briefing; same mistake never twice

Mini glossary

Terminal — the text window inside VS Code where Claude lives. Type, press Enter, done
Markdown (.md) — a plain text file with simple formatting. Opens in any app, survives every tool change
Prompt — fancy word for "what you say to the AI"
CLAUDE.md — the briefing the colleague reads first, every single session
me.md — who you are and how you like to work; written by Claude in your starter interview
Session — one conversation. New session = fresh start, but files and briefing remain

If something feels off

Answer too generic? Give more context — or better, add the context to your briefing once
It made something up? Say so! "That's wrong, please check" works. Trust, but verify — you're the editor
Conversation went sideways? Close the session, start fresh — nothing breaks, your files are safe
Feels slow or stuck? Press Escape once, then ask again more specifically
Real problem? Screenshot + message to Jan. That's the deal

The essentials

Costs — VS Code € 0 · Claude Pro ~€ 20/month. That's the whole bill
Privacy — traffic light, always; red means red (AVG)
You're the editor — nothing goes out to anyone without your eyes on it; the AI never sends anything
Your files, your system — plain text on your Mac; no lock-in, ever
Bigger picture — what you learn here, you'll carry into the team (that's the Adoption-Champion plan)

Why do this at all?

Get hours back

Think with a partner

Lead the change

Build the unimaginable

What is an "AI operating system"?

You start with three things

The desk

The colleague

The briefing

The recipes

The specialists

The plumbing

This is what your very first session looks like

How this fits Van Aanpak

Microsoft 365 + Copilot

Your AI-OS

The traffic light

Green — go

Amber — ask first

Red — never

The path: Starter → Mid → Pro

Set up your AI-OS (~30 minutes)

🎨 Build your brand kit

🖥️ A web page from words

🎙️ Talk instead of type

📊 A deck in 10 minutes

Meeting notes → action list

The Monday-morning week plan

Drafting mails & messages

From messy notes to a clean document

Your content studio: presentations & brochures

The wow one: an online checklist for clients

Make the colleague smarter

Give every project its own room

Build your template library

Connect your Microsoft world: SharePoint files

Teams meetings: transcript in, actions out

Design your own use case

Pass it on

Real examples: one briefing, three skills

Where this can go: from cockpit to automation

Your AI cockpit

Your folder talks to your tools

Routines run on their own

What "connected tools" actually looks like

What "routines run on their own" looks like

Questions you're probably asking

Cheat sheet

Phrases that work well

Mini glossary

If something feels off

The essentials

Stuck? That's what Jan is for.