Skip to content
WizardDynamics
The programme

From zero to AI-native. In three blocks.

A system that sticks — not a flash in the pan. Block I+II back-to-back, Block III one or two weeks later, so what you learned lands in real projects.

IHow the model actually thinks

An IDE is just a window into your folders.

Sounds techie. It isn't. Your folders on the left, Claude in the chat on the right. That's it. We're opening your folders — promise.

Mindset above all

Three sentences that shape every session.

This isn't a tool. It's your 24/7 assistant with a 250 IQ. Your full-time job from today on — a clean folder structure. Claude does the rest.

Company OS

A structure that tidies everything up.

One folder per project. All project folders inside one parent folder. What goes inside is up to you — as long as Claude knows where to look.

CLAUDE.md

Two files, one memory.

Global: how you write. Per project: where you stand. Claude reads both every time it starts — automatically. They're the reading glasses through which it sees your world.

# Firmen-OS · v1
Firma/
├── CLAUDE.md            ← Stil & Briefkopf
├── .claude/skills/      ← Skills
├── Projekte/
│   ├── 2026-001_Kunde_Mueller/
│   │   ├── CLAUDE.md
│   │   ├── Angebote/
│   │   ├── Korrespondenz/
│   │   └── Unterlagen/
│   ├── 2026-002_Schulz_GmbH/
│   └── 2026-003_Relaunch_Bauer/
└── Vorlagen/            ← Stil-Anker
Context hygiene

Like a desk — eventually you have to clear it.

The more tokens in context, the worse the model gets. Claude holds out longest — but everyone eventually drops off. /clear on topic change. /compact past 120k tokens. Better too early than dumb too late.

# Company-OS · v1 — sample structure, adapted to your industry.

IIPrompting through pain

Tools become leverage.

Three principles that change everything: context beats cleverness. Examples beat description. Allow follow-up questions — better yet, force them.

Anatomie · 5 Teile

A prompt is a brief, not a wish.

Five parts every good prompt needs. Role before task. Context before format. Format before constraints.

i. RoleYou are an experienced proposal writer on our team.
ii. ContextClient has been with us since 2024, mid-size budget. Latest proposal: @02_Proposals/2026-03-02.docx.
iii. TaskDraft a follow-up proposal for package 3 (retainer).
iv. FormatAbout 1.5 pages. Letterhead from @Templates/Letterhead.docx. Same tone as always.
v. ConstraintsAsk me 3 questions before you start. On pricing — flag your assumptions.
Skills

Your reusable gold.

A skill is an instruction plus logic that Claude loads automatically on certain triggers. Without skills you use 1 %. With the right skills you lever the same AI a hundredfold.

client-update

Summarise the state of play — what's happened since the last client contact.

proposal-draft

A proposal in your own style — pulls from /Templates automatically.

deadline-check

Scans every project CLAUDE.md for deadlines, sorted by date.

Permission modes

Plan Mode — or — Bypass Permissions.

Control isn't a feature — it's a position on a slider. You decide per task.

Mode 1 · safe

Plan

Claude thinks, writes a plan. Nothing is executed. Default for sensitive client data.

Mode 2 · default

On confirmation

Reads without asking, writes with OK. The standard — where you spend 80 % of the time.

Mode 3 · fast

Accept Edits

File edits without asking, bash with. Gold for drafting & refactoring.

Mode 4 · open

Bypass

Everything without asking. Isolated environments only, always with backup. Never with real client data.

IIIWhen assistants become a team

Your 24/7 staff.

Power features for those who really want to go deep. From „Claude calling Claude" to „a Claude team coordinating itself."

Sub-Agents

When Claude calls a second Claude.

A sub-agent is a specialised Claude that the main Claude spawns for a bounded task. Own brief, own tools, own context window. It works in the back room — and returns only the summary.

  • Keep context clean — token hogs stay in the back room.
  • Work in parallel — three sub-agents research at the same time instead of one after the other.
  • Constrain rights — the research agent can read but not write.
  • Pin down behaviour — one sub-agent per role, not per task.
Agent Teams · bleeding edge

From „Claude calls Claude" to „a Claude team coordinates itself."

Each helper is a full Claude session of its own — same task list, talking to the others directly, addressable individually. Sub-agents for speed. Agent teams for depth.

The CLI · where Claude Code actually lives

Composable with everything.

Output from git, grep, a mail export — piped in, an answer back out. Headless mode for scripts. Runs anywhere a terminal runs — over SSH on your company server, in a container, in the cloud.

# Four lines, four worlds: interactive · one-shot · pipe · write to file.

# 1 · einfach starten
$ claude
> Fasse die Akte zusammen.

# 2 · einmalig, ohne Sitzung (Headless)
$ claude -p "Welche Frist läuft heute?"

# 3 · andere Tools reinpipen
$ git diff Schriftsatz.md \
   | claude -p "Was hat sich geändert?"

# 4 · ausgeben, weiterverarbeiten
$ claude -p "Sachverhalt extrahieren" \
   > sachverhalt.md
Optional · 3 h each

Three add-on modules.

When you want to go even deeper. Mix and match. Book any time after.

+A

Design.

Canva integration, letterheads, client pitches and one-pagers in 10 minutes instead of 4 hours.

+B

Autonomous.

OpenClaw as a 24/7 runtime. Telegram bridge. Claude answers even when you're away.

+C

Marketing.

Win customers with AI. Content, SEO, Ads — the playbook we run daily across 100+ apps.

Three sentences that have to be said in every block

Mindset above all.

  1. i.

    This isn't a tool. This is your 24/7 assistant with a 250 IQ.

  2. ii.

    Your full-time job from today on — a clean folder structure. Claude does the rest.

  3. iii.

    When you're stuck, ask Claude. It will help you set yourself up.

Contact

Let's talk.

30-minute discovery call. We'll figure out whether the programme fits your company — free, no strings.

Book directly

Pick a slot.

Grab a time that works. 30 minutes, video. Confirmed instantly.

Or drop a line

By message.

If no slot fits — one line is enough. We reply within 24 hours.