Manager of Five · BeforeScaling

Leveraging AI as a Knowledge Worker

The concepts, skills, and habits that turn AI into more than a tool — so you get more done, do better work, and finally have something that thinks alongside you.

output mastery →
The Format

One-on-one, by design

Delivered one-on-one to achieve real personalization:

Every session increases your productivity from day one — you leave each one with something that works, not theory for later.

The Big Idea

The Three Levels of AI

Most people stop at the first two levels — and quietly cap how far they can go. The real leverage is at the top, and the top runs on the things only you bring.  Automation → Augmentation → Collaboration

Level 1

Automation of Logistics

The busywork of moving information around — gathering it, copying it, filing it, connecting one system to another. AI does the manual labor, faster and without slips.

e.g. Auto-file every new invoice and log it to your tracker.

Linear
Level 2

Augmentation of Content

Reshaping content that already exists — summarizing, drafting, reformatting, translating. You've already decided what it should say; AI gives it shape.

e.g. Turn a messy call transcript into a clean one-page summary.

Linear
Level 3

Collaboration as a Thinking Partner

A genuine back-and-forth partner. You think with it to reach answers you couldn't reach alone — shaped by your judgment, taste, and experience, the things the model can't know.

e.g. Talk through a hard call with something that knows your business and pushes back.

Exponential

Levels 1 and 2 are real wins — they free your time and teach you the tool. Once they're running, you're free for Level 3 — where your work starts to multiply.

Your Operating System

The 6 Skills — how you operate at every level

01

Context Assembly

Feed the model what only you know — background, constraints, examples, goals.

02

Quality Judgment

Tell genuinely good work from work that just sounds right — and spot what's missing.

03

Task Decomposition

Break work into the right pieces: what to hand over whole, what to split.

04

Iterative Refinement

Refine the output and the prompt — improve the machine, not just the output.

05

Workflow Integration

Wire AI into how you actually work: from one-off chat to repeatable process.

06

Frontier Recognition

Know where the tool's limits are — and whether you've given it enough to do the job well.

What You Bring

The 6 Human Dimensions

AI can match almost anyone's baseline work — but not these. They start to show at Level 2, and become your real edge at Level 3.

01

Taste

Your sense of what's good — the aesthetic and quality bar only you hold.

02

Context / domain expertise

What you know about your field that the model simply doesn't.

03

Judgment

The calls you make on the output: keep, kill, reshape.

04

Interpretation

Reading meaning and nuance the model takes literally.

05

Unique history

Your experience and scars — the situations you've lived that inform the work.

06

Risk profile

How much risk you'll accept — what must be safe vs. where you can be bold.

The Key Separation

Brain vs. Context

Rented · replaceable

The Brain (LLM)

The reasoning engine. General, powerful, stateless. A commodity that gets better and cheaper over time.

Owned · permanent

The Context

Your knowledge, files, examples, rules, history. The thing that makes the work yours. The durable asset.

The brain is rented and replaceable. The context is owned and permanent. Build on what you own.

Foundational Concepts

Vocabulary everyone needs

LLM

The "brain"

A reasoning engine — general, stateless, swappable. Knows nothing about you until you give it context.

Context

What you feed the brain

Its working memory for a task. More isn't always better — there's a sweet spot.

InstructionsDomain rulesExamplesFilesPrior outputsLive dataHistory
quality too little optimal too much
Prompt

The direction you give

Identity, task, context, constraints, output format.

Skill

A packaged job

A reusable instruction set that does a specific job the same way every time.

Artifact

A kept output

A concrete thing you keep — a doc, dashboard, report, file.

Connector (MCP)

A live link

A link to an external system so the brain reads and writes real data.

Scheduler / Trigger

What starts it

Makes work run on its own — on a clock or on an event.

Memory

What persists

Durable knowledge the brain pulls across sessions — in files you own.

Why Context Lives in Folders

Keep your context in plain files — not locked inside an app

  • Portability — files are yours and move anywhere. A better brain appears? Point it at the same folders.
  • No vendor lock-in — the model is a commodity; your context is the moat. Don't hand it to a vendor.
  • Transparency & control — plain text you can open, read, edit, and version. No black box.
  • Composability — the same folder feeds many brains, skills, and workflows.
Agentic Workflows

Give a goal — it runs the whole job

A normal prompt gives one answer. An agentic workflow carries out a multi-step process on its own — choosing steps, using tools, and checking its own work.

1

Trigger

You ask, a schedule fires, or an event happens.

2

Goal + Instructions

The objective + the skill that defines how.

3

Context / Sources

Folders (memory) + live systems (connectors).

4

Tools

Run code, search, edit files, call APIs.

5

Decisions + QA

Branching and checkpoints to catch errors.

6

Transforms

The actual work done on the data.

7

Destination

Where the finished artifact lands.

8

Memory

What persists so the next run is smarter.

Draw it on paper first:  TRIGGER · SOURCES · TRANSFORMS · DECISIONS · DESTINATION
Interpretable Context Methodology (ICM)

You need a folder, not a framework

Numbered folders are the stages, plain-text files carry the context, and every output is editable before the next stage runs — the folder structure does the orchestration.

Layer 0 · Identity
CLAUDE.md — "Where am I?" Routes the agent in.
Layer 1 · Routing
Root CONTEXT.md — "Where do I go?"
Layer 2 · Contract
Stage CONTEXT.md — Inputs · Process · Outputs.
Layer 3 · Factory
references/ — stable rules. Same every run.
Layer 4 · Product
output/ · runs/ — per-run material.

The rule that matters most: keep Layer 3 (stable) separate from Layer 4 (per-run).

workspace/
├── CLAUDE.md        ← L0 identity
├── CONTEXT.md       ← L1 routing
├── references/      ← L3 factory
└── stages/
    ├── 01_research/
    │   ├── CONTEXT.md   ← L2
    │   ├── references/  ← L3
    │   └── output/      ← L4 ▸ review
    ├── 02_draft/  …
    └── 03_final/  …
The Parallel

ICM vs. Agentic Workflow

Agentic pieceICM equivalentWhat they share
TriggerRunning the pipeline (drop input into stage 01)The signal that starts the job
Goal + InstructionsLayer 0/1/2 — identity, routing, contractWhat role to play, what "done" means
Context / SourcesLayer 3 references + Layer 4 inputsWhat the agent must know
ToolsLocal scripts (the non-AI steps)Work that doesn't need the brain
Decisions + QAThe review gate at each output/Quality checked before continuing
Destination / ArtifactLayer 4 output/ · runs/The kept result
MemoryLayer 3 factory + runs/ historyWhat makes the next run smarter

Same anatomy. An agentic workflow is the pieces of one autonomous run; ICM lays those pieces out as folders so a human can see and steer every step.

The Bet

Solve Levels 1 & 2 to free your time and learn the tool. Operate at Level 3 to multiply yourself.

← → or Space to navigate
1 / 12