01In plain terms

What this rung really is

Level 1 is the first step off the chat window. With your workspace already set up, you hand a real piece of client work to an agent that reads your files, makes the changes, and runs the steps — and you drive it by talking, not typing. The whole rung is one proof: it can do work that ships, not a demo.

Two habits make it real and make it safe: describe the outcome out loud instead of typing syntax, and keep every version from the first line so nothing it touches is ever truly lost. Clear this and “AI helps me write” becomes “AI does the work, and I can prove it.”

Feeling behind, or starting over? That's normal — and it isn't starting from zero. Don't re-watch the demos or reorganize your files first. The one move is smaller than the overwhelm makes it look: pick a single task you already have to do today and do that one job through the agent, by voice, in about an hour. One job, one hour, one proof — clear it once and the rest of the ladder comes back to you.

02The move

Brief it like a junior — out loud

The whole skill of this rung is the brief: tell the agent the result you want, the context it needs, and the guardrails — then let it work. A good brief has three parts.

Outcome
what “done” looks like
Context
what it needs to know
Constraints
the guardrails

Say all three out loud and the agent does the rest. Miss one and it guesses.

One real brief, color-coded

Read the brand voice in voice.md and the cities in cities.txt, then draft twelve location pages, one per city each with a unique intro and our services, saved in a /pages folder.”

● context● outcome● constraints
How it all connects
by voiceundo —roll backYousay it out loudYour voicebuilt-in dictationYour agentdoes the real workYour project filesactually changedGit historyevery change savedReal client work you can always undo
03Set it up, step by step

Do this rung, for real

Everything you need is here — no tabs to chase. First, the jargon this rung throws at you, in plain words. Then the steps, with the exact things to paste or say.

A real task an agency owner would actually run

You run SEO for a dentist. Instead of hand-writing 12 location pages, you point the agent at the client's existing site copy and say what you want. It drafts all 12 — each with the right city, services, and a unique intro — reads the brand voice from a file you give it, and saves every version as it goes. A morning of copy-paste becomes a 20-minute conversation. That's L1: a genuine deliverable, done by talking, fully recoverable.

What is

Driving it by voice

Dictating to the agent instead of typing — you talk, your words become the instruction, the agent acts on the outcome you described. Your computer's built-in dictation is enough.

Why it matters here — You think in outcomes (“make all twelve pages match the brand voice”), not syntax. Speaking lets you brief a complex job in the time it takes to say it — the moment the keyboard stops being the bottleneck.

What is

Version control (git)

A folder that quietly snapshots every change, so you can undo anything without losing work. The standard tool is git, and your set-up workspace already has it on.

Why it matters here — Real client work needs an undo button you can trust. This is what makes the agent safe on a client's files — nothing it does is permanent, every step is recoverable.

01

Point the agent at a real task

In your set-up workspace, open the agent in the client's folder and hand it a genuine deliverable — not a toy prompt. The folder already has version control on, so every change is recoverable from the first line.

02

Brief it by voice

Turn on your computer's dictation (Mac: the mic / fn key; Windows: Win + H), put your cursor in the agent, and describe the outcome you want. Don't type syntax — say the job the way you'd brief a junior.

Say it out loud

“Read the brand voice in voice.md, then draft twelve location pages — one per city in cities.txt — each with a unique intro and our services. Save them in a pages folder.”

03

Save the work as you go

When a chunk looks right, have the agent save a snapshot — a “commit.” Each one is a labelled restore point; if a later change goes wrong, roll back and lose nothing. You can just ask the agent to do this by voice.

Save a restore point (or just ask the agent to)
git add .
git commit -m "Draft 12 dentist location pages"
# that snapshot is now permanent and recoverable
04What good looks like

How you know it's working

Before

Before L1: you paste prompts into a browser, copy answers back out by hand, and there's no record of how you got there. The AI “helps” — you still do the work.

After — a good L1 setup looks like
  • A real client deliverable was produced by the agent, not just suggested — you can point to the actual files.
  • You briefed it by talking, not by typing commands or syntax.
  • Every step is saved in version control — `git log` shows the history and you can undo any change.
  • The work lives in a folder on your machine, not trapped in a chat thread you'll lose.

Make it stick. Make this your default for one recurring task you already do — location pages, a weekly report, a content refresh. Don't automate everything at once; run the next real instance of that ONE job this way instead of by hand, and the rung becomes a habit, not an experiment.

05The skills that get you here

Our skills for this rung

01Do a real client task through a frontier model — not a toy prompt
02Drive it by voice, so you describe outcomes instead of typing syntax
03Put the work under version control from the first line

Linked items are founding-circle skills — clone the repo and run ./install.sh from the skills folder. Unlinked items are practices you build by doing.

06Your checkpoint

A piece of real client work done via a frontier model, by voice, in version control — link the commit.

Clear this and you've genuinely cleared the rung — not read about it. Keep the proof; it's how you place yourself on the ladder.