| name | first-real-job |
| description | Do ONE real client task through the agent, by voice, in about an hour — and prove it shipped. The Level-1 on-ramp for anyone who feels behind or is starting over: it shrinks "learn this whole new way of working" down to a single job you already have on your plate today. Walks you through the first fifteen minutes, gives you the exact words to say out loud, saves every step so nothing can be lost, and ends with a checklist that proves the work is real. Use when you're overwhelmed and don't know where to start, when you're re-joining after time away and everything feels like a fire hose, when you've watched the demos but haven't done a real task yourself yet, or when you want your first genuine win instead of another practice prompt. |
first-real-job
The problem this solves: you've seen what the agent can do, but you haven't done a
real piece of your own work with it yet — and the gap between "I get the idea" and "I
did it on a live client task" feels bigger than it is. Maybe you're starting over after
some time away and it feels like drinking from a fire hose: new tools, new words, a
hundred things you could set up. That overwhelm is the thing standing between you and
your first win, and this skill exists to remove it.
The move is deliberately small:
Pick one real task you already have to do today. Do that ONE task through the agent,
by talking to it, in about an hour — and save every step so you can point to the proof.
Not a demo. Not "learn everything first." One genuine deliverable, start to finish, by
voice. When it ships, you'll have done the whole Level-1 loop for real — and you'll never
again wonder whether this works on your work.
Why an owner should care, in money terms: the first real task is the one that turns hours
of your week into minutes for the rest of the year. The point isn't to learn a tool — it's
to take a job you already bill for (a batch of pages, a client email, a report you dread)
and prove the agent can do it to your standard, so the next fifty come back in a fraction
of the time. Everything else you'll build later compounds on this single proof.
If you're starting over, read this first
Restarting is normal. It is not starting from zero. If you stepped away, or the
last attempt fizzled, or you feel a step behind the room — none of that is a problem to
fix before you begin. The people who get the most out of this didn't set everything up
perfectly first; they did one real thing, saw it work, and let momentum carry the rest.
So don't re-watch the demos. Don't reorganize your files. Don't try to remember where you
left off. The one move is this: pick a single task on your plate today and do it through
the agent, right now, by voice. One hour, one job, one proof. That's the entire re-entry
plan — the rest of the ladder comes back to you the moment you clear this once.
Before you start (a 2-minute check)
You need three things, and you almost certainly have them already:
- The agent, open in a folder. Your setup from onboarding put the agent (Claude Code)
in your terminal. Open it inside a folder that holds a real client's work. If you're not
sure it's installed, the setup guide gets you there in about fifteen minutes — do that
first, then come back here.
- A way to talk to it (dictation). You're going to say the job out loud, not type
it. Your computer's built-in dictation is enough — no extra app.
- On a Mac: press the microphone / 🎤 key, or tap the fn key twice, then start
talking. (If it's off: System Settings → Keyboard → Dictation → turn on.)
- On Windows: press Windows key + H to start voice typing, then talk. (First time,
Windows may ask permission to use the mic — say yes.)
- Put your cursor in the agent's prompt, start dictation, and speak your instruction like
you'd brief a smart new hire.
- One real task you can point at. Not a made-up example — something on your list today.
Good first jobs (pick the one that's yours):
- draft a batch of similar pages (location pages, service pages, product blurbs);
- write or rework a client email or proposal section;
- turn a messy set of notes into a clean one-page report;
- refresh the copy on a page to match a brand voice you can hand it.
Keep the first one low-stakes. Real, but not the thing that goes live to a client in
five minutes. You want room to watch it work and roll back freely. Nothing here can be
permanently lost — that's the point of the saving step below — but a calm first run beats
a high-pressure one.
The first 15 minutes (a script you can follow line by line)
Minutes 0–3 — Open the agent where the work lives.
Open the agent inside the folder for the client whose task you picked. It can now read the
files in that folder and make changes there. Turn dictation on (see above) so you're ready
to talk.
Minutes 3–5 — Tell it the outcome, out loud.
Don't describe steps or type commands. Say the result you want, the way you'd brief a
person. Use the "what to say" prompts in the next section — pick the one closest to your job
and change the details to yours.
Minutes 5–12 — Let it work; steer in plain words.
It'll read your files and start producing the work. Watch what it does. If it goes a
direction you don't want, just say so — "no, keep the intros shorter" or "use the pricing
from services.md, not made-up numbers." You're directing by talking, not fixing by typing.
Minutes 12–15 — Save the work as a snapshot.
When a chunk looks right, tell the agent to save it (a "commit" — a labelled restore point).
You don't have to know the commands; you can just ask. From now on, every step is
recoverable — if a later change goes wrong, you roll back and lose nothing.
That's the loop. One real task usually finishes inside the hour. When it does, run the
success checklist at the bottom — if you can tick every box, you've cleared Level 1 for real.
What to say (copy these, swap in your details)
These are spoken briefs — say them out loud with dictation on. They follow a simple shape:
the outcome you want, where to find your inputs, and where to put the result.
For a batch of similar pages:
"Read the brand voice in voice.md and the list of cities in cities.txt. Then draft one
location page per city — each with a unique intro, our list of services, and a clear call
to action in that brand voice. Save them in a folder called pages, one file per city."
For a client email or proposal section:
"Read the brief in this folder and the notes I saved in notes.md. Draft a client email
that covers what we're proposing, in a plain, direct voice — no hype. Keep it under 200
words and leave a bracket where I need to add a number I haven't given you."
For a messy pile of notes → a clean report:
"Take the notes in notes.md and turn them into a one-page client update: a short summary
at the top, then the three things we did, then what's next. Plain language, no filler.
Save it as update.md."
To keep it honest (say this any time):
"Don't invent facts, numbers, or client details. If you need something I haven't given
you, leave a clear [NEED THIS] tag instead of guessing."
To save your work (say this whenever a chunk looks right):
"Save a snapshot of the work so far with a short label saying what you did, so I can roll
back to here if I need to."
If you'd rather run the save step yourself, it's the same two lines on Mac and Windows:
git add .
git commit -m "First real task: 12 location page drafts"
What a good result looks like
By the end of the hour you can point at something real:
- The actual files exist. The pages, the email, the report — you can open them, not just
read about them in a chat window.
- You briefed it by talking. You described outcomes out loud; you didn't type syntax or
memorize commands.
- Every step is saved. Asking the agent to
git log (or running it yourself) shows a
history of snapshots — proof of how the work got made, and an undo button for every step.
- It's on your machine, in a folder — not trapped in a chat thread you'll lose track of.
That's the whole Level-1 proof: a genuine deliverable, done by voice, fully recoverable.
Success checklist (tick all four = Level 1 cleared)
Hit all four and you're standing on Level 1 — the first rung off the chat window. The next
move (writing your standing rules down once so you stop re-briefing) is Level 2, and it gets
easier from here because you're no longer wondering if this works on your work. You've
proven it does.
If it goes sideways (when to stop and re-steer)
Stop the agent and say what you want instead — calmly, in one line — when it:
- drifts outside the task (starts editing files you didn't ask about) → "only work on the
files for this task; leave everything else alone";
- invents details (a stat, a price, a client fact you never gave it) → "don't guess — leave
a [NEED THIS] tag where you're missing something";
- claims it's done without proof → "show me the actual files you changed and save a snapshot."
None of these are failures — they're the normal back-and-forth of directing. Because every
step is saved, the worst case is "roll back and rephrase," never "lost the work." That safety
is exactly why you can be bold on this first run.
Make it your default
The rung sticks when the second real task goes the same way as the first. Pick one job you do
every week — the recurring report, the batch of pages, the weekly client note — and run the
next real instance of it through the agent instead of by hand. Two or three times through and
it stops being an experiment and becomes how you work. That's when an hour saved this week turns
into a day saved every month.
