Day One · Session 01 · Dates TBD
Getting real work out of ChatGPT and Claude
Ship AIPhoenix, ArizonaPress → to begin
The blank box is the whole problem.
Everyone in this room already has an account. Almost nobody has a thread that ended in something they kept.
Four sessions, one business — yours. Every session is free and standalone, each one leaves a folder the next one reads, and catching up late is explicitly allowed.
Act one
Most people are running a search engine that types slowly. Five things separate that from a thread that hands you work.
Nobody writes this in order the first time. You write the task, get something generic back, and add the other four.
Chat as search
Chat as work
Same box, same model. The difference is whether anything survives the tab closing.
Every good thread ends in an artifact you keep.
A file, a page, a draft. If tonight only made you feel productive, it didn't happen.
Act two
The third time you explain your business to a model, you're doing unpaid data entry. A skill is that explanation, written once, read automatically from then on.
It's markdown with a bit of frontmatter. No framework, no platform — a folder your agent reads before it answers you.
Teach it once
/skill-writer
what do you re-explain every time?drafting frontmatter + steps…wrote .claude/skills/brand-voice/SKILL.md
It asks what you keep repeating, takes real examples rather than adjectives, writes the steps as instructions, then tests the result in a fresh thread — because a skill that only works in the thread that wrote it isn't a skill.
The first one everybody should write is voice. Feed it three things you actually wrote and let /brand-voice pull the do/don't pairs out of your own sentences.
Who it's for, in writing
/icp-profile
who pays, and what just happened to them1 disqualifier written downwrote 01-voice/icp.md
Three answers: who pays, the trigger that makes this urgent now instead of someday, and the disqualifier. The disqualifier is the one everyone skips and the one that makes every later draft specific.
Written as a skill, not a note — so the calendar, the outbound and the support agent all read the same definition instead of three different guesses.
Act three
One business, picked tonight, carried through all four sessions. Not a demo project — the thing you'd actually like customers for.
| Tonight | Where it goes | |
|---|---|---|
| Voice | A skill built from your own writing | Every draft, page and reply after this |
| ICP | Who pays, the trigger, the disqualifier | The calendar, the outbound, the agent |
| Positioning | One page you'd defend to a stranger | The hero of the site you ship in 03 |
| The draft | One article, claims marked | The pipeline that drafts them in 04 |
Nothing tonight is a warm-up exercise. Every file in the left column is an input to a session that comes after it.
The one-pager, then the first draft
/positioning-brief → /article-draft
reading brand-voice + 01-voice/icp.mdone-liner failed the swap test — retryingwrote positioning.md, article-01.md
The brief compresses everything above into one line, then applies the swap test: put a competitor's name in it, and if it still reads true it isn't positioning. The article comes out in your voice because the voice is a file it can read.
Drafted tonight, published when the claims check out. Every number in it gets marked verified or unverified before it goes anywhere.
Demos over memos.
Next
Tonight everything happened because you asked. Next session it happens on a schedule — a live content calendar, a weekly report that has already run once, and outbound drafted for you to send.
shipai.club · Free, always