FRIDAY BUILD + FIELD NOTES · ISSUE 06 · SEPTEMBER 18, 2026
I’ve been thinking about the difference between using AI and actually building with it.
A few weeks ago, I built a campaign website for a friend. The website turned out great. But that was not the part I kept thinking about after I closed my laptop.
I had also built the research flow, the page structure, the prompts, and the decision points that got us there.
The site was for her.
The system was mine to keep.
That is the shift I want you to start looking for. Not “can AI do this one thing?” but “if this works, how do I keep it from becoming a one-time win?”
In this issue
[ 01 ] The 10-minute move that turns one good AI result into a repeatable system
[ 02 ] The missing four-letter ingredient when your prompt gives you beige nonsense
[ 03 ] A Project does not make AI smarter. It gives the work somewhere to live.
[ 04 ] Keep this line: one question at a time
01 · THE FRIDAY BUILD
The work you already did is a system. Stop leaving it in the chat.
We do this all the time.
You spend an hour or two working with Claude or ChatGPT until it finally gives you the onboarding email that gets people to reply, the listing description that makes the right buyers ask questions, the intake form that stops the 14-email back-and-forth, or the client follow-up that keeps something important from falling through the cracks.
It works.
You get your life back for five minutes. You close the tab. You move on.
Then the same task lands back on your desk next month, and somehow you are explaining the whole thing from scratch again.
That is the part I want to fix.
When I built that campaign website, I did not just save the final site. I saved what produced it: the research flow, the page strategy, the prompts, the choices we made, and the bits I would absolutely forget the second I started something else.
The outcome was useful once. The system is useful every time.
Steal this: turn the finished work into a Skill
The next time you get AI to an outcome you are happy with, do not immediately move on. Give it one more job.
Paste this into the same chat while it still has the context:
You are helping me turn a good, real result into a process I can reuse.
We just created [DESCRIBE THE RESULT] for [WHO OR WHAT IT WAS FOR]. I want to repeat this process when I need [THE REPEATING TASK], without rebuilding it from scratch.
Use the conversation we just had, the finished result, and any files or examples I attached to document the process. Capture the steps, the decisions that mattered, the questions you need answered before you start, and what a good final result should look like.
Write a reusable Skill I can save and use again. Be clear about what must be reviewed by a human and what you should never assume or invent.
Then read what it gives you.
Add back the parts only you would know. The weird exception. The decision you made because of a specific client. The sentence you always want in the follow-up. The thing it got wrong the first time.
That is not annoying cleanup. That is you turning your judgment into an asset.
Save the finished Skill somewhere you will find it. A Claude Project is a good home if the work repeats around one client, one process, or one part of your business. Give the Project a specific job, add the instructions and source material it needs, then use it on the next real task.
Do not make one giant Project called “My Business” and throw every file you have ever touched into it. That is just a junk drawer with better branding.
Make it specific.
A Project for weekly content. A Project for client proposals. A Project for the intake process that currently lives in your head and four different folders.
Start with one thing you have already done well once.
02 · A USEFUL DETOUR
If your prompt keeps giving you beige nonsense, it is probably missing one of these.
The prompt is not the hard part. The thinking before the prompt is.
Last month, my husband needed to clean up an address list before a marketing campaign. He gave Claude three sentences and got back a clean spreadsheet with a confidence score for every address.
That worked because the prompt had all four parts of RAAC:
Role: “Act as a research marketing assistant.”
Ask: “Review every address on the attached for accuracy. Return a spreadsheet with a confidence score from 1–10.”
Attachments: The actual address list. Not a description of it.
Context: “The goal is to ensure the marketing campaign has the correct targeting addresses.”
Most weak prompts only include the Ask. Then we act surprised when the answer is vague. AI has to guess who it is, what to look at, and why the work matters.
Give it the job, the raw material, and the reason.
You will get less generic back immediately.
03 · WHAT I’M NOTICING
A Project does not make AI smarter. It gives the work somewhere to live.
I showed a room in Tulsa how I set up Claude for my business. The feedback afterward came in two versions:
“I had no idea it could do that.”
And: “That was exciting and completely over my head.”
Both were fair. I showed them a lot.
The first thing I would build is not a complicated automation. It is one Project with one job.
A Project gives the work a permanent place for the instructions, examples, client context, and source files that keep getting lost between chats. It means you stop opening a blank window and reintroducing your own business every Monday morning.
But the value is not that Claude suddenly becomes brilliant because you clicked a new menu item. The value is that you stopped making the work start from zero.
That is the actual pattern underneath most useful AI work.
04 · KEEP THIS
“Interview me one question at a time.”
Put that line in more of your prompts.
It works when you are building Project instructions. It works when you are trying to clarify an offer. It works when you are asking AI to help you see progress you cannot see from inside your own business.
One question at a time keeps the AI from racing ahead with a polished guess before you have done the thinking.
And honestly, you will often figure out what you need before it ever gives you the final answer.
If you do one thing this week, take one chat where AI finally gave you a result you liked and run the Skill prompt above before you close it.
Not next week. Not when the task comes back around. While the work still remembers what happened.
Hit reply and tell me what you are tired of rebuilding from scratch. I read every one.
— Jamie
P.S. The best Project is not the prettiest one. It is the one that has already helped you get a real task done twice. Build the smallest useful version first, then let the work teach you what it needs.