Most AI forgets who you are every time you open it. So people try it once and go back to doing the work by hand.
An agent is different. It is a folder of plain files, and we write them for your business.
Below is every file we put in one, what it does, and why it is there. Nothing is held back. If you want to build this yourself, you have enough here to start.
Agent Workflows
Set the goal. We’ll put AI to work on it.
An agent is an AI helper set up to do a specific job for your business. It uses your information and instructions, starts when you choose, and follows rules for what it can do automatically and what needs your team’s approval.
Illustrative example
Requests come in.
Friday
Lead inquiry form Awaiting reply
4 inquiries
2:15 p.m.
Name
Alex
Your request
Is next Friday at 2 p.m. available?
Saturday
Lead inquiry form Awaiting reply
5 inquiries
9:40 p.m.
Name
Jordan
Your request
Could I book a time next week?
Sunday
Lead inquiry form Awaiting reply
3 inquiries
6:20 p.m.
Name
Taylor
Your request
Do you have anything after 3 p.m.?
↓
Every morning, your agent checks open times and emails options.
Every day
8 a.m.
Check inbox
Gather details
Check times
Email options
It checks its work, follows your rules, and flags anything unclear.
↓
You confirm the booking.
8:17
Friday: 3 pm or 5 pm? Inbox
B
Bookingsto Alex⌄
8:05 a.m.···
bookings@yourcompany.example
Hi Alex,
Thanks for reaching out. Friday at 2 pm is already booked, but we have two openings later that day:
Friday at 3 pm
Friday at 5 pm
Which works better? Just reply with the time. If neither fits, no worries, send us one that does.
Neither time is held yet. Once you reply, Maria will confirm your spot.
Talk soon, Your Company
A
Alexto Bookings⌄
8:17 a.m.···
3 pm is perfect. Thanks for fitting me in!
Reply→ Forward
Your agent finds three stories worth posting about. You pick one.
Every day at 7 a.m. · 3 stories
Story 1
✓ You picked
Small shops use AI to answer customer questions
Why it fitsYour customers ask after hours
Story 2
New booking tools for service businesses
Why it fitsCould make booking easier
Story 3
How shops keep regulars coming back
Why it fitsFits your loyal customers
↓
Your agent researches your pick and drafts a post in your voice.
After you pick
Read your pick
Research
Draft post
Check facts
It checks its work, follows your rules, and flags anything unclear.
↓
You approve, and the post is lined up in your scheduler.
8:17
‹Drafts···
Y
Your companyDraft · waiting for approval
Got a quick question for us after we close?
Some small shops are trying AI to answer customer questions. We’re curious what you’d want.
Would you rather get a quick answer at 9 pm, or a call back from one of us the next morning? Tell us in the comments.
Image
A renewal is coming up.
Customer account
Jordan Lee
Plan renews
Oct 15
Saved spot
Saturdays, 9 am
Account notes
“Moving across town in September.”
Next step
Check in before renewal
↓
A month before, your agent checks the plan and account notes, then drafts a check-in.
30 days before each renewal
Check dates
Review terms & activity
Draft check-in
Flag gaps
It checks its work, follows your rules, and flags anything unclear.
↓
You approve the message.
8:17
Your Saturday spot Drafts
S
DraftSam to Jordan⌄
Hi Jordan,
It’s been so nice seeing you on Saturday mornings this year.
You mentioned you were moving across town this month. With your plan renewing October 15, I wanted to check that Saturdays will still be easy after the move.
Does 9 am still suit you, or would another time work better now? No rush at all.
Thanks, Sam
00
How the three fit together.
An agent that already knows your company, your product, and your tools runs the complicated, repeating work on a schedule. That gives your people their hours back — so they spend them on quality instead of volume, which is how you win trust and build credibility.
Your agents work in your projects, using your skills. All of it routes to human approval.
Your agents
Who does the work.
Marketing agent
Support agent
Sales agent
work in
Your projects
Where the work lives.
Brand profile
Customer accounts
Forecasting
using
Your skills
How it gets done, every time.
Draft the social post
Draft the reply
Write the weekly report
Routes to human approvalNothing goes out until you say yes.
01
The whole shape, on one screen.
One agent, one skill, one project, and every file in each. The lists below say what each file is for.
One agent, one skill, one project. Every file. It is a lot — and building it is exactly what you would be hiring us for.
your-business/
├── agents/
│ └── sales-agent/
│ ├── Memory.mdit remembers your business
│ ├── Learner.mdit fixes its own mistakes
│ ├── Principles.mdit will not break your rules
│ ├── Agent.md
│ ├── README.md
│ ├── Identity.md
│ ├── Purpose.md
│ ├── Soul.md
│ ├── Heartbeat.md
│ └── evals/criteria.md
├── skills/
│ └── write-the-weekly-report/
│ ├── SKILL.md
│ ├── evals/criteria.md
│ ├── learner.md
│ ├── agents/
│ ├── references/only when the job needs it
│ ├── scripts/only when the job needs it
│ └── assets/only when the job needs it
└── projects/
└── new-business/
├── project.md
├── README.md
├── AGENTS.md
├── links.md
├── memory/
├── learning/
└── receipts/
02
Inside one agent
An agent is a folder, not a chat window. These ten files are what turn a general-purpose AI into one that knows your business, keeps its promises, and gets better instead of drifting. Every one of the 56 agents we run has all ten.
Agent.mdWhat it does all day.
README.mdHow to start it up.
Identity.mdWho it is.
Purpose.mdWhy it exists.
Soul.mdHow it talks.
Principles.mdRules it will not break.
Memory.mdWhat it remembers about you.
Learner.mdMistakes it already fixed.
Heartbeat.mdWhat it checks on a schedule.
evals/criteria.mdHow we grade its work.
The same folder, in plain words
Instructions
The job description.
Rules
What it never does, and what needs a person's yes.
Memory
It remembers your business.
Learning
It fixes its own mistakes.
Checklists
Quality checks before anything ships.
Tools
It can act in your systems: web browsing, and computer use. It uses your computer the way a person would.
Routines
It runs on its own at set times.
Sign-off
A person approves before anything goes out.
03
Inside one skill
A skill is one job done the same way every time — write the weekly report, screen a candidate, close the books. It is a much smaller folder than an agent on purpose, so a new one is cheap to add. We run 155 of them across 8 categories, and agents reach for them as needed.
SKILL.mdWhat the job is, and when to use it.
evals/criteria.mdHow we grade the result.
learner.mdMistakes already fixed.
agents/Which agents may run it.
When the job needs it
references/Source material it has to get right.
scripts/Code it runs instead of guessing.
assets/Templates and files it works from.
04
Inside one project
A project is where a body of work lives — the goal, the decisions, the receipts. This is the part most teams skip, and it is why their AI work cannot be picked up by anyone else six weeks later.
project.mdWhat we are building.
README.mdHow to pick it up cold.
AGENTS.mdWhich agents work on it.
links.mdEvery doc and tool it touches.
memory/What the project already knows.
learning/What went wrong, and the fix.
receipts/Proof of every run.
05
Runs where your team already works
These are plain text files, so they are not tied to any one assistant. The same folder works in ChatGPT, Claude, and Gemini — which means switching later costs you nothing, and nobody can raise a price on work you already own.
ChatGPT
Claude
Gemini
Want this built for your team?
We run 56 agents and 155 skills inside our own business. This is how every one of them is built.