2.01
What an agent is made of
Stage one was your bearings, and this stage is what you are building.
An agent is a chat with a job.
It has the same brain as the chat you already use, and what turns it into an agent is four things you give it, the same four you would give a new member of staff.
It knows the world, and it knows nothing about you.
Not your clients, not how you write, not what you decided last Tuesday, and it never will until you give it, which is what these four things are.
The brief
Stage four
The connectors
Stage five
The runbooks
Stage six
Working hours
Stage seven
What we call it
What it does
What it is in Claude Code
When it is built
The brief
What it does. Its standing instructions. Who it works for, how it behaves, what it never does, so you are not explaining what you want every time.
In Claude Code. CLAUDE.md, one short file at the root of the folder, read at the start of every chat.
Stage four
The connectors
What it does. Its logins, so it can do things and not only tell you things. Read the calendar, draft in the inbox, open the files.
In Claude Code. Connectors, one for each thing it reaches.
Stage five
The runbooks
What it does. How you want each recurring job done, and the background on your work it needs to do it.
In Claude Code. Skills, one short file per job, and the documents in the folder.
Stage six
Working hours
What it does. When it does the job. A time attached, so it runs without you and tells you what it did.
In Claude Code. The timer your computer already has, starting a skill at the hour you set.
Stage seven