Keep it alive
Context tells it who you are. Memory tells it what has already happened.
A system that knows your business but not what you decided last month will cheerfully propose the thing you rejected in April, and you will have the same conversation forever. The apps keep a little memory of their own, but it is the AI's choice, hard to see, and it stays in that one product. Your files are the record you can read and carry to a different AI. This lesson adds those files, and the thirty-second habit that keeps them true.
This is a thing you maintain, not a thing you finish. The good news is that most of the maintaining is the AI's job: in a folder it updates the files for you, and in a Project it drafts and you paste it back. Yours is remembering to ask.
MEMORY.md
What you settled, in one running file. Ask for it now: create a MEMORY.md in my AI OS: what we decided, when, and why. Start it with anything worth keeping from this conversation.
Then get into the habit of ending sessions by adding to it. The difference this makes is simple to say: an assistant who knows you versus an assistant who remembers you.
DECISIONS.md
What was decided, when, and why. Newest at the top.
Write an entry when a decision closes off an option. Not every choice. If reversing it would cost real time or money, it belongs here.
The field that matters is why, and being honest in it, including the unglamorous reasons. In six months that is the only part worth anything.
Add one real entry now. Something you decided in the last month.
The payoff: the AI can tell you you already considered that in April, and here is why you said no. Nothing else in this system does that.
STATUS.md, and closing a session
Where things stand. Overwrite it each time; it is a snapshot, not a log.
The habit costs thirty seconds at the end of a session:
Before we stop: update STATUS.md with where things stand, add anything we settled today to MEMORY.md, and if any decision today closed off an option, add it to DECISIONS.md with the honest reason why.
Opening the next one
Read before you write. When you sit back down:
Read STATUS.md and tell me where I left off, then check MEMORY.md and DECISIONS.md before suggesting anything we have already ruled out.
The other two habits
That loop is most of what separates a system from a chat. It means Monday morning does not start with ten minutes of reconstruction. Two more habits keep the rest of it true.
Fix it when it annoys you. Wrong tone, wrong assumption, wrong format: change the line in the instructions rather than correcting it again tomorrow. Every correction you type twice is a line that belongs in a file.
Once a quarter, audit the knowledge. Ask it: look through KNOWLEDGE, which documents are more than a year old and which ones contradict each other. A stale folder quietly poisons everything downstream, because the AI quotes a superseded rate card with exactly the same confidence as a current one.
None of the three habits takes a minute. Skip them for a month and the system decays into a folder of stale files; keep them and it gets better every week without you noticing the effort.
Stuck anywhere in this lesson? Reply to any Build Notes email and tell me where. I would rather fix the lesson than have you quietly give up.