Harlem Labs Advisory · Framework
Context is the product.
Everyone has the same model. The difference between your result and everyone else's is entirely what the machine read before it answered. That layer is the only part you own.
The problem nobody names
Every session starts at zero.
Every AI assistant an operator uses — whichever brand is on the box — opens each conversation as a stranger. Brilliant, fast, tireless, and carrying nothing forward. Not your clients. Not your standards. Not the correction you gave it yesterday.
0
What it knows about your business on arrival
What changes the outcome
A briefed system reads for two seconds first.
Measured on a working Harlem Labs machine, this is what gets read before a single word of the answer is written.
378
lines of rulebook
identity and standards
159
memory files
what already happened
891
permission notes
what it may do freely
33
saved procedures
repeatable moves
9
hard nevers
the only real brakes
The framework
The Briefing Stack
Five layers. Every tool on the market has all five under different names. Most operators have built one.
01
IDENTITY
Who you are, who you serve, what you sell
Custom instructions, or a project file the tool reads first
02
RULES
How the work must be done, and what finished looks like
Voice, format, terminology, spelling, deliverable shape
03
MEMORY
Decisions already made, so they stay made
Saved notes the tool can reach. The layer that compounds
04
LIMITS
What must never happen, under any instruction
A deny list is a lock. A written rule is only a request
05
MOVES
The procedures you run more than once
Saved prompts, skills, commands. Built last, not first
Defense before offense
The order it reads is not the order you build.
It reads
Identity → Rules → Memory → Limits → Moves
Top to bottom, every session, in about two seconds.
You build
Limits → Identity → Rules → Memory → Moves
Put the locks on before you hand over the keys. Moves are the fun part, which is exactly why they are the part everybody builds first.
Test your own setup tonight
Three questions.
1
If you opened a new session right now, what would it already know about your business without you typing anything?
2
What is on your never list — and have you ever actually opened the file it lives in?
3
What correction have you now given the same tool twice, and why is it not written down yet?
You are not using AI.
You are writing the briefing a stranger reads every morning.
Map → Build → Evolve. Capability, not dependency.
Build the layer, not another tool.
Most operators do not need a new platform. They need the briefing layer written down once, in the right order, with the locks on first. That is the work we do.