LANDMARK · 8 MIN · UNDERSTAND
Drawing the system
After this landmark, you can draw an AI feature end to end, mark where it can fail and who would notice, and use that drawing to make decisions about it.
↗
This concept is shared, but the Everyday lens is less central here.
You can still explore it. We’re showing the shared explanation and a related practical view without hiding the knowledge.
This landmark closes the ring by putting the pieces together. A real AI feature is a path: a request arrives, instructions and retrieved context are assembled, a model produces output, that output may trigger tools, and something is returned, stored, or sent. Almost every property people attribute to “the AI” is actually a property of one of those stages — a wrong answer usually comes from retrieval, an expensive feature usually from prompt construction, an unsafe action always from the execution step. Drawing the path is therefore the most useful single artefact you can produce about a system, because it converts an opaque judgment (“is this AI good?”) into answerable ones: what enters the context, what the model can call, which actions are irreversible, what is logged, and who finds out when it is wrong. It is also the thing that makes accountability concrete, because every arrow on the drawing belongs to somebody.
This is the canonical concept. It stays the same across learner lenses so personalization never changes the underlying facts.
◌ EVERYDAY LENS
What this looks like for you
Even for a consumer app, the three-line version is worth knowing: what does it see, what can it do, and where does what I type go? If you can’t answer those, you don’t know what you’ve agreed to.
MAKE A DECISION
A support assistant quotes an out-of-date refund policy to a customer. What is the most useful first diagnostic step?
CARRY THISDraw one AI feature you rely on as boxes and arrows. Mark every irreversible action, every point where untrusted content enters, and every stage that is logged.