Project 42Start learning
Module 1 of 5 · 35 min

Name the Layers in an AI Experience

Separate the model from the interface, context, retrieval, memory, tools, policy, and runtime around it.

Core concept

By the end

You will be able to

  • Distinguish a model from a chatbot, assistant, and copilot experience.
  • Attribute observed behavior to the responsible product or runtime layer.
  • Replace brand-wide claims with exact, version-scoped statements.
  • Record unknown behavior instead of guessing.
01

A model generates; a product surrounds it

A model maps supplied input and context to generated output. A product can add an interface, instructions, retrieval, memory, tools, identity, policy, deterministic code, and an execution environment.

When a product searches, remembers, edits, or sends something, identify the component that performed that behavior. Do not transfer every product capability to the model name.

02

Chatbot, assistant, and copilot describe different experiences

A chatbot is organized around conversational turns. An assistant helps a person complete tasks. A copilot is embedded in a work surface while a person remains responsible for direction and consequential decisions.

Any of these experiences can contain a deterministic workflow or agentic mode. The label alone does not prove who chooses the next step, executes tools, stores state, or approves effects.

03

Describe the exact experience

Name the product surface, version or observation date, configuration, enabled tools, permissions, and journey. Then label each claim documented, tested, inferred, or unknown.

A provider family can contain a model endpoint, consumer assistant, coding agent, framework, and managed agent service with different classifications.

04

Unknown is an honest result

Missing documentation does not prove capability or absence. An untested feature is not tested evidence, and a marketing claim is not a runtime trace.

Record what would resolve the unknown: a first-party contract, controlled run, permission inspection, tool trace, state record, or accountable review.

Practice activity

Map one familiar AI experience

  1. Choose one exact AI product experience and record its surface, observation date, configuration, and enabled capabilities.
  2. Map the model, interface, context, retrieval, memory, tools, policy, deterministic code, and runtime.
  3. For each claimed behavior, record responsible layer, evidence kind, source, and unknowns.
  4. Rewrite one provider-wide or marketing-led statement as an observable scoped statement.

What to produce

  • A complete layer map with no behavior attributed only to a provider name.
  • A claim table that separates documented, tested, inferred, and unknown evidence.

Reflect before continuing

Which capability did you initially assign to the model but later locate in the surrounding product?

Evidence

Sources and verification

Knowledge check

Make it stick.

Pass at 80%

Choose the strongest answer for each question. Your attempts become part of your device-local transcript.

01What does a model classification establish?
02A product is called a copilot. What can you conclude from the name alone?
03Which statement is properly scoped?
04What should happen when documentation does not resolve who stores state?
05Why separate a model from its harness?