What modern AI systems do
Tell which part of an AI system does each job.
In this lesson · F1 of 8
All lessons in this track →Progress for this track is saved in this browser only. No completion badge is issued. View your learning.
Start with the idea
Tell which part of an AI system does each job.
The model writes the answer. The app is the service you use. Other software gives the model information and controls its tools.
Harness: The software that connects a model to instructions, information and tools. Its surrounding controls must enforce permissions.
See the example and check your understanding
A made-up teaching example
Imagine asking a travel assistant if it will rain tomorrow. A weather tool supplies a current forecast. The model turns that forecast into an answer. Writing about rain is different from checking the forecast.
Ask yourself
Without a weather source, does a confident answer prove tomorrow’s weather?
Show the explanation
No. The assistant needs current weather evidence. If that evidence is missing, it should say the answer is unknown.
Start with these checks
- Find the source of current facts.
- Check which tools can act.
- Keep unknown facts marked as unknown.
Use the different practice case below and check every requirement in its full instructions.
The video explains the main idea. Use the sources, instructions and checks below to complete the practice.
Captions appear here during playback.
What modern AI systems do
Tell which part of an AI system does each job.
Video playback is unavailable. Read the descriptive transcript below, go directly to practice, or open the video file.
Source-based teaching example · Fictional data · Reviewed 2026-10-03. Microsoft Mark · locally synthesized. On a small screen, use fullscreen for diagram details or read the transcript.
Read the descriptive transcript
0:00 · What you will learn
Tell which part of an AI system does each job.
What appears on screen
Tell which part of an AI system does each job.
0:05 · The idea
The model writes the answer. The app is the service you use. Other software gives the model information and controls its tools.
What appears on screen
The model writes the answer. The app is the service you use. Other software gives the model information and controls its tools.
0:17 · One useful term
Harness. The software that connects a model to instructions, information and tools. Its surrounding controls must enforce permissions.
What appears on screen
The software that connects a model to instructions, information and tools. Its surrounding controls must enforce permissions.
0:30 · A worked example · 1
Imagine asking a travel assistant if it will rain tomorrow. A weather tool supplies a current forecast. The model turns that forecast into an answer.
What appears on screen
Imagine asking a travel assistant if it will rain tomorrow. A weather tool supplies a current forecast. The model turns that forecast into an answer.
0:42 · A worked example · 2
Writing about rain is different from checking the forecast.
What appears on screen
Writing about rain is different from checking the forecast.
0:49 · Your turn to think
Without a weather source, does a confident answer prove tomorrow’s weather? Pause and decide before continuing.
What appears on screen
Without a weather source, does a confident answer prove tomorrow’s weather? Pause the video if you need more time.
1:06 · Why this answer works
No. The assistant needs current weather evidence. If that evidence is missing, it should say the answer is unknown.
What appears on screen
No. The assistant needs current weather evidence. If that evidence is missing, it should say the answer is unknown.
1:17 · Check 1
Find the source of current facts.
What appears on screen
Find the source of current facts.
1:22 · Check 2
Check which tools can act.
What appears on screen
Check which tools can act.
1:27 · Check 3
Keep unknown facts marked as unknown.
What appears on screen
Keep unknown facts marked as unknown.
1:32 · Now try a different case
Open the lesson and try its different practice case. Use every check in the full task. Watching alone does not mark the lesson complete.
What appears on screen
Open the lesson Try the new case Check and revise The full task and its requirements are in the linked lesson.
Watching explains the example. Complete the lesson practice to check what you can do. Explore all visual lessons →
Before you start
Prerequisite: O2 permissions gate and O3 synthetic workspace. Offline/no-cost route: use paper, a notes editor and the invented facts below. Tools are optional and must already be authorized. No accounts, paid API calls, private workplace data or external actions are required.
What you will be able to do
- Distinguish model, product, retrieval, memory, search, tools, and agents.
- Explain training and inference in plain language.
- Identify capabilities dependent on external data or tools.
Part 1 Map the whole system
Training changes model parameters; inference runs the model on selected context. A product provides account/UI features. A harness assembles instructions, context, tool calls, approvals, state and retries. The execution environment controls where operations run and what files/networks are accessible. Location, weight availability/license rights and frontier/specialized capability are independent dimensions. Publicly downloadable weights can still have proprietary restrictions; inspect the actual license. A laptop app may use a cloud model and local tools. Text generation drafts; reasoning supports harder computation; multimodal capabilities vary by accepted and generated modality; embeddings retrieve similar candidates; reranking orders candidates. Check actual supported features and permissions rather than selecting a universal best model.
Part 2 Worked synthetic example
A fictional travel assistant answers “Will it rain tomorrow?” The model formats prose; a weather tool supplies current data with location/time; the harness chooses the tool and blocks booking; product memory may supply a reviewed preference; logs retain their own record. If no weather source is available, the answer is unresolved. A training cutoff is not current weather evidence.
Part 3 Guided practice and feedback
Label five actions: create prose, fetch weather, reload a prior preference, request a booking, record the tool event. Answers: generation, retrieval/tool, product or harness persistence, proposed external action requiring scope, logging. Conversation history is saved turns; memory is selected persisted facts; retrieval fetches external evidence; compaction compresses prior context and can lose detail.
Review and next step
Self-review records practice; it is not independent assessment, certification or operational approval. Continue to F2 to apply the permission map before a prompt.
Substantive review: 2026-10-02. See the configuration guide. The NIST AI RMF provides voluntary risk-management guidance, not authorization for this exercise.