PMGM7023 Chap.1 Managing an AI Analyst
Managing an AI Analyst
The job the course is preparing you for
The opening session sets out a progression that decides how the rest of the course is organised. Raw records become organised facts once context is attached to them, organised facts become an explanation once a pattern is identified, and an explanation becomes judgment once somebody decides what should happen next.
Each of those transitions can be handed to a machine except the last, and the lecture states the position directly: analysis informs a judgement and does not stand in for one.
That claim is an allocation of work rather than a reassurance. A capable system can move records into a clean table faster than you can describe the table, and it can describe patterns in that table fluently.
What it cannot do is supply context the records never contained, or accept the consequences of acting on a pattern.
Two forms of help that arrive looking identical
Ask a conversational assistant why complaints rose and you receive an explanation of what could be investigated: segment by product, check seasonality, look at policy changes.
Nothing was inspected, so the answer is cheap, reversible and impossible to check by reading it.
Ask a supervised agent the same question, having granted it permission, and it inspects the actual files, writes and runs the analysis, checks the result, revises it, and leaves behind cleaned data, logs, diagnostic views and a written conclusion.
The second form is more expensive and more consequential, and that is exactly its value.
Because files were touched, the files are evidence, and any claim in the summary can be traced to a step that either happened or did not. The lecture summarises the difference as advice against artifacts.
Permission is the first analytical decision
An agent can change files and run tools only because someone allowed it to, so the managerial work moves to the front of the task.
You specify the objective, the boundaries, the checkpoints at which you want to see something, and the evidence the finished work must carry. A vague instruction does not produce a vague answer; it produces a confident and complete set of artifacts that may be wrong, which is considerably harder to notice.
This is why the tutorials begin with a trusted folder and a permission mode rather than with any analysis.
A session started in a folder you chose can only read and change what is inside it, and anything outside requires an explicit approval you will see before it happens.
Reading the session shape as an argument
A typical session runs about two hours of lecture followed by about one hour of tutorial: the conceptual framework first, then the same idea executed at a terminal.
Learning the command first produces a procedure that expires with the tool. Learning the question first turns the command into a detail you can look up, and the reasoning transfers to whatever replaces the current interface. The final two sessions drop the tool work entirely and assess only the reasoning, which is a fair preview of where the marks sit.
What this chapter covers
- 01
Records, facts, explanation and judgment as four distinct levels
- 02
A conversation returns advice; a supervised agent returns inspectable artifacts
- 03
Capability is granted by permission, so boundaries precede analysis
- 04
Deciding which form of help a task needs, and saying why
Choosing the form of help for four tasks
- 3Test whether the answer depends on a file somebody can open.
- 2Test whether anyone will need to rerun the work.
- 3Test whether the deliverable is your own wording or judgment.
Key terms
- AI Agent
- An AI agent is a system that selects and uses tools across several steps to pursue an objective. An assistant that answers one request and stops is not one.
- Artifact
- An artifact is a file the work leaves behind, such as a cleaned dataset, a log or a chart. Artifacts are what make a claim traceable to the step that produced it.
- Permission Mode
- Permission mode is the setting that decides whether each action is approved individually, approved for a session, or allowed automatically.
- Supervised Loop
- A supervised loop is an arrangement in which the system executes while a person sets the objective, reviews the evidence and owns the decision.
Managing an AI Analyst FAQ
What is the difference between AI chat and an AI agent?
A chat returns an explanation of what could be done, and the person carries out the work; the conversation is the whole interface. An agent, once granted permission, inspects files, writes and runs code, checks results, revises and produces deliverables. The lecture frames the difference as advice against artifacts: one leaves you with a suggestion, the other with cleaned data, logs and an owned conclusion.
Why does this course start with permission rather than with analysis?
Because capability only exists where it has been granted. An agent that can change files can do so because a boundary was set, so the objective, the limits, the checkpoints and the required evidence are decided before any work begins. A loose instruction produces confident and complete output that may be wrong, and that is much harder to detect than an obviously vague answer.
Exam move
Before each session, write one sentence naming the decision the session is about. Afterwards, list the two tasks from your own work that fit each form of help and say why. Practise stating a boundary aloud before starting any tutorial exercise, since the same sentence structure reappears in the project specification.
Working through Managing an AI Analyst in PMGM7023? Sia is AskSia’s AI Management tutor — ask any PMGM7023 Managing an AI Analyst question and get a clear, step-by-step explanation grounded in how PMGM7023 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
