Skip to content
NTPU AI Learn

AI 202 Practice

Six short activities that build the judgment behind a good Agent Planning & Reasoning Canvas

Info

Self-check (not scored, not stored). Every activity gives immediate feedback and explains why. If you miss, pick again; an activity is done when every item is right.

Workflow or Adaptive Agent?

Classify each scenario as a Fixed Workflow (every step predefined), a Conditional Workflow (rules choose the branch), an Agentic Workflow (a predefined frame with bounded agent decisions inside) or an Adaptive Agent (the next step depends on goal, state, observations and previous results).

0 of 8 correct

Item 1

Every Friday a script exports the week's timesheets, formats them in the standard template and saves the file to the team folder. Nothing ever changes.

Item 2

A simulated leave request under 3 days goes to the team lead; 3 days or more goes to the department head. A rule decides the path.

Item 3

A simulated support inbox: a fixed sequence receives, classifies and files each message. At one step an agent may choose which of three approved knowledge sources to search for a missing fact.

Item 4

You ask an agent to find out why a simulated supplier's delivery dates keep slipping. What it checks next depends on what it finds: contracts, shipping logs, emails or a call to a person.

Item 5

A form is converted to PDF, stamped with the date and archived. Every submission follows the same three steps.

Item 6

A simulated invoice is routed by amount and currency: small domestic invoices are auto-filed, large or foreign ones go to finance. AI only extracts the fields.

Item 7

An agent is asked to prepare a market briefing. It decides on its own which sources to search, whether the evidence is enough, when to ask a person and when to stop, within a step and cost budget.

Item 8

A report pipeline always runs collect, analyze, draft and review in that order. During drafting an agent may decide to run one extra check on a figure if a number looks inconsistent.

Goal, Subgoal or Action?

A simulated task: "Recommend one venue for a 40-person workshop next month, within budget." Sort each item into the part of the plan it belongs to: Goal, Subgoal, Action, Constraint or Success Criterion.

0 of 10 correct

Item 1

Recommend one venue for a 40-person workshop next month.

Item 2

Find out which venues are free on the workshop dates.

Item 3

Query the shared booking calendar for venue availability next month.

Item 4

Total cost must stay under the approved budget of 60,000.

Item 5

The recommendation lists at least two options with price, capacity and availability, each backed by a source.

Item 6

Confirm that each remaining venue holds at least 40 people.

Item 7

Do not book or pay for anything. A person makes the booking.

Item 8

Compare the two shortlisted venues on price and travel time.

Item 9

A person reads the recommendation and accepts it without needing to redo the research.

Item 10

Get one reviewed venue recommendation ready for the organizer to decide.

Trace a ReAct Loop

Read a simulated agent trace and classify each line as a Reasoning step, an Action, an Observation, a State update or a Next-step decision.

0 of 10 correct

Item 1

The customer asks why the order is late. I need the order status before I can answer.

Item 2

lookup_order(order_id = 4471)

Item 3

Result: status = shipped; carrier note = "weather delay".

Item 4

Known so far: order 4471 shipped, delayed by weather, promised date not yet compared.

Item 5

The delay may or may not break the promised date. I cannot tell the customer anything before I compare them.

Item 6

Next: check the promised delivery date in the order record; do not reply yet.

Item 7

get_promised_date(order_id = 4471)

Item 8

Result: promised date = 3 days ago; new estimate = tomorrow.

Item 9

Known so far: order is late by 3 days; estimated arrival tomorrow; cause is weather.

Item 10

Enough evidence to answer. Draft an honest reply with the new date; a person approves any compensation offer.

Search the Next Step

A simulated agent must find out why a monthly figure looks wrong and has five candidate actions. For each branch decide whether to keep it as promising or eliminate it, using cost, risk, reversibility and evidence quality.

0 of 8 correct

Item 1

Read the source spreadsheet and compare the disputed cell with last month. Cost: low. Risk: none, read-only. Reversible. Evidence: direct.

Item 2

Overwrite the disputed cell with the value the agent guesses is correct. Cost: low. Risk: high. Not reversible. Evidence: none.

Item 3

Ask the data owner which system produced the figure. Cost: medium (waits for a reply). Risk: low. Reversible. Evidence: strong.

Item 4

Rebuild the whole report from scratch with a new method. Cost: very high. Risk: medium. Evidence: none that the old method is wrong.

Item 5

Run a read-only comparison of the figure against the raw transaction export. Cost: medium. Risk: none. Evidence: strong, independent source.

Item 6

Send the corrected number to all department heads now to save time. Cost: low. Risk: high. Not reversible. Evidence: not yet checked.

Item 7

Search an unofficial web forum for similar problems. Cost: low. Risk: low. Evidence: weak, source quality unknown.

Item 8

Check the change log of the spreadsheet for edits made this month. Cost: low. Risk: none. Evidence: direct.

Verify or Reflect?

For each simulated situation choose the best move: Verify (is this result correct?), Evaluate (how much progress have we made?), Reflect (why did the strategy work or fail, and what should change?), Ask a human, or Stop.

0 of 8 correct

Item 1

The agent quotes a tax rate from a web page and wants to use it in a calculation.

Item 2

Three of five subgoals are finished, but the budget is half spent. Is the agent on track?

Item 3

The same search query has returned nothing useful three times. The agent keeps repeating it.

Item 4

The agent is asked to decide whether a policy exception should be granted to a named employee.

Item 5

The success criterion was met and the reviewer has accepted the summary.

Item 6

A generated table of figures looks very convincing, but nobody has compared it with the source.

Item 7

After a full run the agent notes that most of its time went to a step that added no new evidence. What should it do before the next run?

Item 8

The step limit is reached and the goal is still not met, and no new evidence has appeared for four steps.

Re-plan or Continue?

Inspect the new state in each simulated situation and decide whether the original plan is still valid. Choose Continue, Re-plan, Escalate to a human, or Stop.

0 of 8 correct

Item 1

The step finished normally and the result matches what the plan expected.

Item 2

The approved data source has been taken offline for maintenance, but a second approved source holds the same data.

Item 3

Midway the user says the report should now be for the board, not for the team, and it must be much shorter.

Item 4

New evidence suggests that the contract has a penalty clause. The next planned action would cancel the order.

Item 5

The plan has used its full cost budget, and two subgoals remain that would need much more.

Item 6

A human reviewer rejects the intermediate summary because it left out the main risk.

Item 7

The user gives two instructions that contradict each other, and the agent cannot satisfy both.

Item 8

A tool returned a slightly slower response than usual, but the result is complete and correct.

Sync my progressJoin a session to keep progress across devices and share it with your instructor

Info

After you join, your completed lesson numbers are sent to your instructor together with your learner code, so they can see your progress. Use the codes your instructor gave you, and do not enter your name, company secrets or other personal data. Progress is kept with the rest of that session's data.

What next?