Skip to content
NTPU AI Learn

AI 301 Practice

Seven short activities that build the judgment behind a good Multi-Agent Architecture 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.

One Agent or Many?

Decide the simplest architecture that fits each simulated use case: a Single Agent, an Agent + Tool (one agent that calls a tool), an Agent + Human (one agent whose result or decision a person owns) or a real Multi-Agent design. Choose Multi-Agent only when there is a clear reason such as different expertise, different permissions, parallel work, independent verification or organizational separation.

0 of 8 correct

Item 1

Summarize one simulated meeting transcript in a fixed format for the team.

Item 2

Convert a simulated invoice amount into another currency using an official exchange-rate service.

Item 3

Draft a reply to a customer complaint that may include a refund. A person must decide whether the refund is granted.

Item 4

A simulated report needs a researcher who reads the public web, an analyst who uses internal finance data that must never reach the researcher, and an independent verifier.

Item 5

Rewrite a paragraph in a friendlier tone.

Item 6

Check the stock level of an item in a simulated inventory system and report it.

Item 7

Decide whether to uphold a simulated appeal from a named employee. The agent may prepare the evidence.

Item 8

Three regional analyses of the same simulated dataset must run at the same time, and each result is checked by a separate reviewer.

Design the Agent Team

A simulated grant-proposal team has four roles: Researcher, Writer, Verifier and Coordinator. Assign each responsibility to the role that should own it. If a line describes a role doing something another role already owns, mark it as Duplicated: it is an overlap to remove.

0 of 10 correct

Item 1

Find candidate sources and list them with links.

Item 2

Write the proposal sections using only the evidence that was supplied.

Item 3

Check each cited fact against its source and report every mismatch.

Item 4

Assign tasks, track status and assemble the final package for the human owner.

Item 5

The Writer also runs its own searches for extra sources while drafting.

Item 6

The Verifier also rewrites the weak paragraphs in the draft.

Item 7

Report which claims have no supporting source.

Item 8

Decide which sections are drafted first and which can be worked on in parallel.

Item 9

The Researcher also drafts the executive summary.

Item 10

Adapt tone and structure to the funder's template.

Delegate or Keep?

A simulated supervisor agent is producing a research report. For each task decide whether it should Keep it (do it itself), Delegate it to another agent, Parallelize it across several agents, or Escalate it to a human.

0 of 8 correct

Item 1

Fix a typo in the section the supervisor just wrote.

Item 2

Extract the statistics from a PDF. Only the Data agent has the PDF reader tool.

Item 3

Summarize five independent reports that do not depend on each other, and time is short.

Item 4

Approve an exception that publishes embargoed figures early.

Item 5

Reformat a table that is already in the supervisor's own context.

Item 6

Verify twelve citations, each of which can be checked independently.

Item 7

Check whether the draft follows legal-compliance rules that only the Compliance agent has been given.

Item 8

Send the finished report to the external client.

Agent Communication Contract

A message contract says what each kind of message must carry. Read each simulated agent message and identify its message type: TASK_REQUEST, TASK_RESULT, STATUS_UPDATE, EVIDENCE, QUESTION, ERROR, ESCALATION or HANDOFF.

0 of 8 correct

Item 1

"Please compile three sources on topic X by Friday. Constraint: public sources only. Return a list with links."

Item 2

"Done: three sources compiled. Status: complete. Open issue: source 2 is behind a paywall."

Item 3

"In progress: two of three sources found, no blockers."

Item 4

"Claim: turnout rose 4%. Source: official register, table 3, retrieved today."

Item 5

"Do you mean the 2023 or the 2024 dataset? I cannot start without knowing."

Item 6

"Cannot open the register: access denied. No result produced. Retrying will not help."

Item 7

"The two sources disagree on the total and policy does not say which to prefer. A person must decide."

Item 8

"Transferring ownership of section 2 to the Writer. Attached: task, objective, context files, constraints, expected output and open issues."

Shared or Private?

A simulated team of agents works on one task. Classify each piece of information as Shared context (all agents see it), Private context (only one agent needs it), Shared knowledge (a common reference source), Human-only (a person keeps it or decides it) or Restricted (only a specific agent or tool may access it).

0 of 10 correct

Item 1

The shared goal of the task and its deadline.

Item 2

An agent's draft working notes and discarded ideas.

Item 3

The approved style guide and glossary that all agents consult.

Item 4

The final decision on a staff member's performance rating.

Item 5

Real names and student numbers of learners, needed only by the Intake agent.

Item 6

Which subtasks are done, in progress or blocked.

Item 7

The Analysis agent's intermediate calculation scratchpad.

Item 8

The official policy handbook kept as a reference.

Item 9

Credentials and access keys used by one tool.

Item 10

Whether to sign a contract with a simulated vendor.

Choose the Orchestration Pattern

Choose the pattern that best fits each simulated scenario: Supervisor / Worker (a central agent assigns and aggregates), Planner / Executor (one plans, others execute), Specialist Team (a fixed sequence of specialists), Peer-to-Peer (equals talk directly), Hierarchical (leads over sub-teams) or Hybrid (central control plus bounded peer collaboration).

0 of 8 correct

Item 1

A central agent splits a customer-feedback analysis into five parts, hands them to workers and merges the results.

Item 2

One agent writes a plan for a data migration, other agents carry out the steps and report back, and the planner checks the results.

Item 3

Research, then analysis, then verification, then reporting, each by a different specialist, always in that order.

Item 4

Three brainstorming agents of equal standing exchange ideas directly, with no coordinator, on a low-risk creative task.

Item 5

A program lead agent directs three department leads, each of whom manages its own specialists. Conflicts go upward.

Item 6

A supervisor controls the overall process and the permissions, but two writing agents may negotiate wording directly within a limit.

Item 7

An audited review where one central agent must log every assignment and control all permissions.

Item 8

A large simulated organization with many teams and hundreds of tasks, where a single supervisor would be overloaded.

Resolve the Multi-Agent Conflict

Two or more simulated agents disagree or overlap. Follow the resolution flow: Apply a rule if a policy resolves it, Compare evidence if the evidence can settle it, let an Evaluator decide if neither does and the impact is low, Assign one owner when the conflict is duplicated work, or Escalate to a human when the decision is high impact or cannot be settled.

0 of 8 correct

Item 1

Two agents disagree on a figure. Company policy says an audited report always takes priority over an estimate.

Item 2

Two agents give different totals. No policy applies, but one shows its calculation from the source table and the other shows nothing.

Item 3

Two reviewer agents give contradictory recommendations on a low-impact task. The evidence is equal, no policy applies, and an evaluator agent holds the rubric.

Item 4

Two agents disagree on whether to reject a simulated applicant. The evidence is inconclusive and the outcome affects a person.

Item 5

Two agents are both drafting the same section and their versions differ.

Item 6

One agent wants to finish fast and another wants a full check. Policy says accuracy comes before speed for published figures.

Item 7

Two agents disagree about why sales dropped. One cites a dated system log and the other cites a rumor.

Item 8

An agent asks to spend above its budget and another agent objects. The amount is above the authority of any agent.

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?