Give an agent a job you can check.
Create a one-page Topic 2 study guide from the instructor’s page. Write a specific brief, run it, then use the source to ask for one precise revision.
Work in pairs: one person directs the agent; the other checks its sources. Switch roles for the revision. Working alone? Do both passes.
- 1. Write the brief3 minutes
- 2. Talk to the agent5 minutes
- 3. Check and revise7 minutes
Your source and your finished page
Use this instructor page only
Topic 2 Classroom Deck — Agent Foundations and Human Review. Use the student-facing material in these five sections, including their lesson reveals:
- Screen 3: the LLM’s answer
#llm-alone— what was and was not checked - Screen 4: the agent’s moves
#llm-agent— observation, proposal, permission - Screen 6: tool and skill
#tool-skill— an action versus a procedure - Screen 8: look, decide, repeat
#loop-write— the loop and its stop rule - Screen 10: the four-part brief
#brief-why— outcome, evidence, limits, stop
Give the agent the full page URL, not just “my instructor’s page.” Citation format: Screen 6, #tool-skill, with a link. Do not follow other sites linked from the deck.
“Done” means you can check all five
- Scope: title it “Topic 2 study guide — agent foundations so far.” This covers today’s concepts, not all Topic 2 material.
- Content: explain LLM, agent, tool and skill in plain language using the course’s working model.
- Application: include one look → decide → repeat example, its stop rule, and one action requiring human permission.
- Retrieval: add two self-check questions with short answers. Cite the supporting screen/anchor for each definition, example and answer; flag unsupported details.
- Format: one readable Letter or A4 page, including sources. Aim for at most 400 words; confirm the actual page count in print preview.
1. Write directly to the agent 3 minutes
Use commands such as “Create…,” “Use only…,” and “Stop when….” Complete all four parts in your own words. Your partner should be able to name the result, source, permissions and stopping point without guessing.
Complete all four parts first. This checks completeness, not quality.
One possible brief — compare after writing yours
Outcome: Create a one-page study guide for classmates on Topic 2 agent foundations so far. In at most 400 words, explain LLM, agent, tool and skill; give one loop example with a stop rule and a human permission boundary; add two questions with answers.
Evidence: Use only https://monmouthaiteaching.com/learnai/week-02-agent-foundations.html, student-facing Screens 3, 4, 6, 8 and 10 (#llm-alone, #llm-agent, #tool-skill, #loop-write, #brief-why). Cite screen numbers and anchor links for definitions, examples and answers. Flag gaps instead of adding outside facts.
Tools + limits: You may read that public page or the relevant public text I supply. Return the draft in this conversation. Do not follow other links, read unrelated files, write files, email or publish.
Stop: First confirm the source you could access. If you cannot read it, ask me for the public text and wait. After the draft, source list and uncertainties, stop for my review.
This worksheet has no submit action or intentional storage. Copy your work before leaving or reloading. It does not contact an AI. What you paste into an external assistant is handled by that service; a local interface does not imply local-only processing. Never paste grades, credentials, private files or peer work.
2. Start a fresh conversation and send it 5 minutes
- Open your already-working, instructor-approved agent. Start a new task/conversation so an old task does not supply hidden assumptions. For a command-line agent, launch it in a new, empty practice folder. No installation or purchase is needed for this exercise.
- Paste your brief. Grant only read access to the named public source. If it requests unrelated access or actions, stop and ask the TA. Your brief is an instruction; it does not replace the tool’s permission controls.
- Check its source access before trusting the draft. Ask: “Which of the five sections could you actually read? Show the section names and flag any missing text.” Inspect the read/browse result if your tool exposes one. A claim to have read a page is not itself proof.
- If the page cannot be read: open the source yourself and paste only the relevant public lesson text, with screen/anchor labels. Tell it to use only that text. Record this as supplied-context practice; it does not demonstrate a tool read by the agent.
Timebox: if the run stalls or access is blocked, use the sample below. Spend the remaining time checking evidence, not fixing an account.
No account or device? Use this simulated draft excerpt.
Teaching simulation — not a recorded live run. Use your written brief and the instructor page (or the printed source excerpts from the TA). A partner with access may also run your brief while you direct and check.
Simulated first draft excerpt
“An agent is a model that can use observations to choose a next move. Once it finds a broken link, it may fix and publish the page without waiting for a person. A tool is a reusable set of instructions; a skill is one action that returns a result.”
Draft cites: Screen 4, #llm-agent; Screen 6, #tool-skill.
Check two claims in this excerpt. Write the specific follow-up you would send, then hand-revise the excerpt using the source. Work on paper if needed. Label your result simulated practice; do not claim an agent ran or that you produced a complete live guide. Finish the one-page guide when access is available.
3. Check two claims. Ask for one revision. 7 minutes
Switch roles. Pick two claims from the draft and locate the supporting screen yourself. Do not ask the same assistant to be the only judge of its own answer.
Send a specific follow-up
Replace the brackets: “In [section], you wrote [claim]. Screen [number/anchor] says [source evidence]. Revise [specific change]. Keep the original source and permission limits, keep it to one page, and tell me what changed.”
If both claims are supported, request one concrete improvement to the audience, example, question or layout. Explain why. “Make it better” is not a specific instruction.
Before you call the guide ready
Over one page? Ask the agent to shorten repetition while keeping the required ideas and citations. Recheck the preview; do not shrink the text until it becomes unreadable. This worksheet is separate from your one-page guide.
Keep for yourself: your brief, revised guide, two checks and revision note. This is in-class practice, not a new graded submission. Nothing here uploads a chat log. On the simulated route, keep your brief, corrected excerpt, two checks and proposed follow-up instead.
Tell your partner: “My instruction changed ___, and I checked it by ___.”
Return to Screen 12: what does the agent’s report actually prove?