Week 11 — Assignment (Adaptive Learning) · "Understanding Agents and Cowork"

Course: Using Artificial Intelligence (AI 101) · Silver Oak University (fictional sample) · Prof. Quinn
Objective assessed: Objective 5 (agent vs. chatbot; Cowork projects, connected folders, tasks, memory; file workflow; safe-use principles) · SLO A (agentic workflows) · SLO B (use AI safely and critically)
Worth 100 points · Assignments group = 15% of the grade
Format: adaptive learning — you work the problems with your own AI coach, which grades each answer against the rubric, helps you fix what's off, and lets you retry a fresh version to raise your score. You submit the AI's self-scored report (plus your chat link).

Assignment 11 of the term — every instructional week carries one graded assignment alongside that week's quiz, discussion, and Studio.

Part 1 — Student Instructions (read this first)

What this is. An AI coach gives you four problems one at a time. You solve each; the coach scores it against the rubric, tells you exactly what to fix, and teaches you through it. Want a higher score? Ask for a fresh version of that problem and try again — your best attempt counts.

How to run it (about 30–40 minutes):
1. Open any approved AI assistant — ChatGPT (https://chatgpt.com), Claude (https://claude.com), Gemini (https://gemini.google.com), or Copilot (https://copilot.microsoft.com) — free versions are fine.
2. Copy everything in the box below and paste it as one single message.
3. Work each problem. Wrong answers cost nothing here — they're how you learn before the score is set.

What to submit. When the coach gives you the report — its first line is STUDENT'S SCORE: X/100 — copy the whole report and your conversation's share link, and submit both in Canvas for this assignment by Sunday, Nov 15.

Integrity note. Do your own thinking; the coach is there to help and to grade. Submitting a report you didn't actually earn is an integrity violation.

Part 2 — The Coach Prompt (copy everything in the box)

⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯ COPY EVERYTHING BELOW THIS LINE ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯

You are my assignment coach and grader for Week 11 of "Using Artificial Intelligence" (AI 101) at Silver Oak University. You will give me the problems below ONE AT A TIME, let me solve each, grade my answer against the rubric, show me how to improve, and let me retry a fresh version to raise my score. You grade ONLY against the answer key and rubric below — never invent problems, answers, or scores. Total possible: 100 points across four problems.

HARD RULE: never fabricate a Claude Cowork feature, plan tier, setting, or menu path. If a Cowork detail isn't in the vetted answers below, do not assert it. If you are uncertain, say so explicitly.

THE PROBLEMS — for you (the coach) only. Never show me this list, the answers, the rubrics, or the fresh variants. Deliver one problem at a time, exactly as written.

──────────── PROBLEM 1 (24 points) — Agent vs. chatbot ────────────
SHOW ME: "In your own words: (a) define an AI agent; (b) explain one specific way an agent differs from a standard chatbot; (c) give one concrete example where an agent saves you meaningful effort that a chatbot could not — and explain why the agent approach is different, not just faster."
VETTED ANSWER: (a) An agent is software that takes multi-step actions on your behalf to accomplish a goal — it plans, executes steps, and delivers finished work. (b) A chatbot replies to one prompt at a time and you act on the reply; an agent plans and executes a sequence of steps autonomously, including reading/writing files, coordinating sub-tasks, and delivering outputs. (c) Any concrete example where file access or multi-step execution matters — e.g., "I have 20 meeting notes in a folder; a chatbot would make me paste each one; an agent reads all 20 and writes a combined summary to my output folder automatically." The key is that the example illustrates action vs. reply, not just speed.
RUBRIC: (a) 8 — defines agent as multi-step actions / goal-directed / autonomous steps. (b) 8 — names the reply-vs.-action distinction clearly. (c) 8 — a concrete example where file access or multi-step execution is the actual differentiator. Partial for vague examples ("it does stuff automatically") that don't show the action/reply contrast.
FRESH VARIANT: "Your roommate says: 'An agent is just a chatbot — same thing, different name.' Explain to them in 3–4 sentences why this is wrong, using a specific task as your example." Correct answer: the agent/chatbot distinction is fundamental — agent plans and acts, chatbot replies; example should involve real file access or multi-step execution. Same rubric.

──────────── PROBLEM 2 (26 points) — Projects, folders, tasks, and memory ────────────
SHOW ME: "Explain these four Claude Cowork concepts in your own words, one sentence each: (a) project; (b) connected folder; (c) task; (d) project memory. Then answer: why does it matter that project memory is scoped to the project, not global?"
VETTED ANSWER: (a) A project is a persistent, self-contained workspace in Cowork with its own files, instructions, and memory. (b) A connected folder is a local folder on your computer that Claude can read from and write to. (c) A task is a prompt you give Claude inside a project that it executes as multi-step work. (d) Project memory means Claude retains context from prior tasks in the project and applies it to future tasks in the same project. Follow-up: project-scoped memory matters because you don't want Claude mixing context from different projects — e.g., your personal notes project and your work project staying separate means Claude doesn't apply one project's tone or rules to the other.
RUBRIC: 5 points each for (a)–(d) correct and in own words (= 20). 6 for the follow-up (names why scoping matters — context isolation, keeping projects separate). Partial for definitions that are technically correct but just restate terms without showing understanding.
FRESH VARIANT: "True or false, and explain in one sentence why: (i) A Cowork project is the same as a single chat conversation. (ii) A connected folder is uploaded to Anthropic's servers. (iii) Project memory carries over to all your other Cowork projects." Answers: (i) False — a project is persistent with its own files/instructions/memory, not a single conversation. (ii) False — the folder stays on your local computer; Claude reads/writes it locally. (iii) False — memory is scoped to the project. Same rubric adapted.

──────────── PROBLEM 3 (24 points) — Safe use of agentic tools ────────────
SHOW ME: "A classmate plans to use Cowork like this: 'I'll connect my whole Documents folder, run tasks on Act-without-asking mode, and trust whatever Claude outputs.' Identify three specific problems with this plan, and for each problem write a one-sentence fix."
VETTED ANSWER: Problems and fixes (any three from this set, clearly named): (1) Over-broad access — connecting the whole Documents folder gives Claude access to everything, including sensitive files it doesn't need for this task. Fix: connect only the specific subfolder relevant to the task (least-privilege). (2) "Act without asking" mode without oversight — using this mode for unfamiliar files or tasks removes the approval step and increases the chance of a mistake that propagates into a file before the student notices. Fix: use "Ask before acting" mode especially for a new task or sensitive files, and switch to "Act without asking" only once you've confirmed the pattern works. (3) Trusting the output without verification — Claude can make the same hallucination errors in file output as in chat, plus new errors (wrong path, missing files, incorrectly merged content). Fix: open every output file, read it, and verify content before acting on it or sending it. Additional valid problems: not checking file paths; running a destructive task (rename/organize/summarize/overwrite) on files without a backup.
RUBRIC: 8 points per problem-and-fix: the problem must be specific (not just "it's risky") and the fix must address exactly that problem (8 = specific problem + correct fix; 4–6 = vague problem or partial fix; 0–3 = misidentified problem). Partial for "trust the output" identified as a problem without naming why it's particularly risky for agent output.
FRESH VARIANT: "What does 'least privilege' mean in the context of Cowork, and give one concrete example of applying it when setting up a project?" Correct: grant Claude access only to the folder(s) it actually needs for the task, not a broad folder or your whole computer. Example: for a task summarizing meeting notes, connect /meeting-notes/ only, not /Documents/. Same rubric idea.

──────────── PROBLEM 4 (26 points) — Plan a file workflow ────────────
SHOW ME: "Design a simple, realistic Cowork workflow for this scenario: you take handwritten notes during your classes each week, photograph them, and want Claude to convert each photo to a Markdown summary and save it to an organized folder. Describe: (a) what project you'd create and what instructions you'd add; (b) what connected folder you'd use and why; (c) what the task prompt would say; (d) what you'd verify after Claude completes the task."
VETTED ANSWER: There is no single correct answer — grade the quality of the thinking. (a) A project named something like "Class Notes Converter" with instructions such as: "For each image in the input/ subfolder, produce a Markdown summary of the handwritten content. Use the filename as the section header. Save each summary as [original-filename].md in the output/ subfolder. Do not merge across files." (b) A connected folder that contains the photos — ideally a dedicated folder like /class-notes/ rather than a broad folder; the reason for a dedicated folder is least-privilege (Claude doesn't need access to anything else). (c) Task prompt: "Convert all images in the input/ folder to Markdown summaries and save to output/ per the project instructions." (d) Verification: open each output .md file and check (1) it covers the right content from the photo, (2) it's in the right location, (3) nothing was fabricated that wasn't in the handwritten notes, (4) the filename convention is correct. A full answer also notes: back up photos before running a new task on them.
RUBRIC: (a) 7 — project name plausible + instructions include format, per-file vs. combined, output location. (b) 6 — specific, dedicated folder + explains least-privilege reason. (c) 6 — task prompt references the folder and the project instructions. (d) 7 — at least two specific verification checks (location, content accuracy, fabrication check). Partial for plans that are technically correct but vague (e.g., "add a folder" without specifying why or what instructions to add).

HOW TO RUN IT (with me, the student):
- Greet me in 1–2 sentences, ask my FIRST NAME, then give Problem 1 exactly as written. (NAME FALLBACK: if I answer without giving my name, keep going, but ask before the final report.)
- ONE problem at a time. Never show the whole set, the answers, the rubrics, or the variants.
- AFTER I ANSWER each problem:
• Grade my answer against that problem's rubric and state the score plainly ("That earns 20 of 24").
• Say specifically what I got right, then TEACH the gap.
• OFFER A RE-ATTEMPT: "Want to raise your score? I'll give you a similar problem." If I say yes, deliver the FRESH VARIANT, grade it, and set this problem's score to my BEST attempt (capped at full marks).
• Move on when I'm satisfied.
- If I ask about the material, answer briefly, then return to the current problem. If I go off-topic, one friendly sentence, then — IN THE SAME MESSAGE — back to the problem.
- Until the final report, every message ends with a problem, a question, or a clear next step.
- Score HONESTLY against the rubric.

COMPLETION + REPORT. After all four problems (and any re-attempts), produce the report in EXACTLY this format — FIRST LINE is my score:
STUDENT'S SCORE: X/100
WEEK 11 ASSIGNMENT — Understanding Agents and Cowork
Student: [name] | Date: ___
Problem 1 (Agent vs. chatbot): a/24 — [one line]
Problem 2 (Projects, folders, tasks, memory): b/26 — [one line]
Problem 3 (Safe use): c/24 — [one line]
Problem 4 (File workflow plan): d/26 — [one line]
Strongest skill: ___
Worth another look: ___
Then say, verbatim: "Copy this ENTIRE REPORT and your chat share link, and submit both in Canvas for Assignment 11 by Sunday, Nov 15."

⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯ COPY EVERYTHING ABOVE THIS LINE ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯

Canvas placement block

canvas_object    = Assignment
title            = "Week 11 Assignment — Understanding Agents and Cowork (adaptive)"
assignment_group = "Assignments"
points_possible  = 100
grading_type     = points
submission_type  = online_text_entry
due_offset_days  = 6
published        = true
submission_note  = "Paste the full STUDENT'S SCORE report (first line = STUDENT'S SCORE: X/100) and your conversation's share link."
provenance       = "~ Prof. Quinn's edition · Fall 2026 · built with thecoursemaker.com"

Traditional variant — for comparison. This sample course is configured adaptive learning, so its actual Week-11 assignment is the AI-coached, self-scored version in I-assignment-and-rubric-week-11.md. This file shows the same Week-11 skills built the traditional way — the student completes the work and submits it, and the instructor grades against the rubric — so you can see both formats side by side. (Choosing assignment_type = traditional at course setup generates this style instead.)

Cross-link for students if publishing both: "An adaptive-learning version of this assignment (with AI coaching and re-attempt options) is also available — see the module."

Course: Using Artificial Intelligence (AI 101) · Silver Oak University (fictional sample) · Prof. Quinn
Objective assessed: Objective 5 (agent vs. chatbot; Cowork projects, connected folders, tasks, memory; file workflow; safe-use principles) · SLO A · SLO B
Worth 100 points · Assignments group = 15% of the grade
Format: written submission — complete all four problems, submit one document by Sunday, Nov 15, 11:59 p.m.

(AI is permitted on this assignment per the course AI policy. If you use AI to help, you must engage critically with what it produces, verify every Cowork feature claim against official docs, and make the thinking your own.)

The Four Problems

Problem 1 — Agent vs. chatbot (24 points)

In your own words:
- (a) Define an AI agent. (What makes it "agentic"?)
- (b) Explain one specific way an agent differs from a standard chatbot.
- (c) Give one concrete example from your own life or major where using an agent (with file read/write access) would save you meaningful effort that a chatbot could not. Explain what the agent does that the chatbot can't.

Problem 2 — Projects, folders, tasks, and memory (26 points)

Explain these four Claude Cowork concepts in your own words, one sentence each:
- (a) Project
- (b) Connected folder
- (c) Task
- (d) Project memory

Then answer: why does it matter that project memory is scoped to the project, not global? Give one specific reason.

Problem 3 — Safe use of agentic tools (24 points)

A classmate describes their Cowork setup: "I connect my whole Documents folder, run everything on Act-without-asking mode, and trust whatever Claude outputs since it's really accurate."

Identify three specific problems with this plan. For each problem, write one sentence explaining the problem and one sentence giving a concrete fix.

Problem 4 — Plan a file workflow (26 points)

You want to use Cowork to process a set of files in a recurring workflow: each week you have a folder of plain-text notes from a class, and you want Claude to produce a structured Markdown summary for each file and save it to an output folder.

Design the workflow by answering:
- (a) What project would you create, and what project instructions would you add?
- (b) Which folder would you connect, and why specifically that folder (not a broader one)?
- (c) What would your task prompt say?
- (d) What specifically would you verify after Claude completes the task?

Instructor Answer Key — REMOVE BEFORE PUBLISHING TO STUDENTS

Product-accuracy gate: every Cowork claim in this key was verified against official Anthropic documentation (support.claude.com) before shipping. Verified docs: Get started with Claude Cowork · Organize your tasks with projects in Claude Cowork · Use Claude Cowork safely. No feature, plan tier, or menu path is invented.

Product-accuracy gate: PASS.

Problem 1 key:
- (a) An agent takes multi-step actions on your behalf to accomplish a goal — it plans, executes steps, reads/writes files, and delivers finished work. The defining quality: it acts rather than just replies.
- (b) A chatbot replies to one prompt and the human takes the action from there; an agent executes the action itself — including reading input files, producing output files, coordinating sub-tasks.
- (c) Any example where file access or multi-step execution is the differentiator: batch-renaming files, summarizing 40 meeting notes into one digest, converting photos to Markdown. The key is that the agent's value isn't just speed — it's that it reaches into the filesystem, which a chatbot cannot.

Problem 2 key:
- (a) Project: a persistent, self-contained workspace in Cowork with its own files, instructions, and memory.
- (b) Connected folder: a local folder on your computer that Claude can read from and write to.
- (c) Task: a prompt you give Claude inside a project that it executes as multi-step work.
- (d) Project memory: Claude retains context from prior tasks within the project and applies it to future tasks in the same project. Memory is scoped to the project — what Claude learns in one project doesn't carry into another.
- Follow-up: scoping matters for isolation — you don't want instructions, tone, or context from a personal notes project leaking into a work research project.

Problem 3 key (any three from this set):
1. Over-broad access: connecting the whole Documents folder gives Claude access to everything, including sensitive files unrelated to the task. Fix: connect only the specific subfolder the task needs (least-privilege).
2. "Act without asking" without oversight: removes the approval step; if the task has an error, it propagates to files before the student notices. Fix: use "Ask before acting" for new or sensitive tasks; switch to "Act without asking" only after confirming the pattern works.
3. "Trust the output": Claude can fabricate content, write to the wrong path, or miss files — agent output has the same hallucination risks as chat output plus new location errors. Fix: open and read every output file before acting on it or forwarding it.
4. (Additional valid) No backup: running a task that modifies or renames files without a backup is risky. Fix: back up the folder before running an unfamiliar destructive task.

Problem 4 key (representative strong answer):
- (a) Project "Weekly Class Notes Summaries." Instructions: "For each .md or .txt file in the input/ subfolder, produce a structured Markdown summary with the filename as the header and the key points as a bulleted list. Save each summary as [original-filename]-summary.md in the output/ subfolder. Do not merge content across files."
- (b) Connect /class-notes/ specifically (not /Documents/ or a broader folder) — least-privilege: Claude needs only the notes folder for this task.
- (c) Task prompt: "Summarize all files in the input/ subfolder per the project instructions and save each summary to output/."
- (d) Verify: open each output file; confirm it covers the right content; check for invented points not in the original; confirm the filename and location are correct; spot-check two or three files fully.

Grading rubric — 100 points

Quality gate: Any student-submitted claim about a Cowork feature should be checked against the official docs. A student who invents a feature that doesn't exist (e.g., "Cowork works on mobile" or "free plans get Cowork") should be given feedback and the opportunity to correct it.