Distinguish an LLM answer, a tool request, and an agent loop.
Week 02 · Agent foundations
From Answer to Action Agent loops + human review
Follow one fictional broken-link case to see what an agent observes, what it may request, what people must review, and how work becomes inspectable.
Before Meeting 1
Set up Claude Code before Tuesday. Meeting 1 opens by asking you to run claude --version. If you cannot install it or sign in, you are not behind: bring the exact error text and take the no-install route, which earns the same credit. Bring a browser and something for short notes; both meetings use fictional files and synthetic traces.
By the end
Four distinctions make agent work reviewable.
The goal is not to memorize product vocabulary. It is to recognize the loop, set authority, and ask for evidence.
Label input, context, model proposals, tool observations, and human review.
Give an agent the least authority sufficient for a task.
Use plans, diffs, tests, and human decisions to make changes inspectable.
Class materials
Jump back to any activity.
The classroom deck contains the full prompts, staged reveals, discussion protocols, rationales, and final study states needed to revisit each concept.
Week 2 homework
Prepare one coding agent—or document the blocker safely.
Complete the Agent Setup Readiness task before Week 3. This is a completion check: readiness, access diagnosis, and safe boundaries matter more than whether one product installs successfully.
One synthetic file. One read-only prompt. Five evidence items.
Choose Claude Code or Codex only if your current device and account support it. Work through the Set Up Claude Code workshop before Tuesday’s class—it covers installing, verifying, and what to do if you cannot get access. Then launch the tool in a disposable folder and ask it to inspect the supplied synthetic file without editing or acting.
A · INSTALLEDB · BLOCKED + TRANSCRIPTC · SUPERVISED LAB
Open Agent Setup Readiness→Optional depth
Keep advanced agent topics outside the required route.
These resources extend the course model. They are not required before Meeting 2 and do not replace the classroom activities.
Optional · not assessed
Choose a resource only if it supports your own goal. External sites may have different privacy, access, and account requirements.
View optional practice and advanced resources
GitHub + Agents review labLong-form standalone practice · optionalAI workflow skill mapReusable workflow patterns and when to use themPrintable workflow mapPDF version for offline reviewMultimodal and embodied agentsSurvey guide for agents beyond text ↗Agent mindset casesAdditional boundary and review scenariosAgentic AI backgroundExternal technical course · advanced ↗Agentic engineering primerOptional engineering perspective ↗
Next in the course
Preview Week 3→Week 3: AI Ethics and Responsible Use
Use principles, lived realities, and power structures to evaluate university AI policy.