Georgia State University — J. Mack Robinson College of Business PATH — Pathways for AI Training & Hiring CIS 4394 Agentic AI  ·  Fall 2026  ·  Dr. Xinyu Fu
Week 11 · Module 4 · Deployment & Applications

When the agent can spend.

A shopping agent that finds, compares and buys is the whole course in one transaction: tools, protocols, memory, evaluation, and, above all, the human gate. Every other week you could fix a bad run by running it again. This week the run moves money, and “mostly works” stops being an acceptable answer.

📌 This week's logistics:
  • Format this week: a short lecture on agentic commerce, a hands-on trace lab (page 03), then project time in class; I go desk to desk to unblock. Bring your laptop with your capstone running.
  • Before class, five minutes at home: create a free LangSmith API key, create an Observability agent named shop-agent, and install langsmith. The steps and the add-on download are at the top of 03 · Trace it.
  • Group Assignment 3 is due this week, on your own section's iCollege page (team submission plus peer evaluation).
  • Capstone Milestone II is due next week: Weeks 12 and 13 are Capstone Studios: build time in your own section, no new lecture material.
  • No quiz this week.
  • Two make-up quizzes are offered: one in Week 13 and one in Week 14. You may take both; the better of the two replaces your lowest score from Quizzes 1–4.
The big question
Discussion questionYour agent has worked all semester. Today you give it a payment method. Name the single property of the task that changed, and every design decision that must change with it.
Irreversibility. Everything else follows from it. A failed summary is retried; a failed place_order() is a return, a chargeback, a phone call, and sometimes a loss. So: retries need idempotency (a retried order must not become a second order); the loop needs a hard spend cap enforced in the harness, not requested in the prompt; the trace stops being a debugging convenience and becomes the evidence that answers “who authorized this?”; and the human gate moves from a nice-to-must the thing the system is designed around. Reliability that was “good enough for a draft” is now the difference between a product and a liability.
Interactive · the whole course in one purchase

One transaction, ten weeks

A small task (“restock the espresso pods for the break room, under $60, here by Friday”) touches nearly everything you have built since Week 1. Click each stage to see which week it came from.

Start here

Pick a stage

Each stage of a purchase maps onto a layer you have already studied. The new part this week is what happens at the stage where money moves, and what must be true before you let the agent get there.

What changes

Browse-and-click vs delegate-and-approve

The old shape: a person shops
Human: searches, opens five tabs, compares, decides, types a card number, clicks Buy. Merchant sees: a browser session, a card, a shipping address. Consent is the click. Identity is the card.

Everything the merchant needs to trust the sale is bundled into one human sitting at one screen. Fraud controls are built around that assumption.

The new shape: an agent shops
Human: states an intent and limits. Agent: searches, compares, builds a cart, comes back for approval, then transacts. Merchant sees: an automated client. Is it the customer's agent? Acting within what the customer allowed? Or a bot? Consent is now a claim that somebody has to verify.

The click is gone, so consent must be represented explicitly, as something a merchant can check. That is precisely what the new payment protocols on page 01 are trying to standardize.

Read this week as one sentence: agentic commerce is the moment the industry must write down, in a machine-checkable form, everything a human click used to imply.
See it move · an agent that buys

Where the money moves, and who can still say no

An office-supplies agent with a budget. Press Play to watch a purchase flow through an intent mandate, a cart, a cap check and a payment credential, then watch the same agent misread a catalog listing. The question the week keeps asking: at each arrow, what can go wrong here, and who pays when it does?

Try this: in the second scenario the agent is never “wrong” in its own trace. Point to the single control that caught the error, and say what would have happened without it.

This week's pages

Work through in order

Why managers care

The three questions a CFO asks first

Cost

Not “what does the model cost” but what does one completed task cost: tokens per step, times steps, times attempts, plus the human minutes spent reviewing and fixing. Until you can quote that number, you cannot compare the agent to the person doing the job today.

Trust

Trust here is a set of enforced limits. A spend cap, an allow-list of merchants, an approval step on anything irreversible, and a kill switch. Each one is a line of code somebody can point at.

Auditability

When a purchase is disputed, someone must be able to reconstruct it: what the user authorized, what the agent saw, what it chose, and what it paid. Your Week 2 evidence standard (state before → observation → available actions → selected action → result → state after) becomes a financial control.

← CourseAll weeks