Saginaw Valley State UniversityZahnow Library
Imagination LabAPPLIED AI LITERACY · A LIBRARY LEARNING GUIDE
Zahnow Library guides/Applied AI literacy
LEARN · EXPERIMENT · REFLECT

Applied AI lessons

10 lessons and six hands-on experiments. Work through them in order, at your own pace.

Lesson 5 of 10

Directing assistants and agents

In plain termsWhen you ask an AI to just answer, one good prompt is enough. When you ask an agent to do a multi-step task, it needs more context: what the goal is, what tools it may use, what “done” looks like, and when to stop and check with you. PASTAS is a checklist for writing that kind of brief, much like instructions you’d give a new teammate.

The context layer

PASTAS: how to write an agent prompt

Prompt engineering optimizes a single instruction. An agent plans, uses tools, and acts across multiple steps, so context engineering designs everything around it before it runs.

P · Purpose and Audience

Define what the agent is for and who it serves.

A · Authority

Set what the agent may decide and where its role ends.

S · Structure and Style

Shape how the agent organizes its work and how it sounds.

T · Tools and Access

Specify which tools and sources it can reach.

A · Accountability and Safeguards

Set the boundaries it observes and who answers for its output.

S · Signals and Review

Make uncertainty visible and route work to a human reviewer.

Practice: a PASTAS agent brief

Purpose and audience: You are a research assistant helping a [year] [major] student prepare a literature review on [topic].
Authority: You may suggest sources and outlines. You may not write final paragraphs for submission.
Structure and style: Return a table of 8 sources (author, year, claim, method), then a 5-point outline. Plain, academic tone.
Tools and access: Use only peer-reviewed sources you can link to. No paywalled guesses.
Accountability and safeguards: Do not invent citations. If unsure a source exists, say so.
Signals and review: Mark each claim as verified or needs-checking, and end with three questions I should ask a librarian.

You’ll learn: How to set up an agent so its work is checkable before it starts.

Sources: Lo, L. S. (2026). From Prompt to Agentic AI: The PASTAS Checklist for Context Engineering. portal: Libraries and the Academy, 26(3), pre-print. Lo, L. S. (Aug 1, 2025). Why Prompt Engineering is Still Crucial for Agentic Systems. LinkedIn.

In plain termsIn agent mode, ChatGPT can take steps on your behalf, like browsing the web, opening files, or drafting a plan. You give it a goal; it returns a draft for you to review. The important habit: you’re the reviewer. Read what it did before you send, submit, or share anything.

ChatGPT Work

Hand it a goal, get back a draft to review

Agent mode can connect to apps like Google Drive, Gmail, Calendar, and Slack (availability varies by plan and admin settings). It asks before important actions.

  1. 1
    You give it a goal and some files
  2. 2
    It plans and gathers context
  3. 3
    It uses your tools and apps
  4. 4
    You review, redirect, and approve

What it could help you get done

Study guide

From your notes and readings.

Class presentation

A first-draft deck.

Data analysis

Charts from a spreadsheet.

Research brief

Sources in a cited summary.

Weekly plan

A schedule from your deadlines.

Job search kit

Tracker, résumé drafts, interview prep.

Takeaway: You stay in charge: it drafts, you decide.

Sources: OpenAI: Get started with ChatGPT Work; chatgpt.com/work.

Check your understanding

1. You’re about to let an AI agent handle a multi-step task for you. What sets you up for success?
2. Why paste the assignment instructions into your prompt (instead of just the question)?
PreviousNext: experiment