Right Now, the Paradigm Is Shifting
Left side — a person typing a question into ChatGPT at a laptop. Right side — the same person standing like an orchestra conductor, directing multiple AI agents performing different tasks. A large arrow in the center. Professional, minimal illustration style.
Image Prompt: "Split illustration, left side: a researcher typing a question into a chatbot on a laptop screen, simple and passive interaction. Right side: the same researcher standing like an orchestra conductor, directing multiple AI agents represented as glowing orbs performing different tasks (coding, searching, writing). A large arrow in the center pointing from left to right. Clean, professional, minimal flat illustration style, blue and white color palette."
Before: The AI Tool User Era
😅 A Familiar Scene
- Ask ChatGPT a question → copy answer → paste into document
- "Fix this code" → manually apply the result to the file
- Search → request summary → search again → summarize again...
- Every step is manually connected by a human
A person going back and forth between ChatGPT and a Word document, copy-pasting. Arrows tangled everywhere, person looks exhausted. Simple diagram style.
Image Prompt: "Diagram showing a frustrated researcher sitting at a desk, manually copying text from a ChatGPT window to a Word document, then to a code editor, then back to ChatGPT. Multiple tangled arrows showing the manual back-and-forth workflow. Simple, clean diagram style with subtle humor, muted colors."
After: The Research Director Era
🎬 The New Landscape
- "Find 5 relevant papers on this topic and summarize them" → Agent searches, downloads & summarizes
- "Analyze this data, visualize it, and draft a report" → Agent runs the entire pipeline
- "Find bugs in the code, fix them, and run the tests" → Agent executes autonomously
- The human sets the goal and verifies the result
A person sitting comfortably, looking at a dashboard. Multiple AI Agents performing paper search, code writing, data analysis, report drafting. Clean, futuristic feel.
Image Prompt: "A researcher comfortably sitting in a modern office, looking at a dashboard screen. Around them, multiple AI agents (represented as friendly robot assistants or glowing orbs) are simultaneously performing tasks: one is searching academic papers, another is writing code, another is analyzing data charts, another is drafting a report. Clean, futuristic but warm illustration style, blue and green tones."
Understanding by Analogy: The Film Director
🎬 A Film Director...
- Doesn't act in scenes (= doesn't code everything)
- Doesn't operate the camera (= doesn't search everything)
- Sets the vision, orchestrates the team, and takes responsibility for quality
🎯 A Research Director...
- Doesn't write all the code personally
- Doesn't read every paper manually
- Sets the research direction, orchestrates AI Agents, and takes responsibility for accuracy
Left — a film director in a chair shouting "Action!" directing a crew. Right — a researcher directing multiple AI Agents on screens. 1:1 visual parallel.
Image Prompt: "Split comparison illustration. Left: a film director sitting in a director's chair on a movie set, saying 'Action!' with camera crew, actors, and lighting team around them. Right: a researcher at a modern desk, directing multiple AI agents displayed on screens - one agent searching, one coding, one analyzing. Visual parallel between the two scenes. Professional, warm illustration style."
Tool vs Agent — The Core Difference
| Category | AI Tool | AI Agent |
|---|---|---|
| Analogy | A power drill | A skilled carpenter |
| Behavior | Only works when the user presses the button | Given a goal, plans and executes on its own |
| Example | "Translate this sentence" | "Read this paper and create a Korean summary report" |
| On failure | Outputs an error message | Analyzes the cause and tries a different approach |
| Memory | Starts fresh every time | Remembers context from previous tasks |
| Tool use | None (it IS the tool) | Uses search, code execution, file management, etc. |
Left — a power drill (tool), a person holding it and doing manual work. Right — a skilled carpenter (agent), independently building furniture from a blueprint. Simple analogy illustration.
Image Prompt: "Split comparison illustration. Left side labeled 'Tool': a person holding a power drill, manually drilling into wood - they control every movement. Right side labeled 'Agent': a skilled carpenter independently building furniture from a blueprint - they plan, measure, cut, and assemble on their own. The person just provided the blueprint. Clean, minimal illustration, educational style."
Human-in-the-Loop (HITL)
Agent Starts
➔
Result
➔
⚠️ Human Review
➔
Continue or Redirect
➔
Final Output
💡 When HITL Is Needed
- When interpreting insights from data (statistical significance judgment)
- When ethical judgment is required (personal data inclusion)
- When making creative direction decisions (choosing research hypotheses)
- Before results are published externally (papers, reports)
⚠️ What Happens Without HITL?
- AI cites non-existent papers (hallucination)
- Biased data leads to incorrect conclusions
- Copyright-infringing content generated
- Credibility and academic integrity compromised
Pipeline diagram. Automated steps by AI Agent with "Human Review" checkpoints in between. Checkpoint has human icon with green checkmark / red X.
Image Prompt: "A horizontal pipeline diagram showing an AI agent workflow. Multiple automated steps (search, analyze, generate) shown as connected blue boxes with arrows. At key decision points, there are 'checkpoint' gates with a human icon, a green checkmark and a red X. The human figure is reviewing the intermediate output before allowing the pipeline to continue. Clean, professional infographic style."
🛠️ Vibe Coding with Google AI Studio
📚 Google AI StudioCreate a web app that plots $y=ax^2+bx+c$ based on the input of $a, b, c$
Google AI Studio
Web App UI created by Gemini
Web App UI created by Gemini
Web App Code created by Gemini
Local Environment Setup (Windows) (1/3)
https://www.python.org/downloads/windows/
Find Python 3.12.10 via Ctrl+F
py -3.12 --version
py -m pip --version
Local Environment Setup (3/3)
python -m pip install numpy pandas matplotlib jupyter ipykernel- Download and install:
https://antigravity.google/download - Log in with your Google account.
- Open your
UST_Lecture_AgenticAIfolder. - Follow instructions to verify
.venvis recognized.
🎯 First Mission: Your First Directed Agent
"Activate venv, write
demo_basic.py that plots $y=x^2$ using numpy/pandas/matplotlib, saves as demo_basic.png, and execute it."sequenceDiagram
participant U as You (Director)
participant A as Antigravity Agent
participant T as Terminal
U->>A: Natural Language Prompt
Note over A: Plan -> Code -> Execute
A->>A: Create demo_basic.py
A->>T: Run python demo_basic.py
T-->>A: Generates demo_basic.png
A-->>U: Review Output & Chart
Antigravity IDE
Ask Antigravity to create the code
Approve the execution of Antigravity
Final confirmation of the execution
Created code and plot created by the code
🗣️ Discussion
Is AI a research assistant or a crutch? Let's define the boundary!
📚 UST LMS → This Course → Discussion → Week 1
graph LR
Step1[1. Go to Discussion Forum in UST LMS] --> Step2[2. Read AI Panel
Opinions: Iron Man, Captain America, Hulk] Step2 --> Step3[3. Vote with Reaction ("1,2", "3", or "None")] Step3 --> Step4[4. Reply with your
Stance & Rationale] style Step4 fill:#ffe082,stroke:#ffb300,stroke-width:2px
Opinions: Iron Man, Captain America, Hulk] Step2 --> Step3[3. Vote with Reaction ("1,2", "3", or "None")] Step3 --> Step4[4. Reply with your
Stance & Rationale] style Step4 fill:#ffe082,stroke:#ffb300,stroke-width:2px
UST LMS → This Course → Discussion
Week 1 Discussion
Dicussion Topic → Reply
Reply format: the numbers of panels you agree / your opinion