AI mascot

Generative · Agent · Agentic

Osama Ali

linkedin.com/in/os3li

Which AI tool should I use?

Confused ghost ? ? ?
? ? ? ? ?

The tool matters less than the level of AI powering it.

Generative → Agent → Agentic
INTRO

Generative · Agent · Agentic:
What’s the Difference?

A visual guide for Product Managers & UX Designers.

📜 🖊️
AI mascot explaining the three levels
⚡ Not All AI Is Equal.

There are 3 levels: Generative, Agent, and Agentic. Each changes your workflow in fundamentally different ways.

🔄 One Input, Different Outputs.

Give the same prompt to three AIs and you’ll get three wildly different results. The difference isn’t the prompt — it’s the level of intelligence behind it.

🧑‍💻 Human-in-the-Loop.

AI can hallucinate — confidently generate wrong answers. Every level still needs a human to verify, guide, and decide. You are the quality gate.

Level 1 — You Prompt, It Produces

Generative AI.

Definition: AI that creates content — text, images, designs — from a single prompt. One input → one output.

Think: ChatGPT, Midjourney, Figma AI — tools that produce when you ask.

“Same prompt, different result every time. No memory, no learning, no initiative. You are the brain — I’m just the hands.”

🔑 Key: Same prompt → Different result each time. No memory, no learning, no initiative. You are the brain.
👻 Ghost AI
GPT-4o
🧠 GPT-4o
🟣 Claude 3.5
💎 Gemini Pro
🦙 LLaMA 3
AI mascot
👻 Ghost AI is ready. What would you like to create?
Level 2 — The Tool User

AI Agent.

An AI model — but now with hands. It can search the web, read files, execute code, and call APIs. You delegate a task, it uses tools to finish it.

Nature: Limited Autonomy. You set the task, it picks the tools.

Ghost Agent — VS Code
EXPLORER
📁 src
📄 app.js
📄 utils.js
📁 tests
📄 app.test.js
app.js
1const express = require('express');
2const app = express();
3
4app.get('/api', (req, res) => {
5 res.send(data); // ❌ bug: data is undefined
6});
7
TERMINAL
$ ▌
Ghost agent Agent Workflow
🔍 Read error log…
🧠 Recall: similar bug in utils.js (2 days ago)
🔧 Apply fix → async/await pattern
✅ Run tests → 12/12 passed
Level 3 — Goal → Plan → Execute → Learn

Agentic AI.

Give it a business goal, not a task. It analyzes data, builds a plan, designs solutions, tests them, and iterates — until the target is hit.

Jira — Onboarding Sprint 14
UX-347 🔴 Critical
60 % of users click “Skip” on onboarding
TO DO 🤖 AI Agent
60 % skip rate
Analyst 📋 Reading Jira ticket…
📄 problem.md
→ Figma Design — 3 Variations
📄 solution.md — received
✅ UX-347 Resolved — Winner deployed
onboarding-flow.fig
Variation A
Variation B
Variation C
+10 % Projected Conversion Lift
Designer 🎨 Generating variations…
🎨 Generating variation A…
🎨 Generating variation B…
🎨 Generating variation C…
📄 variants.md
📄 heatmap-results.md — received
🚀 Deploying Variation B…
📄 solution.md → sent to Jira
Agent Reasoning
📖Read Jira ticket
🎨Generate variations
🔥Fetch heatmap data
✅Deploy winner
Var A
Var B
Var C
QA 🔥 Fetching HotJar data…
📄 heatmap-results.md
📤 Sending results to Figma…
🎯 You define the What
🤖 AI figures out the How
✅ You approve the plan
Side by Side

See the Difference.

3 levels of AI — autonomy, memory & initiative compared.

1 / 7