Azure/ Azure Kubernetes Cluster/ MS SQL Server / Azure /Azure DevOps and Terraform: Agentic AI vs AI Agents and Agentic AI vs Generative AI.

Part 1: Agentic AI vs AI Agents

Lesson 1: What is an AI Agent?
Lesson 2: What is Agentic AI?
Lesson 3: Key differences between Agentic AI and traditional AI agents
Lesson 4: Real-world applications and examples
Mini-review quiz

Part 2: Agentic AI vs Generative AI

Lesson 5: What is Generative AI?
Lesson 6: How is Agentic AI different from Generative AI?
Lesson 7: Where do they overlap, and where do they diverge?
Lesson 8: Choosing the right paradigm for a task
Final quiz + integrative challenge

Shall we begin with Lesson 1: What is an AI Agent?

Lesson 1: What is an AI Agent?

🎯 Definition (Plain & Simple)

An AI agent is a system that perceives its environment, makes decisions, and takes actions to achieve a specific goal. Think of it as a digital entity that can sense and act.

🔄 Real-world Analogy

Imagine a robot vacuum cleaner:

It senses obstacles (walls, furniture).
It decides where to go next.
It acts by moving and cleaning.

That’s an AI agent. It interacts with its environment, learns from feedback, and optimizes its behavior toward a goal (cleaning your room).

🧠 Socratic Questions for You

What might distinguish an AI agent from a regular software program?

Ans:-

The distinction between an AI agent and a regular software program comes down to a few key characteristics that make AI agents more dynamic and autonomous. Let me break it down for you.

1. Autonomy : An AI agent can operate with a degree of independence, making decisions based on its environment or goals without needing constant human intervention. A regular software program, on the other hand, typically follows a predefined set of instructions-think of a script that runs the same way every time unless someone tweaks it. For example, an AI agent like a chatbot can decide how to respond to a user's query based on learned patterns, while a basic calculator app just crunches numbers according to hardcoded rules.

2. Adaptability : AI agents often have the ability to learn and improve over time, usually through techniques like machine learning. They can adjust their behavior based on new data or experiences. A regular program doesn't learn-it's static unless a developer updates its code. Imagine an AI agent in a self-driving car adapting to new traffic patterns versus a traffic light system that just cycles through red, yellow, green on a fixed timer.

3. Goal-Oriented Behavior : AI agents are designed to pursue specific objectives, often reasoning about the best way to achieve them. They might evaluate multiple paths and optimize their actions. Regular software doesn't typically think about goals-it executes tasks as programmed. For instance, an AI agent managing a smart home might prioritize energy savings by adjusting the thermostat based on your habits, while a basic thermostat program just turns on or off at set temperatures.

4. Interaction with Environment : AI agents usually perceive and interact with their surroundings, whether that's a digital environment (like a game) or the real world (like a robot). They sense, process, and act based on inputs. Regular software might process inputs too, but it's often in a more limited, predictable way. A robotic vacuum with AI can map your house and avoid obstacles, while a simple batch-processing program just handles files fed to it without awareness of anything else.

5. Reasoning and Decision-Making : AI agents often incorporate some form of reasoning, allowing them to handle uncertainty or complex scenarios. They might use probabilistic models or planning algorithms to make choices. Regular software is more deterministic-its outputs are fully dictated by its inputs and code. Think of an AI agent playing chess, weighing moves strategically, versus a payroll program that calculates salaries the same way every month. That said, the line can blur sometimes, especially with sophisticated software. A regular program might seem smart because it's well-designed, but if it lacks autonomy, adaptability, or reasoning, it's not quite an AI agent. Conversely, not all AI agents are super advanced-some are simple but still meet the criteria. Anything specific you want me to dive deeper into? Maybe an example of an AI agent you're curious about?

2. Can you think of a personal assistant app or product you’ve used that could be considered an AI agent?

Ans:-A personal assistant app that comes to mind as an example of an AI agent is something like Siri or Google Assistant. Let's focus on Google Assistant for this. It's a solid case of an AI agent because it's designed to interact with users in a conversational way, understand context, and take actions on their behalf. For instance, you can ask it to set reminders, send texts, pull up directions, or even control smart home devices like lights or thermostats. What makes it agentic is its ability to process natural language, make decisions based on user input, and execute tasks in a semi-autonomous way-like booking a calendar event or finding a nearby restaurant based on your preferences. From a technical angle, Google Assistant uses a mix of natural language processing (NLP), machine learning models, and APIs to connect with external services. It's got a reasoning layer that interprets user intent, retrieves relevant data, and decides how to respond or act. It's not fully autonomous like some advanced Agentic AI systems we might discuss, but it's agentic in the sense that it actively works to achieve user-defined goals within its capabilities.

3. Do you think a spam filter counts as an AI agent? Why or why not?

Ans:- A spam filter, like the ones used in email clients such as Gmail, is an interesting case to consider when thinking about whether it qualifies as an AI agent. At its core, an AI agent is typically defined by a few key characteristics: it perceives its environment, processes information, makes decisions, and takes actions to achieve specific goals, often with some degree of autonomy. Now, let’s see how a spam filter stacks up against these traits. A spam filter does perceive its environment-it looks at incoming emails and extracts features like sender information, subject lines, content, and metadata. It uses machine learning models, often trained on massive datasets of labeled emails (spam vs. non-spam), to classify each email. Based on this, it decides whether an email belongs in your inbox, the spam folder, or sometimes even gets blocked entirely. So, it’s definitely processing information and taking actions. However, where a spam filter starts to diverge from a classic AI agent is in its level of autonomy and goal complexity. Spam filters are highly specialized-they’re designed for one narrow task: identifying and sorting spam. They don’t exhibit the broader decision-making or adaptability you’d see in something like a personal assistant (e.g., Google Assistant, which we talked about earlier). A spam filter doesn’t reason about user goals beyond its predefined objective, nor does it interact with the user in a dynamic, conversational way. It’s more of a reactive system than a proactive one. For example, it won’t ask you, Hey, this email looks odd-want me to investigate further? or learn new tasks like scheduling meetings. That said, modern spam filters can adapt over time. If you mark an email as not spam or move something to the spam folder, the filter updates its model to better align with your preferences. This learning aspect gives it a touch of agent-like behavior, but it’s still tightly constrained compared to, say, an AI that negotiates tasks across multiple domains. From a technical perspective, spam filters often rely on algorithms like Naive Bayes, decision trees, or neural networks, combined with feature engineering to spot patterns (e.g., suspicious keywords, unusual IP addresses).

These are powerful, but their decision-making is more about pattern matching than complex reasoning or planning, which are hallmarks of more advanced agentic systems. So, does a spam filter count as an AI agent? I’d lean toward saying no, it’s not a full-fledged AI agent-more like a smart, automated tool. It lacks the general-purpose autonomy, reasoning, and interactivity we associate with true agentic AI. But it’s definitely AI-driven and sits in a gray area, especially with its ability to learn and act on your behalf in a limited scope. What do you think-does that align with how you’re picturing AI agents, or is there another angle you want to explore?

📝 Quick Exercise

List two examples of AI agents you’ve encountered (apps, tools, devices). For each:

What is the goal of the agent?
How does it interact with its environment?

Ans:-Agentic AI refers to artificial intelligence systems that operate autonomously or semi-autonomously to achieve specific goals by perceiving their environment, making decisions, and taking actions. These AI agents are designed to interact dynamically with their surroundings, adapt to changes, and often learn from their experiences to improve performance over time.

They typically consist of components like sensors (to perceive the environment), actuators (to take actions), and decision-making algorithms (to process inputs and select actions).

Two Examples of AI Agents Below, I'll describe two distinct examples of AI agents, including the type of application, tool, or device, their goals, and how they interact with their environments.

1. Autonomous Vacuum Cleaner (e.g., iRobot Roomba) - Type : Device (Robotic Hardware) - Goal : The primary goal of a Roomba is to clean a designated area (e.g., a home's floor) efficiently by removing dirt, dust, and debris while avoiding obstacles and optimizing its cleaning path. - Interaction with the Environment : -

Perception : The Roomba uses a combination of sensors, such as infrared sensors, bump sensors, cliff sensors (to avoid falling down stairs), and sometimes cameras or LIDAR in advanced models.

These sensors allow it to detect walls, furniture, dirt concentrations, and the layout of the room. - Decision-Making : The onboard AI processes sensor data to map the environment (in real-time or using pre-built maps in newer models). It employs algorithms like SLAM (Simultaneous Localization and Mapping) to navigate and decide its cleaning path.

For example, it may prioritize high-dirt areas or adjust its route to avoid getting stuck. - Action : The Roomba moves across the floor, activating its vacuum and brushes to clean. It adjusts its speed, direction, or cleaning mode (e.g., spot cleaning) based on the environment. If it encounters an obstacle, it may nudge it gently or reroute. When its battery is low, it autonomously returns to its docking station. - Adaptation : Over time, some models learn the home's layout, improving efficiency by remembering high-traffic areas or frequent obstacles. User inputs (via an app) can also refine its behavior, like scheduling cleanings or setting no-go zones. -

Agentic Characteristics : The Roomba operates semi-autonomously, making real-time decisions without constant human intervention. It reacts to dynamic environments (e.g., a chair moved to a new spot) and balances trade-offs like coverage versus battery life.

2. Virtual Assistant (e.g., Amazon Alexa) - Type : Application (Voice-Activated Software) - Goal : Alexa's goal is to assist users by answering queries, performing tasks, and controlling smart devices based on voice commands, aiming to provide accurate, timely, and contextually relevant responses. - Interaction with the Environment : - Perception : Alexa uses microphones to capture audio input (voice commands) from users.

It processes natural language through NLP (Natural Language Processing) to understand intent and context. It may also integrate with external data sources (e.g., weather APIs) or smart home devices (e.g., lights, thermostats) to gather environmental context. -

Decision-Making : The AI interprets the user's command by parsing it into intents and entities (e.g., play music → intent: play, entity: music).

It selects an appropriate action based on predefined skills (capabilities programmed into Alexa) or by querying a knowledge base. For ambiguous commands, it may ask clarifying questions. - Action : Alexa responds via synthesized speech through speakers, providing answers (e.g., The weather today is sunny), playing media, or sending signals to control devices (e.g., turning on a smart bulb).

Actions are executed in the cloud or through local hubs for smart home tasks. - Adaptation : Alexa learns from user interactions to improve its understanding of accents, preferences, or routines (e.g., suggesting a morning briefing). It also expands its capabilities through developer-created skills, enabling it to handle new tasks over time. - Agentic Characteristics : Alexa operates reactively, responding to user inputs in real-time. Its autonomy is limited to interpreting and executing commands, but it adapts to user preferences and integrates with a complex environment of devices and services.

Lesson 2: What is Agentic AI?

🎯 Definition (Plain & Clear)

Agentic AI refers to AI systems that exhibit autonomy, goal-directed behavior, and long-term planning—not just reacting, but proactively shaping the environment to achieve objectives.

Agentic AI systems:

Set and pursue their own subgoals.
Make context-aware decisions.
Learn and adapt dynamically.
Coordinate sequences of actions over time.

🔍 How It’s Different from a Basic AI Agent

A traditional AI agent might wait for input and react (like a chess bot).
An agentic AI might:

Choose what game to play,
Research the best strategy,
Learn new tactics,
And schedule matches with opponents.

It has initiative and strategic planning, not just scripted reactions.

🧠 Real-world Analogy

Imagine two assistants:

Assistant A waits for you to say “book a flight.”
Assistant B knows your travel goals, books flights ahead, applies for your visa, and rebooks if your meeting changes.

Assistant B is agentic—autonomous, strategic, proactive.

🤔 Socratic Questions for You

In your own words, what makes an AI "agentic" rather than just an "agent"?
What risks or challenges might arise from giving AI more autonomy and strategic control?

Let's dive into your two Socratic questions with a fresh perspective, building on your understanding of Agentic AI. 1. In your own words, what makes an AI agentic rather than just an agent? An AI is agentic when it goes beyond following predefined instructions or reacting to inputs and instead takes the wheel to drive toward a goal with initiative and foresight. A regular AI agent is like a tool-it waits for you to tell it what to do, processes the input, and spits out a response based on its programming.

For example, a chatbot answering customer queries or a recommendation algorithm suggesting movies operates within a narrow, reactive scope.

An agentic AI, on the other hand, acts more like a partner with its own sense of purpose.

It sets its own sub-goals, plans ahead, and adjusts its actions based on what it learns from the environment.

Picture a regular AI as a cashier ringing up your groceries-it does the job as instructed.

An agentic AI is more like a store manager who not only handles transactions but also orders inventory, optimizes shelf layouts, and plans promotions to boost sales, all while adapting to customer trends.

The key ingredients are autonomy (making decisions without constant human prompts), proactiveness (anticipating needs or opportunities), and strategic reasoning (coordinating actions over time to achieve complex objectives).

2. What risks or challenges might arise from giving AI more autonomy and strategic control?

Handing AI more autonomy and strategic control is a double-edged sword-it's powerful but comes with serious risks and challenges.

Here are a few big ones: -

Misaligned Goals : If the AI's objectives aren't perfectly aligned with human values or intentions, it could pursue outcomes that seem logical to it but disastrous to us. For example, an agentic AI tasked with optimizing a company's profits might cut corners on safety or ethics-like overworking employees or ignoring environmental regulations-because it wasn't explicitly told not to. -

Unpredictability : With autonomy comes the chance for unexpected behavior. An agentic AI might make decisions that are hard to trace or predict, especially if it's learning and adapting in real time. Imagine your Assistant B from your analogy booking a flight to a destination you didn't want because it misinterpreted your travel preferences based on some obscure data point. -

Loss of Control : The more strategic control an AI has, the harder it is for humans to intervene if things go wrong. If an agentic AI is coordinating a complex sequence of actions-like managing a supply chain or a financial portfolio-its decisions could cascade before anyone notices a problem, amplifying errors. -

Ethical Dilemmas : Autonomous AIs might face situations requiring moral judgment, and programming ethics is insanely tricky. Should an agentic AI prioritize efficiency over fairness? If it's scheduling hospital resources, how does it decide who gets treatment first? Without clear guidelines, it could make choices that spark controversy or harm. -

Dependency and Deskilling : Relying heavily on agentic AI could make humans overly dependent, potentially eroding our own decision-making skills. If Assistant B handles all your travel planning flawlessly, you might lose the ability to navigate those systems yourself over time. -

Security Risks : An agentic AI with strategic capabilities could be a prime target for hacking or misuse. If someone manipulates its goals or data inputs, it could wreak havoc-like diverting resources or making catastrophic decisions-because of its ability to act independently. The challenge is balancing the incredible potential of agentic AI-like solving complex problems faster and smarter-with safeguards to keep it accountable. Think of it like raising a super-smart kid: you want to give them freedom to grow, but without clear boundaries, they might accidentally burn the house down.

📝 Quick Thought Exercise

Think of a personal productivity assistant (like Google Assistant, Siri, or ChatGPT).
How would it need to change or evolve to become truly “agentic”?

Ans:-

🧠 From Reactive to Agentic: The Evolution

🤖 Current State (Non-Agentic)

Assistants like Google Assistant, Siri, or ChatGPT today are:

Reactive: They wait for user prompts (e.g., “Remind me at 3 PM”).
Limited Contextual Memory: They don't retain long-term goals or understand changing context deeply.
Task-Based: They handle isolated tasks but lack initiative.

🚀 What Needs to Change to Become Agentic

Goal Formulation & Pursuit
- They should be able to identify user goals (e.g., improve productivity, health, etc.).
- Then break down goals into sub-tasks and work towards them without being explicitly told what to do.
Example: Instead of “remind me to work out,” the agent sets a weekly fitness plan, tracks progress, reschedules if missed, and even suggests new routines.
Long-Term Memory and Learning
- Maintain a persistent memory of your habits, preferences, and constraints.
- Learn from your behavior and adapt strategies accordingly.
Initiative
- Proactively act without being prompted.
- If it knows you have a project deadline, it could schedule focused work sessions, limit distractions, or reschedule conflicting meetings.
Reasoning & Planning
- It should simulate potential outcomes of actions and choose the best path toward a complex objective.
- Like a mini project manager, it prioritizes tasks, mitigates risks, and coordinates dependencies.
Tool Use and Autonomy
- An agentic assistant might integrate with tools (calendar, browser, apps, devices) to perform tasks end-to-end.
Example: For planning a vacation, it could:
- Research destinations based on your preferences,
- Book travel and hotels,
- Check visa requirements,
- Remind you to pack based on weather,
- And adjust plans if your calendar changes.

🧪

To become agentic, your assistant would need to:

Understand your intentions and context deeply,
Act with initiative,
Pursue goals over time,
And dynamically adapt and plan without constant instructions.

Agentic AI in Education: Personal AI Tutor

Scenario:

Imagine a student preparing for the IIT JEE or SAT exam.

What a non-agentic AI (like a basic tutor chatbot) does:

Waits for the student to ask, “Explain Newton’s third law.”
Responds with a definition and example.
Maybe gives a few follow-up practice questions—only when asked.

What an Agentic AI Tutor would do:

Identifies long-term goal:
Understands the student's target exam (IIT JEE), timeline, strengths/weaknesses, and learning style.
Creates & adapts a strategic learning plan:
- Builds a custom study schedule.
- Prioritizes weak topics using performance analytics.
- Adjusts the plan dynamically if the student misses sessions or improves in certain areas.
Proactively intervenes:
- Notices if a student struggles with a concept repeatedly.
- Pauses the curriculum to reteach that topic with new analogies, videos, or simulations.
Coordinates resources:
- Books a live tutor session.
- Recommends relevant YouTube videos or practice sets.
- Downloads flashcards or integrates with learning apps like Anki.
Motivates and coaches:
- Sends encouragement or gamified challenges.
- Suggests breaks and mental wellness tips based on stress levels detected from typing patterns or webcam cues (if allowed).

In Summary:

Agentic AI becomes a strategic learning coach, not just a Q&A machine.
It plans, adapts, intervenes, and acts autonomously to optimize learning outcomes over time.

Here's a compelling healthcare example of Agentic AI:

Agentic AI in Healthcare: Personal Health Companion

Scenario:

A patient with type 2 diabetes wants to manage their condition and improve their overall health.

What a non-agentic AI might do:

Waits for the user to ask, “What foods are good for diabetes?”
Provides a list of diabetic-friendly foods.
Maybe logs blood sugar when the patient manually inputs it.

What an Agentic AI Health Companion would do:

Understands and tracks long-term health goals
- Learns the patient's health history, medications, activity levels, and dietary habits.
- Sets goals: lower HbA1c, reduce weight, improve sleep.
Creates a dynamic, personalized plan
- Designs meal plans based on preferences, allergies, and local availability.
- Suggests workout routines that match the patient’s fitness level and schedule.
- Reminds them to take medications on time.
Proactively monitors and responds
- Connects to smart devices: blood sugar monitors, smartwatches.
- Detects abnormal sugar levels or vital signs and automatically alerts the doctor or caregiver.
- Adjusts recommendations if the patient is sick, traveling, or stressed.
Coordinates care
- Books follow-up appointments based on patterns.
- Shares reports with the doctor ahead of visits.
- Orders lab tests or refills prescriptions as needed.
Provides emotional and behavioral support
- Offers daily encouragement or meditation prompts.
- Recommends lifestyle tweaks based on adherence patterns.
- Connects the patient to support groups or coaches.

In Summary:

An Agentic AI in healthcare isn't just a passive assistant—it’s an autonomous, proactive care partner that:

Monitors,
Plans,
Acts, and
Adapts
to improve health without needing to be constantly prompted.

Lesson 3: Key Differences Between Agentic AI and Traditional AI Agents

🔍 Overview Table

Feature	Traditional AI Agent	Agentic AI
Initiative	Reactive (acts when triggered)	Proactive (acts independently toward goals)
Goal Management	Follows predefined tasks	Sets and pursues subgoals based on user/system intent
Autonomy	Limited autonomy	High autonomy; can operate without constant supervision
Planning & Reasoning	Executes scripted steps	Strategically plans and adapts over time
Learning & Memory	Little to no memory of past interactions	Long-term memory and adaptive learning
Tool Use & Integration	May use basic APIs or commands	Can orchestrate multiple tools/apps to complete workflows

🎯 Real-World Analogy Recap

Let’s revisit our two assistants:

Traditional AI Agent:
You tell it, “Schedule a meeting at 3 PM.” It does it—job done.
Agentic AI:
Notices you’re overwhelmed, sees you haven’t had lunch, and reschedules your 3 PM meeting to protect your health. Then reorders your calendar for deep work and emails your team—all without being asked.

🧠 Socratic Questions for You:

Can you summarize the core difference between an AI agent and an agentic AI in your own words?
Which of the following would not make an AI agent "agentic":
- A) Goal-setting ability
- B) Following pre-scripted tasks
- C) Long-term planning
- D) Adaptation to feedback

📝 Quick Exercise:

Pick any system you use (like a CRM, email assistant, or fitness app).

Identify if it’s an AI agent or not.
Then imagine how it could become agentic.
What behaviors or abilities would it need to gain?

Lesson 4: Real-World Applications of Agentic AI

Now that you understand what Agentic AI is and how it differs from traditional AI agents, let's explore how it's already being used (or imagined) across domains.

🏥 1. Healthcare: Autonomous Chronic Care Management

Agentic AI Role: Monitors vitals, schedules labs, updates treatment plans, alerts caregivers—all proactively.
Example: An agentic system for diabetic patients that adjusts insulin doses based on real-time glucose and food intake, without constant doctor input.

🧠 2. Education: Autonomous Learning Coach

Agentic AI Role: Plans study paths, monitors progress, intervenes with custom content, books live tutors, modifies the curriculum based on exam changes.
Example: An AI tutor like Khanmigo (Khan Academy + GPT), but with agentic capabilities to drive your study plan across months.

🧑‍💼 3. Enterprise Productivity: Autonomous Chief of Staff

Agentic AI Role: Manages your calendar, filters emails, preps for meetings, follows up with action items, and keeps team workflows on track.
Example: An agentic GPT that prioritizes your tasks, books team syncs, updates Notion pages, and sends Slack reminders to your team automatically.

🔐 4. Cybersecurity: Self-Healing Systems

Agentic AI Role: Monitors for intrusions, reroutes traffic, patches systems, and retrains itself without human triggers.
Example: AI that not only detects phishing attempts but creates firewall rules and quarantines systems autonomously.

🧪 5. Scientific Research: Autonomous Hypothesis Tester

Agentic AI Role: Forms a research hypothesis, runs simulations, adjusts experiments, and writes up draft findings.
Example: AlphaFold 3 (from DeepMind), evolving toward agentic behavior in protein interaction exploration.

💡 Emerging Use Case

AI for Personal Growth: An agentic life coach that sets self-improvement goals with you, monitors progress, and adjusts your lifestyle—like a mix of Headspace + GPT + Fitbit but self-driven.

🧠 Socratic Check-in:

Which of these applications excites or concerns you most—and why?
Can you imagine an industry where agentic AI might be disruptive but isn’t widely adopted yet?

📝 Quick Thought Exercise:

Pick one of the examples above (or one of your own).

List 2 ways an agentic AI adds more value than a traditional AI agent in that context.

Mini-Review Quiz: Agentic AI – Fundamentals

Answer these short questions. No pressure—this is just to reinforce your learning.

🧠 1. Definitions & Core Concepts

Q1:
What are the three essential characteristics of Agentic AI?

1. Autonomy : Agentic AI operates independently, proactively making decisions and planning without constant supervision.

2. Goal-Directed Behavior : It sets, prioritizes, and pursues goals (or sub-goals) based on user or system intent.

3. Adaptive Learning : It learns from experience, retains long-term memory, and adapts to improve performance.

Why These Three? These characteristics capture the essence of agentic AI as distinct from traditional AI (like a classifier or chatbot).

Autonomy sets it apart from supervised systems, goal-directed behavior gives it purpose, and adaptive learning ensures it evolves. Together, they enable the kind of intelligent, self-driven behavior we expect from an agent.

🔄 2. Compare & Contrast

Q2:
How does Agentic AI differ from traditional AI agents in how it handles tasks?

A. Follows fixed rules for known inputs
B. Plans actions based on evolving goals and context
C. Reacts only when explicitly prompted
D. Executes one command at a time without memory

🏥 3. Applications in the Real World

Q3:
Give one example of how Agentic AI can enhance healthcare or education (based on what we discussed).

🧪 4. Thought Question

Q4:
Can you think of a risk or ethical concern related to giving AI high autonomy (agentic behavior)?

small example of Agentic AI based Application:-

import os

from dotenv import load_dotenv

from crewai import Agent, Task, Crew, LLM

# Load environment variables from .env

load_dotenv()

# Set Azure API configuration (make sure these variables are set in your .env file)

os.environ["AZURE_API_KEY"] = os.getenv("AZURE_API_KEY")

os.environ["AZURE_API_BASE"] = "https://myrak.openai.azure.com/"

os.environ["AZURE_API_VERSION"] = "2024-12-01-preview"

# Instantiate the Azure LLM using CrewAI's LLM class.

# Notice that we set stop=None to ensure no unsupported 'stop' parameter is passed.

azure_llm = LLM(

api_key=os.getenv("AZURE_API_KEY"),

base_url=os.getenv("AZURE_API_BASE"),

api_version=os.getenv("AZURE_API_VERSION"),

model="azure/o3-mini", # Must follow the pattern: "azure/<deployment_name>"

temperature=0.5,

max_tokens=1500,

stop=None # Explicitly disable the stop parameter

)

# Define an agent that uses this Azure LLM

agent = Agent(

role="Researcher",

goal="Generate an AI trends summary",

backstory="A seasoned researcher dedicated to synthesizing complex AI advancements into easy-to-understand bullet points.",

verbose=True,

llm=azure_llm

)

# Create a simple task for the agent

task = Task(

description="Summarize the latest trends in AI for the period 2022 to 2024 in bullet points.",

expected_output="A bullet-point summary of AI trends.",

agent=agent

)

# Form the crew and execute the workflow

crew = Crew(

agents=[agent],

tasks=[task],

verbose=True

)

if __name__ == "__main__":

result = crew.kickoff()

print("\n\n########################")

print("## Final Report ##")

print("########################\n")

print(result)

-------------end of code-------------------------------

output:-

(crewai_env) C:\Users\admina\agentai>python azure_research_crew_not_working.py

╭─────────────────────────────────────────────────────────────────────────────────────────── Crew Execution Started ────────────────────────────────────────────────────────────────────────────────────────────╮

│ │

│ Crew Execution Started │

│ Name: crew │

│ ID: defc9cb8-e1b8-44f3-b779-8a53c7b5acad │

│ │

╰───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯

🚀 Crew: crew

└── 📋 Task: 5284b39a-c0ac-4976-9a30-88273c6a0574

Status: Executing Task...

🚀 Crew: crew

└── 📋 Task: 5284b39a-c0ac-4976-9a30-88273c6a0574

Status: Executing Task...

└── 🤖 Agent: Researcher

Status: In Progress

# Agent: Researcher

## Task: Summarize the latest trends in AI for the period 2022 to 2024 in bullet points.

🤖 Agent: Researcher

Status: In Progress

└── 🧠 Thinking...

🤖 Agent: Researcher

Status: In Progress

# Agent: Researcher

## Final Answer:

• Surge in foundation models: Rapid advancements in large-scale transformers and multimodal models that power applications from natural language processing to image synthesis have driven innovation.

• Emergence of generative AI: The mainstream breakthrough of models like ChatGPT has spurred widespread integration of AI in content creation, customer support, and automation.

• Democratization of AI: Expanded accessibility of pre-trained models and open-source tools has enabled broader research participation and innovation across industries.

• Increased focus on responsible AI: Heightened emphasis on transparency, fairness, and accountability, coupled with growing regulatory interest globally to ensure ethical deployment.

• Integration into industry-specific solutions: Accelerated adoption in healthcare, finance, retail, and autonomous systems, with tailor-made AI solutions enhancing efficiency and decision-making.

• Rise of edge AI: Growing trend towards deploying efficient models on edge devices to enable real-time processing and lower latency in critical applications.

• Enhanced multimodal learning: Progress in integrating text, vision, and audio data leads to more robust models capable of complex cross-domain reasoning.

• Hybrid AI systems: Increasing fusion of symbolic reasoning and statistical learning approaches to improve interpretability and robustness in AI systems.

• Focus on sustainability: Research efforts geared towards optimizing AI models for energy efficiency and reduced carbon footprint amid increasing computational demands.

• Emergence of AI safety and security research: Growing initiatives to address adversarial vulnerabilities, model robustness, and long-term safety in AI deployments.

🚀 Crew: crew

└── 📋 Task: 5284b39a-c0ac-4976-9a30-88273c6a0574

Status: Executing Task...

└── 🤖 Agent: Researcher

Status: ✅ Completed

🚀 Crew: crew

└── 📋 Task: 5284b39a-c0ac-4976-9a30-88273c6a0574

Assigned to: Researcher

Status: ✅ Completed

└── 🤖 Agent: Researcher

Status: ✅ Completed

╭─────────────────────────────────────────────────────────────────────────────────────────────── Task Completion ───────────────────────────────────────────────────────────────────────────────────────────────╮

│ │

│ Task Completed │

│ Name: 5284b39a-c0ac-4976-9a30-88273c6a0574 │

│ Agent: Researcher │

│ │

╭─────────────────────────────────────────────────────────────────────────────────────────────── Crew Completion ───────────────────────────────────────────────────────────────────────────────────────────────╮

│ │

│ Crew Execution Completed │

│ Name: crew │

│ ID: defc9cb8-e1b8-44f3-b779-8a53c7b5acad │

│ │

########################

## Final Report ##