AI Trends 2025 🚀

🦜 LangChain's State of AI Agents Report

Source: LangChain State of AI Agents Report (2024)

🦜 LangChain's State of AI Agents Report

Source: LangChain State of AI Agents Report (2024)

🦜 LangChain's State of AI Agents Report

Source: LangChain State of AI Agents Report (2024)

🦜 LangChain's State of AI Agents Report

• 51% of surveyed companies have integrated AI agents into production environments.
• 78% of organizations are actively developing AI agents for near-future deployment.
• 90% of non-tech companies are either using or planning to use AI agents.

Source: LangChain State of AI Agents Report (2024)

The Next Buzzword of 2025

Agentic AI

The future of autonomous, intelligent systems.

Media Mentions of Agentic AI

• Looking At The Crystal Ball: 2025 Predictions For Agentic AI (21.01.2025)
• 5 Predictions for AI in 2025 (16.01.2025)
• AI Agents Are Here. How Much Should We Let Them Do? (15.01.2025)

🎖️ Honorable Mention 🎖️

Agentic AI

Source: AI Agents vs Agentic AI, What's the Difference and Why Does It Matter?

Agentic AI vs AI Agents

Agentic AI vs AI Agents

Agentic AI

Agentic AI Actions

Agentic AI Actions

Source: Devin AI

How can we build Agentic AI

• Model (LLM): The "brain" generating text/logic.
• Orchestration: Multi-step logic & planning
  • ReAct (ReAct)
  • Chain-of-Thought (CoT)
  • Tree-of-Thoughts ( ToT)
• Tools:
  • Extensions (agent-side, real-time external calls)
  • Functions (model outputs arguments, executed client-side)
  • Data Stores (vector DB, document retrieval)

How can we build Agentic AI

Source: Agents

Extensions

Extensions

Bridging between the model and the outside world

• Teaches the agent how to use the API endpoint
• Teaches the agent what arguments are needed

Extensions

Modular tools that give agents real-world capabilities

Example

import vertexai
import pprint

PROJECT_ID = "YOUR_PROJECT_ID" 
REGION = "us-central1"
vertexai.init(project=PROJECT_ID, location=REGION)

zefix_extension = Extension.create(
    manifest={
            "name": "zefix",
            "description": "Access to the swiss central 
                business name index",
            "api_spec": {
                "openApiYaml": ZEFIX_OPENAPI,
            },
            "auth_config": {
                "auth_type": "HTTP_BASIC_AUTH",
                "httpBasicAuthConfig": {
                  "credentialSecret": ZEFIX_SECRET,
                },
            },
        }
)

response = zefix_extension.query("Search for Renuo")
print(response.steps[-1].parts[-1].text)
                

😵‍💫 DEMO

Link: Zefix Extension Notebook

Functions

Functions

Modules that accomplish a specific task

• Model decides which functions to use and when
• Executed on the client side

Functions

API endpoint execution is handled by the client

Functions

• API Layer Control: Execute API calls through middleware or frontend frameworks
• Security & Auth: Handle restricted APIs not accessible to agent infrastructure
• Operation Timing: Manage batch operations and human-in-loop review flows
• Data Transformation: Apply additional processing to API responses
• Rapid Development: Iterate on agent logic without deploying API infrastructure

Example

from vertexai.preview.extensions import Extension

def list_extensions():
    """Provides a list of Vertex AI extensions available
     to the user.

    Returns:
        list[tuple[str, str, str]]: A list of tuples where 
        each tuple contains (in order) the display name, 
        the resource name, and the creation time of an extension.
    """
    extensions_list = Extension.list()
    return [(
        e.display_name, 
        e.resource_name, 
        e.create_time.strftime("%m/%d/%Y, %H:%M:%S")
        ) for e in Extension.list()]
                

Example

model = GenerativeModel("gemini-1.5-flash-001")

user_prompt_content = Content(
    role="user",
    parts=[
        Part.from_text("Which Vetex AI extension do I have access to ?"),
    ],
)

support_tool = Tool(
    function_declarations=[list_extensions_function],
)

response = model.generate_content(
    user_prompt_content,
    generation_config=GenerationConfig(temperature=0),
    tools=[support_tool],
)


for function_call in response.candidates[0].function_calls:
    if function_call.name == "list_extensions":
        api_response = list_extensions()

    response = model.generate_content(
        [
            user_prompt_content,  
            response.candidates[0].content,  
            Content(
                parts=[
                    Part.from_function_response(
                        name=function_call.name,
                        response={"content": api_response},
                    ),
                ],
            ),
        ],
        tools=[support_tool],
    )

    print(response.text)
                

😵 DEMO

Link: Function Calling Notebook

Data Stores

Data Stores

Connecting agents to real-time data sources

• Typically implemented as a vector database
• RAG (Retrieval Augmented Generation) is a common use case

Data Stores

Tools Recap

Summary & Closing

• Agentic AI: Multi-step, self-improving AI that goes beyond chat.
• Core building blocks: Model, Orchestration, Tools, & Learning Loop.
• Practical Use Cases: Customer service, dev ops, marketing, healthcare, etc.

Summary & Closing

Questions? 🤔