Intro to AI Agents and Agentic AI

Explore the core building blocks of AI agents, including the environment with sensors, the model, decision-making logic, actuators, execution mechanisms, and feedback for goal evaluation.

Explore how AI agents adapt to dynamic environments by integrating digital data and real-world factors to make intelligent recommendations, from self-driving cars to shopping assistants.

Explore how ai agents collect input from surroundings with sensors, enabling perception and actions, from cameras and lidar to web scrapers and apis.

Explore how AI agents interpret multimodal sensor data with a model that acts as a brain, turning perceptions into informed decisions and real-world actions.

The lecture demonstrates how a decision making logic layer uses structured rules to evaluate model outputs and choose actions, with trading and spam filtering examples.

Actions by AI agents shape the environment via actuators, while the decision logic layer uses an LLM to classify emails as spam and move them to the spam folder.

Agents observe the environment after acting, process fresh input data, and refine their behavior through a continuous feedback loop to improve future performance.

Explore AI agent structures by complexity, starting with simple reflex agents and advancing to advanced designs to help you choose and build the right agent for your objectives.

Explore five defining traits of ai agents—profile and persona, memory, reasoning, actions, and learning capabilities—and see how they shape environments, interact with users, and improve through observation.

The simple reflex agent reacts to immediate temperature perceptions using hard coded rules, like activate heating below 20°C, with no memory, learning, or internal reasoning.

Introduce model-based reflex agents that build an internal world model with memory, using a robot vacuum cleaner to map rooms and obstacles, and to guide cleaning decisions, unlike learning agents.

Watch how goal-based agents search for action sequences and plan to minimize travel time, updating routes as external conditions change, though they may overlook subtleties like user comfort.

Compare and evaluate multiple criteria to maximize expected utility, surpassing a simple goal based approach. Utility based agents weigh return, volatility, liquidity, and default risk to determine overall utility.

Learning agents extend utility-based and goal-based capabilities with a learning element that updates the internal model from feedback, enabling adaptation and self-refinement in unfamiliar environments.

Explore how AI agents acquire knowledge by connecting to external systems. Learn how human knowledge is incorporated into agents to leverage their full potential.

Highlight the human in the loop concept and how humans train, monitor, and refine AI agents in development, deployment, and production. Align with customer satisfaction scores and feedback.

Discover how AI agents learn from other AI agents, external sources like the internet, datasets, web searches, and APIs, with human feedback guiding rapid improvement and sharing solutions across agents.

Distinguish llms ai workflows from ai agents, explore the React and Riwu frameworks, compare single-agent and multi-agent systems, and examine chain-of-thought reasoning that enhances decision making and interpretability.

Clarifies the differences between llms, ai workflows, and ai agents, highlighting agentic versus non-agentic ai. Shows how llms respond to prompts, workflows stay static, and agents act autonomously.

AI agents reason about the environment and then act to schedule meetings, pick a meeting location, check weather, reserve a cafe, and send invitations, all within the ReAct framework.

Riwu enables agents to plan all steps before action, avoiding updates from tool outputs during execution. Compared to React, Riwu sticks to the original plan, but trades flexibility for consistency.

Emphasize single agent systems as a lean starting point for artificial intelligence workflows. A personal assistant example shows how one agent can respond to emails, book venues, and schedule meetings.

Specialized ai agents orchestrate their tasks inside a project workflow, using tool calls and subtasks to tackle complex goals, starting with a single agent and evaluating manager versus decentralized structures.

Explore practical aspects of creating and working with ai agents by selecting a model, choosing tools, configuring prompt instructions, and adding guardrails.

Define clear objectives and ensure high quality data to implement AI agents in your business. Choose an appropriate agent structure, prototype, and monitor KPIs with human oversight.

Evaluate AI models by balancing accuracy, speed, cost, and scalability; prototype with the most capable model, establish a baseline, then experiment with alternatives to optimize efficiency.

Explore how external tools and APIs let agents access data and perform actions within software, using data tools for retrieval and action tools for tasks like scheduling and tickets.

Learn prompt engineering best practices to configure agent system prompts, break down tasks, define clear actions, and anticipate edge cases for reliable, high quality AI agent workflows.

Master zero-shot, one-shot, and few-shot prompting to configure AI agent systems, and compare these techniques with fine-tuning to understand how guidance shapes performance.

Explore chain of thought reasoning as a prompt engineering technique that requires AI to explain intermediate steps, improving transparency and supporting agentic AI design.

Adopt layered guardrails for AI agents, combining data privacy, content safety, misalignment prevention, brand alignment, and accuracy to reduce risk and prevent hallucinations.

Integrate human intervention early in artificial intelligence deployment to catch failures with human oversight. Implement escalation mechanisms that hand control to humans when uncertain.

Evaluate ai agents by balancing accuracy, speed, coherence, cost, and safety to fit their purpose. Prioritize user experience and real-world feedback to improve tone, memory, and contextual awareness.

Explore building ai workflows and agents with n8n, a drag-and-drop automation tool that needs little to no coding, and start from scratch or test a simple ai agent.

Explore the four node types in n8n—triggers, action nodes, logic nodes, and the AI agent node—and how they connect to form a workflow.

Explore the nar den canvas to build node-based workflows that turn chat into automated emails, with memory, Google Sheets contact lookup, and Gmail delivery.

Define the agent's role with a system prompt to act as a personal assistant drafting work emails, using user context and providing a structured greeting, body, closing, and signature.

Connect the AI agent to its brain by linking to an OpenAI API model, create a credential between Naarden and your OpenAI account, and select GPT four.

Add memory to your n8n agent using a simple memory with a five-interaction context window, then verify it remembers user-provided details like the name.

Connect the agent to a contacts database via Google Sheets and to Gmail to draft and send emails. Configure subject and message fields with from AI function to personalize recipients.

Configure an output node with manual mapping and a json field to surface agent responses into chat, using Google Sheets and Gmail, test by emailing Ned at 10:00 Monday.

Explore the infrastructure behind AI agents, including APIs, cloud services, knowledge integration, and deployment, to confidently set up, manage, and scale agent systems.

Connect your AI agent to external LLM services via APIs, selecting providers based on speed, accuracy, and cost. Scale capabilities without building your own model.

Cloud services offer scalable infrastructure for AI agents, letting you pay for compute power as you use it, with AWS, Azure, and Google Cloud Platform supporting GPUs and tools.

Connect AI agents to databases, ERP systems, knowledge bases, and brand guidelines to empower them with 360-degree visibility for reasoning.

Explore frameworks for building AI agents, including Lang chain with chains and agents, Lang Chain Graph, memory and tool calling, and orchestration, Microsoft Autogen, Cru AI, Google ADK, Flow Wise.

Deploy your AI agent from a notebook to a live environment via a chat interface, using AWS Lambda, bedrock SageMaker, and Lex for scalable, reliable web communication.

AI agents offer cost efficiencies by automating tasks and enabling 24/7 customer service, while delivering real-time insights to decision makers and freeing humans for meaningful, strategic work that drives growth.

AI agents work alongside humans, observe, and handle routine tasks faster, while people focus on high-value work. Establish escalation and human oversight to align AI decisions with real world nuances.