AI Agents & Workflows - The Practical Guide

Discover the core building blocks of AI agents and how to provide the right context to an LLM for tasks, with a user-focused view on steering and control.

Understand how developer-provided system instructions shape agent behavior by configuring the system prompt and invisible task-specific directives, balancing helpfulness with safeguards.

Understand how memory in AI agents persists across sessions, including built-in memory in Claude Cowork and manual memory via a memory folder and claude.md or agents.md, with context engineering considerations.

Explore the building part of AI agents and workflows, and learn the key considerations, decisions, and concrete examples across different complexity levels.

Explore building AI agents and workflows with visual no-code tools like n8n or with code using libraries for full control; vibe coding is also discussed.

Decide who builds it and where the agent loop runs. Choose between no-code or coded builds on your machine or a remote cloud, for shared or personal data and workflows.

Explore choosing building blocks for AI agents, including models, system instructions, tools, memory options, and optional skills or sandbox execution, to tailor agents for specific workflows.

Lay the foundation for AI workflows and agents, compare no-code and code, and programmatically use OpenAI APIs and local LLMs, with structured input/output, control flow, human-in-the-loop, and Slack integrations.

Compare no code and code-based approaches to building ai workflows and agents. No code needs no programming knowledge; code-based workflows require basic programming skills and offer language flexibility.

Build ai apps and workflows by integrating large language models via provider APIs and managing app logic. Learn how applications like ChatGPT interact with OpenAI models through APIs.

Compare proprietary models accessible via paid APIs with open-weight LLMs you can run locally using Ollama. Assess memory needs, parameters, privacy benefits, and choose capable options like DeepSeq R1.

Learn to build a lightweight text summarizer workflow in n8n that uploads a markdown or text file, extracts its content, and generates a model-powered summary with a dynamic prompt.

Explore a code-based workflow that uses the OpenAI SDK to summarize text. Inspect the main.py flow: parse input, load source, summarize, and write the result.

Discover a development environment for AI workflows using Python for its simplicity. Learn to edit code with VS Code and interact with models via OpenAI APIs or locally with Ollama.

initialize a new python project with uv init, creating a per-project virtual environment to isolate dependencies, then build a first basic ai-powered workflow starting from main.py.

Set up the OpenAI API, manage credits and budgets, and understand token-based pricing and model selection, while securely handling API keys and exploring alternatives like Google Gemini.

Build a simple python workflow that takes user input, calls a large language model via the openai api, and outputs a ready-to-share x post.

Learn to send http requests to the OpenAI api using the modern responses endpoint, including json payloads, authorization headers, and secure api key handling with python-dotenv.

Explore how to choose and use an AI model by weighing task complexity against model price, from 4o-mini for summaries to O3 or GPT-4o for advanced planning.

Explore prompt engineering for specialized workflows, assign roles like a social media manager, and craft concise, readable posts with dynamic topic input and topic XML tags.

Learn to extract and use an LLM response by navigating its JSON structure, accessing the output, content, and text fields, and building reusable workflows with prompt engineering.

Learn to send requests to the OpenAI API to interact with a large language model, understanding the workflow and consulting official docs for roles, chat history, images, and audio.

Learn to call the OpenAI API from Python using the official OpenAI SDK, install the package, create a client, and access responses via the output_text field.

Explore how few-shot prompting enhances AI-powered workflows by using example-based prompts, delimiters, and iterative prompt engineering to generate more engaging posts.

Master dynamic prompt generation by outsourcing examples to a separate json file, loading and parsing it, and injecting those examples into your prompt to keep it shorter.

Build multi-step, multi-model workflows that fetch website html, extract the core content with a language model, then summarize and generate an X post.

Compare multi-step workflows and agentic systems, noting AI agents are autonomous programs that create and execute plans with instructions, tools, and language models, while some steps may themselves be agents.

Learn to run open large language models locally with Ollama, download Gemma 3 models, manage memory and context windows, and use multi model workflows for extraction, summarization, and post generation.

Enforce structured outputs in ai workflows by defining a json schema, extracting text from PDF invoices with the PyPDF package, and guiding the model to return data in that shape.

Master structured outputs with the OpenAI SDK and Pydantic in Python by defining an invoice schema to parse responses and insert data into a database.

Explore a multi-step workflow for generating a blog post from an outline, merging it with example posts in markdown, and delivering raw markdown output via an enhanced prompt.

Generate images in a workflow with the OpenAI image API and GPT Image 1 by prompting, setting size and format, and saving the thumbnail next to the outline.

Master configuring workflow steps to run sequentially, in parallel, conditionally, or repeat steps within AI agent workflows, handling dependencies and refining results.

Manage the control flow of workflows and agents by looping article draft generation, evaluation with feedback, and up to three cycles, then run thumbnail and LinkedIn post creation in parallel.

Add a human in the loop to guide AI workflows, allowing input and confirmation during execution, with options to accept, override, or request new drafts.

Learn to integrate external services such as Slack into your workflows, authenticate with APIs via tokens, and add human-in-the-loop checks before sending Slack notifications.

Explore building fully autonomous ai agents and agentic systems by exposing tools to large language models, and connect multiple agents with memory management for short- and long-term memory.

Build an AI agent in n8n using a chat trigger and the AI agent node, connect a model like GPT-5.6 Sol, and attach tools for summarizing text and creating PDFs.

Explore how an ai agent in n8n uses two tools to summarize text and create pdf, orchestrating prompts, models, and data in a visual workflow.

Run the n8n agent to orchestrate tool calls, summarize text with the summarize text tool, and optionally generate a pdf in a practical workflow demonstration.

Explore building a task-specific ai agent in n8n by configuring the system message with instructions, tools, and optional memory, and by passing extra information to tailor behavior.

Explore moving from task-specific to general agents by leveraging core building blocks—model, tools, instructions—and adding general features like agent skills, memory, and AGENTS.md-based guidance for versatile task handling.

Explore a generic agent project with AGENTS.md based startup context and user instructions. It uses tools like read, write, edit, and bash, and stores memory as a markdown file.

Demonstrates running a general agent from a project folder, loading skills and a report template, analyzing a csv, and generating an html report.

Analyze an example sandbox implementation and explore sandboxed AI workflows with humans in the loop and permissioned tool calls, using a sandbox folder to restrict reads and writes.

Explain how to expose tools to a large language model, craft prompts that request tool use, and let the AI application invoke real functions while the model only generates tokens.

Build a tool-aware agent from scratch that uses a getTemperature tool and prompt engineering to prompt a GPT-4o model and execute the function, demonstrating tool use.

Learn to expose tools as functions to a large language model with OpenAI's function calling, and manage tool calls in a Python chat agent workflow.

Build a versatile AI agent that autonomously uses multiple tools to perform customer service tasks, such as verifying customers, retrieving orders, checking refund eligibility, and issuing refunds, in a database.

Build reusable artificial intelligence components with a base tool class and specialized tools, then create a customer service agent and register multiple tools for flexible workflows.

Build a multi-agent system that creates a refined research plan with user input, stores plans in a SQLite database, and performs a web search with a final summary.

Demonstrates a multi-agent workflow with a web search agent deriving terms. A summary report agent then crafts a concise report from Brave search results.

Compare universal and specialized agents in ai workflows, showing how collaboration among specialized agents yields deterministic results and reduces trust risks versus a single all-purpose agent.

Explore how AI agents manage memory, distinguishing short-term memory for a session (chat history) and long-term memory stored in databases for future use across agents.

Discover how to build AI-powered workflows by equipping tools and describing intent to call them, while recognizing that workflows and agents blur but automation remains the goal.

Explore building AI agents and workflows from scratch in Python, wiring AI models with tools and prompts, and using LangGraph, LangChain, and SDKs to simplify development.

Explore popular libraries and frameworks for building AI agents, including the Vercel AI SDK, LangGraph, Pydantic AI, CrewAI, and the Eve framework, with two example frameworks for Python and TypeScript.

Compare libraries and frameworks for AI-powered workflows, using the Vercel AI SDK as a library for OpenAI and Anthropic, versus opinionated frameworks like CrewAI and Eve for building agents.

Explore CrewAI as another framework for building AI agents, comparing it with Eve and highlighting how a non-file-system design offers different approaches to solving agent problems.

Learn to get started with CrewAI for building AI agent workflows, including installing uv, creating a research_crew project, and selecting an OpenAI model like GPT-4o; compare with LangGraph and LangChain.

Explore how the CrewAI framework builds a crew of agents, with researcher and a reporting analyst registered via decorators, configured from agents.yml and tasks.yml, and dynamically populated with topic placeholders.

Define tasks for multiple agents with a tasks.yaml, including placeholders and an expected output, and see how CrewAI funnels results between agents and writes to report.md.

Install the CrewAI tools package and expose Brave search via the tools parameter, using BRAVE_API_KEY in .env to enable integrated web search and reporting.

Run a crew with CrewAI.run, customize dynamic inputs in main.py to drive topic-based research and generate a report. Then compare SDKs like CrewAI or LangGraph with building your own solution.

Build your own ai workflows and agents by applying the course concepts to real problems, using autonomous agents with multiple tools or multi-step ai workflows to solve tasks.