AI Agents Foundations
By Wanderson Leandro de Oliveira
AI Agents Foundations is the on-ramp to this engineering track: not a collection of frameworks to memorize, but the mental model and technical groundwork you need before you can design a real agent. You will start by pinning down what an agent actually is — how it differs from a chatbot and from a deterministic workflow, the perceive-reason-plan-act-observe-reflect loop, levels of autonomy, and honest use cases and limitations. From there the course builds the backend skills agentic applications lean on hardest: HTTP and webhooks, async processing with queues and idempotency, and streaming responses over SSE and WebSockets — because LLM calls are slow, expensive and occasionally fail, and your architecture has to expect that. A dedicated Python module takes you from typed, Pydantic-validated data through an async LLM client, ending in a hands-on lesson where you build a resilient client with retries, backoff, timeouts and streaming. The next two modules cover LLM fundamentals engineers actually need — tokens, context windows, generation controls, open versus closed models — and model selection: choosing by cost, latency and capability, building gateways with fallback and cascading, and matching workload to model size. The course closes with prompting and structured outputs applied specifically to agents, culminating in a capstone where you design a full system prompt and Pydantic output contract, including error cases, for a support-ticket triage agent. By the end, you will have the vocabulary, the backend instincts and the working code patterns to start building real agents in the courses that follow.
Course content
What Is an AI Agent
- 🔒 Agent vs. chatbot vs. workflow text
- 🔒 Autonomy levels and the agent loop at a glance text
- 🔒 Use cases and limitations text
Backend Foundations for Agentic Apps
- 🔒 HTTP, REST and webhooks for agent backends text
- 🔒 Async processing, queues and idempotency text
- 🔒 Streaming responses: SSE and WebSockets text
Python for Agent Engineering
- 🔒 Typed Python with Pydantic for agents text
- 🔒 Building an async LLM client text
- 🔒 Hands-on: a resilient LLM client with retries text
LLM Fundamentals for Engineers
- 🔒 Tokens, context windows and inference text
- 🔒 Generation controls: temperature, top-p and structured output text
- 🔒 Open vs. closed models and licensing text
Model Selection and Routing
- 🔒 Choosing a model by cost, latency and capability text
- 🔒 Model gateways, fallback and cascading text
- 🔒 Small vs. large models and workload classification text
Prompting and Structured Outputs
- 🔒 Prompt engineering for agents text
- 🔒 Structured outputs with JSON Schema and Pydantic text
- 🔒 Capstone: design a prompt and output contract for an agent text