Agentic AI Development
- August 26, 2026
- By Bishal Saha
- Intermediate
- 5 min
Agentic AI Development
For intermediate developers who already know the LLM/GenAI basics and want to go deep on what makes a system agentic: planning, tool use, memory, multi-agent orchestration, MCP, evaluation, security, and production deployment. All code is Python.
What you'll be able to build by the end
You'll be able to design and ship an autonomous, tool-using AI agent from scratch — no framework required — then rebuild it on top of production frameworks like the Claude Agent SDK and LangGraph. You'll finish having shipped two full capstone systems: a multi-agent research assistant that plans, retrieves, synthesizes, and writes reports end to end, and a production code-review agent that integrates with GitHub via MCP, runs under guardrails, and requires human approval before acting.
Prerequisites
- Python 3.11+
- Working familiarity with LLM APIs (chat completions, function/tool calling)
- An Anthropic API key (Claude) for the primary code track
- Docker for the deployment chapters in Modules 9, 13, and 14
What this series covers — and what it doesn't
| Dimension | Covered? | Where |
|---|---|---|
| Security | yes | Module 8 (prompt injection, tool-use exploitation, sandboxing, least privilege); applied in both capstones' guardrail chapters (41, 50) |
| Operations | yes | Module 7 (tracing, logging tool calls, cost/latency tracking); applied in chapters 23 and 51 |
| Resilience | partly | Error handling in chapters 7 and 40 (handoffs, error recovery); retries/idempotency in chapter 26. No dedicated chaos/DR treatment — out of scope for a tutorial series |
| Scale | partly | Module 9 (caching, model routing, rate limiting, queues). Covers patterns, not load-tested numbers — any figures shown carry their method per the site's benchmark rules |
| Cost | yes | Token/cost accounting from chapter 1 onward; chapter 27 is dedicated to it |
| Lifecycle | partly | Deployment in chapters 26, 28, 42, 52. Migration/rollback/backup of agent state is touched in chapter 28 but not treated in depth — see the capstone wrap-ups for what a real deployment would still need |
Chapters
Module 1: Foundations: From LLMs to Agents
1. Generative AI Primer for Agent Builders — what generative AI and LLMs are, how tokens/embeddings/attention actually work, and why their limits are exactly what make agents necessary
2. What Is Agentic AI — definitions, the autonomy spectrum, and when agents are the wrong tool