Agentic Architecture: AI Agent Design Patterns in Python [10/2026, ENG]

Обучающие видео » Компьютерные видеоуроки и обучающие интерактивные DVD » Machine/Deep Learning, Neural Networks

Agentic Architecture: AI Agent Design Patterns in Python

Год выпуска: 10/2026
Производитель: Udemy
Сайт производителя: https://www.udemy.com/course/agentic-architecture-ai-agent-design-patterns-in-python/
Автор: Laurentiu Raducu
Продолжительность: 1h 42m
Тип раздаваемого материала: Видеоурок
Язык: Английский
Описание: This course teaches the architecture that makes an AI agent predictable, affordable, and safe—what comes after the demo, when most agent demos fall apart the first time a real user writes something unexpected. You build one system from start to finish: the support desk of a small online bookshop that reads customer tickets, looks up orders, checks a refund policy, and writes replies, with some tickets simple, some needing judgment, and one order hiding an instruction that tries to trick the agent into issuing a refund. Along the way you learn workflows versus agents and how to choose the simplest shape that works; the agent loop built by hand in about thirty lines of Python, then with the Claude Agent SDK, with turn, budget, and time limits that really fire; tool design with usable descriptions, strict schemas, errors as results, and write tools that validate every call in code; workflow patterns including prompt chaining with gates, routing with structured output, parallel sectioning, and voting; multi-agent patterns like orchestrator and workers, subagents with their own context, and an evaluator that grades another model's work; state and memory covering the context window as working memory, sessions that resume and fork, external memory, and retrieval versus preloaded documents; guardrails including least privilege, human approval for risky actions, hooks as policy and audit trail, and prompt injection through tool results; and evaluation and operations with eval suites graded by code, traces for every run, cost per task, and the failure modes to design around.

Every pattern is a small Python program, and every run in the videos was recorded as it happened—each model call, tool call, result, and cost—nineteen runs in total for seventy-one cents at API prices. When a run went wrong, you see that too: an orchestrator inventing facts because its workers lacked context, a subagent searching ten times for something that isn't there, an evaluator stricter than the policy, and a memory that overruled what the customer just said, with each failure paired with the design rule that fixes it. The complete demo project is included as a download with every pattern as its own module, an offline test suite, an eval suite, and all recorded traces. It was recorded with the Claude Agent SDK 0.2.164 for Python in October 2026, running about 1 hour 40 minutes across 35 lectures of two to four minutes, each with a short recap and many with a quick question.
[spoiler="Содержание"]
1 - Foundations
1. Welcome What You Will Build (02:38)
2. Workflows vs Agents Who Decides the Next Step (02:41)
3. The Augmented LLM Model, Tools, Knowledge, Memory (02:42)
4. The Demo Project A Bookshop Support Desk (02:17)

2 - The Agent Loop
5. The Agent Loop, Built by Hand (02:24)
6. Real Run The Hand-Built Loop Answers a Customer (02:23)
7. The Same Agent with the SDK's Built-In Loop (02:49)
8. Stopping Conditions Turns, Budgets and Final Answers (02:37)

3 - Designing Tools
9. A Tool Is an Interface for a Model (02:50)
10. Schemas and Errors as Results (03:02)
11. Tool Granularity Fewer, Stronger Tools (03:07)
12. Read Tools and Write Tools (03:03)

4 - Workflow Patterns
13. Prompt Chaining with Gates (02:59)
14. Routing with Structured Output (02:54)
15. Real Run Routing Twelve Tickets (02:48)
16. Parallelization Sectioning and Voting (02:55)
17. Choosing a Workflow Pattern (02:43)

5 - Multi-Agent Patterns
18. Orchestrator and Workers (03:21)
19. Subagents and Context Isolation (03:12)
20. Real Run A Subagent Researches the Policy (02:30)
21. Evaluator and Optimizer Generate, Critique, Revise (03:17)
22. What Multi-Agent Really Costs (03:01)

6 - State and Memory
23. The Context Window Is Working Memory (02:54)
24. Sessions Resume and Fork (02:47)
25. External Memory Notes That Outlive a Session (03:08)
26. Retrieval Tools or Preloaded Context (02:50)

7 - Guardrails and Human in the Loop
27. Least Privilege for Agents (03:08)
28. Human Approval with can_use_tool (03:27)
29. Hooks as Policy and Audit Trail (03:07)
30. Prompt Injection Through Tool Results (03:17)

8 - Evaluation and Operations
31. Evals A Test Set with Deterministic Graders (03:10)
32. Real Run The Eval Suite (02:37)
33. Tracing and Cost per Task (02:58)
34. Failure Modes and How to Design Around Them (03:19)
35. The Architecture Checklist (02:43)
[/spoiler]
Файлы примеров: не предусмотрены
Формат видео: MP4
Видео: AVC, 1920x1080, 16:9, 30fps, 1015kbps
Аудио: AAC, 44.1Hz, 128kbps, stereo
[spoiler="Скриншоты"]








[/spoiler]