Recommended Trending Building Agents with the Claude Agent SDK
Design and ship custom AI agents with the Claude Agent SDK. Build agent loops, define tools, manage memory and sub-agents, evaluate behavior, and deploy multi-agent systems that solve real engineering tasks autonomously.
Oleksii Anzhiiak
Software Architect, Senior .NET Engineer & Co-Founder
By the end you'll be able to
- Build production AI agents on top of the Claude Agent SDK
- Wire agents to MCP servers, custom tools, and external systems
- Reason about agent loops, error recovery, and long-running tasks
- Operate agents safely — guardrails, human-in-the-loop, observability
- Ship the kind of agent that gets paid-tier hiring premium in 2026
Is this course for you?
This is right for you if you…
- You finished #25 (Claude API) or #22 (RAG/Agents) and want the Anthropic-native agent layer
- You're a Senior engineer aiming at AI-engineering roles where 'has shipped agents' is the differentiator
- Your team is being asked to build agents and you want to be the one writing the architecture, not the demo
Don't take this course if you…
- You're new to AI APIs — agents are the deep end. Take #25 (Claude API) and #22 (RAG/Agents) first
- You expect agents to 'just work' on real tasks out of the box — they don't. The course is about engineering them so they DO work
- You don't write production code — agents in production need the same engineering discipline as anything else; without it, they fail loudly
Who teaches this
Oleksii Anzhiiak
Software Architect, Senior .NET Engineer & Co-Founder
Oleksii Anzhiiak is a Software Architect, Senior .NET Engineer, and Co-Founder of ToyCRM.com and ProfectusLab. With over 15 years of experience, he specializes in distributed systems, cloud infrastructure, high-load backend development, and identity platforms. Oleksii designs complex architectures, builds secure authentication systems, and develops modern engineering education programs that help students achieve real career results.
Currently leads architecture for ToyCRM.com — a multi-tenant CRM platform built on .NET by our team. The same patterns and design decisions used there appear directly in the courses: identity & auth, distributed services, code review culture. You learn from engineers actively shipping production code, not from a textbook.
Syllabus
Seven modules to design, evaluate, and ship production-ready Claude agents:
- 1 Module 1: Agent SDK fundamentals — installation, agent definition, system prompts, choosing models for cost/quality
- 2 Module 2: Tool design — schema design, side-effect boundaries, parallel tool calls, validating LLM-produced inputs
- 3 Module 3: Agent loops and control flow — turn limits, stop conditions, recovery from tool errors, human-in-the-loop checkpoints
- 4 Module 4: Memory, context, and sub-agents — working memory vs long-term, sub-agent delegation, parallel branches, context compaction
- 5 Module 5: Evaluation and observability — building evals, golden traces, regression suites, tracing, structured logging, cost dashboards
- 6 Module 6: Multi-agent orchestration — handoffs, shared state, task graphs, deciding when to split vs use a single agent
- 7 Module 7: Production deployment — sandboxing tools, secret handling, rate limit and retry strategy, scheduled and event-driven agents
Prerequisites
Building with Claude API strongly recommended. LLM-Powered Apps or MCP Servers also helpful.
Strong backend experience in Python or TypeScript. Familiarity with REST APIs, async programming, and basic LLM concepts.
What you'll build
You leave with a working AI agent built on the Claude Agent SDK — handling a real multi-step task end-to-end (analyse, plan, execute, verify, report) with proper guardrails. The portfolio piece for the highest-paying AI engineering roles in 2026.
- Claude Agent SDK fundamentals
- Tool definitions and the agentic loop
- Long-running tasks: pause/resume, state, persistence
- Multi-agent orchestration and subagents
- Observability and human-in-the-loop patterns
Where this fits in your career
Read alongside this course
Durable Execution for Agents: The Fifth Discipline Your .NET Background Already Prepared You For
Spec is the truth. Context is the assembly. Evals are the proof. OpenSpec is the operating system. The substrate underneath — the thing that keeps a multi-step agent alive across retries, restarts, and the human-in-the-loop who walks away from their desk at 5pm — is durable execution. Every .NET engineer who's ever shipped a MassTransit saga already understands the pattern; here's how it maps onto serious AI agents in 2026, and why it's the fifth discipline that finally closes the arc.
OpenSpec in 2026: The Operating System for Spec-Driven Development
Six weeks ago I installed @fission-ai/openspec. Yesterday I shipped a 14-file change in 90 minutes from a 200-line spec, in a brownfield codebase three engineers have been editing for two years — no merge conflicts, no review escalation. This is the senior-architect deep-dive on why OpenSpec is the first SDD tool that doesn't collapse under production reality.
Evals in 2026: The Test Suite for Systems That Aren't Deterministic
Your AI feature worked yesterday and fails today. No code change, no prompt change, no model change. That's what life without evals looks like. This is the third leg of the spec → context → evals trinity — and the discipline most teams skip.
First lesson on us. Decide after meeting your instructor.
Sit in on the first session. If after lesson 1 you decide the instructor isn't the right fit, you don't pay for it — and no awkward conversation. (Trial offer applies to courses with more than 5 lessons; this one qualifies.)
Pricing & what's included
What's included
- 14 live sessions × 2 hours each — taught by the instructor, not a recording playback
- Slide deck for every session — yours to keep and refer back to
- Working code files and any data files used in class — cloned to your machine
- Weekly homework with personal code review from the instructor
- Recording on request — give the teacher a heads-up ("can't make Tuesday, please record it") and the session is recorded for you
Frequently asked questions
How much time per week will this take?
Plan for the live sessions plus roughly 1.5–2 hours of practice per session. Most students who finish on schedule put in 4–6 hours a week total. If you put in less, you still finish — it just takes longer.
What if I miss a class?
Tell the teacher BEFORE the session — "I can't make Tuesday, please record it" — and the teacher records that lesson and sends it to you. Recordings aren't a default catch-up archive; they're produced on request when you give a heads-up. After the lesson you do the homework, bring questions to the next session.
What's included in the price?
Live sessions with the teacher; the slide deck for each session (yours to keep); the working code files and any data files used in class; weekly homework with code review; and recordings of the sessions you give advance notice for. Anything beyond that — a certificate, alumni access, mock interviews — is listed explicitly in "What's included" above the FAQ for the courses where it applies.
What if I sign up and the teacher isn't a fit?
For courses with more than 5 lessons, the first session is effectively a trial — if after lesson 1 you decide the teacher isn't a fit, you don't pay for it. We'd rather you walk away after one lesson than push through 9 weeks of bad chemistry. (Courses with 5 or fewer lessons are short enough that the standard pricing applies — the trial offer doesn't make sense at that length.)
Can I get a refund partway through?
Yes, while more than half the course is still unused — i.e. you've attended fewer than 50% of the lessons. The refund covers the unused portion proportionally. Past the halfway point we don't refund, on the assumption that the value has been delivered.
Can I switch the language of instruction?
The live sessions run in the language listed under "Available Languages" above. Slides and code files are typically available in all four (English, Russian, Ukrainian, Georgian). Many students attend sessions in one language and read materials in another — that's normal.
Will I be ready for a real job after this course?
One course rarely gets anyone hired by itself — for any field. What this course gives you is the foundation a junior hiring manager expects: working code you wrote yourself, the vocabulary to read other people's code, and the practice habits that make you employable. The honest answer to "am I ready?" is on our roadmap (link in the page header) — open the level you want to reach and read the "You're ready when" checklist.
Can I pay in instalments?
Yes for courses with a "Monthly Payment" option in the at-a-glance ribbon — usually two or three equal monthly payments. Courses without that option are paid in one go. If the price is the blocker, write to us — we look at every case.
You may also like
Building LLM-Powered Apps: RAG & Agents
Build production-grade AI applications using large language models. Cover vector databases, retrieval-augmented generation (RAG), autonomous agents, tool use, evaluation, and deployment patterns.
Building MCP Servers & AI Tool Integrations
Master the Model Context Protocol (MCP) — Anthropic's open standard for connecting AI models to external tools and data. Build custom MCP servers, expose APIs to Claude, and architect next-gen AI integrations.
Design Patterns in C#: From Theory to Practice
Master classic GoF design patterns in C#. Learn how to apply creational, structural and behavioral patterns to build clean, flexible and maintainable systems.