Claude agent engineering: Master Claude Ecosystem in Seven Weeks

Claude Agent Engineering: Master the Claude Ecosystem in Seven Weeks

Learn to use Claude Code, Cowork, Skills, and loops effectively, then build and demonstrate your own agent system.

                                                                                   Seven-week live cohort · 30 participants · $250 USD
                                                                                   3 January–14 February 2027
                                                                                   Sundays, 6:00–9:00 pm EET · Tuesday office hours, 7:00–8:00 pm EET

This cohort develops two connected capabilities: using Claude’s tools effectively in your daily work and building agent applications you can control through code.

We begin with Claude Code and Cowork, covering their setup, instructions, permissions, and practical workflows. You’ll then study Agent Skills from the fundamentals and learn how to design loops that execute, verify, retry, and stop. The later weeks focus on programmatic development with the Claude Agent SDK and managing longer workflows through context engineering.

Sessions combine technical explanations, live demonstrations, and practical exercises, supported by weekly office hours.

Learn through a shared project—and build your own

Throughout the cohort, I’ll develop a deep-search agent as a running example: a system that plans research, searches and reads sources, organizes evidence, identifies gaps, and produces reports with citations.

The early modules demonstrate relevant instructions, Skills, and workflow patterns. The main application build takes place during the Agent SDK and context-engineering sessions.

You’ll develop your own project in parallel. Choose a specialized research agent or another suitable workflow, apply what you learn as the cohort progresses, and demonstrate your system in Week 7. You’ll receive the shared project’s source code and can bring questions about your own implementation to office hours.

The seven-week curriculum

Week 1 · 3 January — Use Claude Code effectively

Set up Claude Code and learn how to work with it across everyday development tasks. We’ll cover how it uses your repository and context, how to give it clear instructions, and how to review its work.

Topics include installation and configuration, project structure, persistent guidance, implementation and debugging workflows, permissions, tool access, and session management.

You’ll leave with: A configured development environment and a practical approach to using Claude Code in your own projects.

Week 2 · 10 January — Organize and run work with Claude Cowork

Learn how to give Cowork a task, provide the material it needs, guide execution, and evaluate the result.

We’ll cover global instructions, task-specific guidance, multi-step workflows, scheduled tasks, recurring work, and using Skills. You’ll also work through access permissions and review points so you can decide where human involvement is needed.

You’ll leave with: A Cowork workflow with clear instructions, expected outputs, and a process for reviewing its execution.

Week 3 · 17 January — Build Agent Skills from scratch

This module assumes no previous experience with Skills. We’ll begin with what they are and when to use them, then build and test a Skill step by step.

Topics include SKILL.md, descriptions and activation, progressive disclosure, reference material, scripts, output requirements, and packaging. We’ll also investigate Skills that fail to load, activate for the wrong requests, or produce inconsistent results.

You’ll leave with: A reusable Skill you have built and tested, plus an approach to improving and maintaining it.

Week 4 · 24 January — Design loops: execute, verify, retry, stop

Explore how agents continue working across multiple steps and determine whether another iteration is useful.

We’ll cover loop structure, state, verification, completion criteria, retries, recovery, and stopping conditions. Practical examples will show how to identify repeated work, handle failed actions, and introduce human review when needed.

You’ll leave with: The ability to design and test a loop with explicit behavior for progress, failure, and termination.

Week 5 · 31 January — Build with the Claude Agent SDK

Move into programmatic agent development using Python. The deep-search agent becomes our main implementation example as we connect research tasks to tools and manage execution from code.

Topics include agent instructions, custom tools, external integrations, sessions, runtime state, execution events, permissions, and error handling. You’ll apply the same foundations to your own application.

You’ll leave with: A working programmatic agent that uses tools and executes its core workflow.

Week 6 · 7 February — Engineer context for longer workflows

Apply four context-engineering operations:

Write → Select → Compress → Isolate

We’ll examine how to store useful state outside the active context, retrieve information for the current step, compress growing histories, and isolate large outputs or separate workstreams. We’ll apply these techniques to the deep-search agent and assess their effects on results and usage.

You’ll leave with: An explicit context-management approach for your project and a system ready to demonstrate.

Week 7 · 14 February — Demonstrate your project and receive feedback

Present the agent you have developed during the cohort. Show the problem it addresses, run a representative example, and explain the main implementation decisions.

Each participant will have a short demonstration slot with feedback. We’ll discuss results, failure cases, limitations, and practical next steps, followed by closing Q&A.

You’ll leave with: Feedback on your system and a clearer direction for continued development.

What your enrollment includes

• Six 3-hour live teaching sessions
• One 3-hour project demonstration and feedback session
• Seven 1-hour group office hours
• 28 scheduled live hours in total
• Lifetime access to the recorded technical sessions
• Private Slack group throughout the cohort
• Technical documentation and complete slide decks
• Source code for the shared deep-search agent
• Implementation examples, prompts, commands, and exercises

Office hours provide a shared space for troubleshooting and discussing implementation choices. Allow additional time between sessions to practice and develop your project.

Who should attend

This cohort is for software engineers, AI engineers, data scientists, and technical professionals who want to use Claude’s tools more effectively and build their own agent applications.

You should be comfortable reading and modifying Python, running terminal commands, installing dependencies, and working with basic APIs and JSON. Previous experience with Claude Code, Cowork, Skills, or the Agent SDK is not required.

Bring a laptop with Python, Git, and a code editor, along with access to the Claude products used in the sessions and your own API access. Use non-sensitive material for exercises and project work.

Schedule

• Live sessions: Sundays, 3 January–14 February 2027, from 6:00–9:00 pm EET.
• Office hours: Tuesdays, 5 January–16 February 2027, from 7:00–8:00 pm EET. The final office hour provides follow-up support after the demonstrations.

If you miss a technical session, you can catch up through its recording and accompanying materials.

About your instructor

Youssef Hosni is an AI engineer at Solita, with a research background in generative AI and agentic systems at Aalto University. He also run To Data & Beyond, where I publish technical resources and teach practical AI engineering workshops.

This cohort brings together material I have developed and taught individually, with time to cover the tools in depth and support your implementation. My previous Agent Skills workshop sold out all 30 seats.

Enrollment and refund policy

Enrollment is $250 USD, with places limited to 30 participants.

API usage costs are not included. You are responsible for charges incurred while building and testing your agent.

Refund requests must be submitted at least one week before the course starts. The refund deadline for this cohort is 27 December 2026.

Join the cohort starting 3 January to develop your Claude skills, build your own agent, and demonstrate your work on 14 February.