A working demo cannot tell you whether an agent is safe to operate. This course makes you implement the controls, test the failure paths and assemble evidence a production review can inspect.

View the course on Leanpub

Leanpub’s normal pay-what-you-want price is $49 USD minimum, $149 suggested, before applicable tax. The course is available now; Leanpub displayed 92% published when checked on 22 September 2026. Check the storefront for the latest publication status and price.

Read the free CC-001 lesson and practical lab on Leanpub. This is a written sample, not a sample video, worksheet download or assessment. Scroll to "Free sample" on the storefront.

What you will build

The course has eight modules and 31 numbered lessons, with written instruction, worked examples, practical labs, assessments and capstones. Supporting video accompanies the course; a video is not promised for every lesson.

The final deliverable is a production-readiness dossier for an agent system. It records the agent loop, tool and permission boundaries, model-routing policy, context controls, evaluation cases, observability plan, recovery procedure and release evidence. A reviewer should be able to follow what happened and decide whether the system is ready.

The curriculum covers:

  1. Foundations of a production agent
  2. The Claude Code primitive stack, including MCP and hooks
  3. Owning the runtime with the Claude Agent SDK
  4. Securing and governing the agent boundary
  5. Evaluating agent behaviour before and after release
  6. Observability, reliability and cost
  7. Scaling to a team-owned agent platform
  8. Remediating anti-patterns and planning the migration

Who should take it

Software engineers, technical leads, platform engineers, solution architects and security engineers who are moving beyond a successful experiment. You should be comfortable with terminals, Git, configuration files, tests and normal software-delivery workflows. Prior Claude Agent SDK experience is not required.

Start with the free CC-001 written lesson and practical lab on the storefront if you want to see the teaching style before buying. You can complete that sample lab using a blank document as your Agent Loop Canvas.

About the instructor

Thomas De Vos is an architect and technical author with 25+ years of experience, a Google Cloud Generative AI Leader and a Google Cloud Authorised Trainer. His teaching focuses on engineering decisions, practical controls and the evidence needed to operate AI systems responsibly.

Course, book and purchase terms

The companion book is a separate, optional purchase. It is not included in course enrolment and is not required. The course adds hands-on work and assessments rather than narrating the book.

Leanpub offers a 60-day refund policy, subject to its terms and course exclusions. A course is not refundable after you generate its certificate or fail its last or sole permitted attempt. Read Leanpub’s current refund policy and the attempt limits on the storefront before purchasing.

See the full curriculum and enrol on Leanpub.

For the engineering argument behind the course, read the launch article.