AI Agentic Engineering Workshop

Overview

Duration: 3 Days

Course time: 9:00 am - 5:30 pm

---------------------------------------------------

Course Synopsis

Coding agents no longer just suggest, they build, refactor, make architectural choices, and commit code. The question has shifted from whether an agent can write software to how we ensure the software it writes is good, safe, and maintainable.

This workshop teaches AI agentic engineering deliberately: designing the context, constraints, and control mechanisms that shape how a coding agent operates within a real codebase. The format blends concept sessions with hands-on paired coding (~60% of workshop time) and reusable artifacts participants take back to their teams.

Key Takeaways

At the end of this programme, you will be able to:

• Apply a systematic mental model for harness design across two axes: Computational vs Inferential, and Feedforward vs Feedback

• Work fluently with the five primary harness mechanisms, CLAUDE.md, MCPs, Skills, Hooks, and Subagents

• Build a working harness system across a real codebase covering architecture, quality, and compliance

• Produce reusable artifacts (harness profile cards, system diagram, packaging strategy) that translate directly to your workplace

• Understand AI agentic engineering as a design discipline scalable across your organisation

Who Should Attend

Developers, tech leads, and architects who are already working with AI coding agents, such as:

• Software Engineers and Senior Developers

• Tech Leads and Engineering Managers

• Solution Architects

Prior experience with Claude Code or a similar agent is helpful but not required.

Prerequisites

• Participants are expected to attend all 3 days continuously

• Working knowledge of Python or JavaScript (cohort language confirmed one week before)

• Familiarity with Git, Docker, and basic CI/CD concepts

• A laptop with administrative privileges and an active Claude subscription (~USD 20 for the training period)

Course Structure

This programme covers the following themes across three days:

• Thoughtworks AI-first sensible defaults and Tech Radar framing

• Anatomy of MCP: trust boundaries, tool manifests, and subagent delegation

• CLAUDE.md and Skills.md authoring for mono and polyrepo environments

• Structured Prompt Driven Development (SPDD) and the REASONS Canvas

• TDD in an agentic loop: red-green-refactor with Claude Code

• AI DevSecOps: Semgrep, Trivy, SAST, and merge gate patterns

• Zero Trust for AI Agents: identity, access scoping, and observability

• Evolutionary architecture and responsible AI policy for teams

What To Bring: 

No physical materials required. Participants should complete environment setup before Day 1: Docker Desktop, Python 3.11+ or Node.js 20+, Git, and an active Claude subscription. Pre-reading links will be shared one week before the course.