Course Outline
From autocomplete to agents: Understanding agent failures
• Anatomy of a coding agent: model, harness, tool surface, context, and permissions
• Positioning of specific tools: Claude Code, GitHub Copilot, Cursor, Codex CLI, and Gemini CLI
• A taxonomy of failures: incorrect context, mismatched tools, lack of feedback, and unbounded autonomy
Demonstration: Executing the same task with optimal versus suboptimal configurations, shown side by side
Context engineering
• Treating the context window as a finite budget: determining what content justifies its inclusion
• AGENTS.md, CLAUDE.md, .cursor/rules, and copilot-instructions.md as varied filenames representing a single concept and source of truth
• Defining conventions, build and test commands, and architectural boundaries
• Balancing retrieval with explicit context; task decomposition and the use of sub-agents
Lab: Author repository context for an unfamiliar Python service, then re-execute a previously failing task to compare results
Reusable workflows and Agent Skills
• Selecting the appropriate abstraction level: instruction files, skills, custom commands, or standard scripts
• Structure of a skill: triggering mechanisms, instructions, bundled scripts, and progressive disclosure
• Ensuring portability across different tools and identifying where vendor lock-in may occur
• Versioning, reviewing, and distributing workflows across a team; addressing common anti-patterns
Lab: Create and test a reusable workflow that enforces a specific house coding standard
MCP: Connecting agents to real systems
• MCP architecture: clients, servers, tools, resources, and prompts; utilizing stdio and HTTP transports
• Identifying high-value servers: Git hosting, issue trackers, databases, browsers, and internal APIs
• Determining when a CLI or script is more effective than an MCP server
• Maintaining tool-surface hygiene: understanding why an excessive number of tools reduces reliability
Lab: Configure MCP servers and manage a ticket end-to-end, covering issue creation, branching, patching, testing, and pull request submission
Feedback loops and evaluation
• Using tests, types, and linters as ground truth for agents; employing test-first work as a control mechanism
• Leveraging CI as an outer loop and maintaining review discipline for agent-generated diffs
• Developing golden-task evaluation sets: defining metrics and detecting regressions
• Treating cost and latency as primary performance metrics
Lab: Construct a small evaluation set and benchmark two different agent configurations against it
Security and guardrails
• Mitigating prompt injection risks via issues, pull requests, READMEs, dependencies, and fetched pages
• Implementing permission models: allowlists, approval processes, read-only tools, and network egress control
• Managing secret hygiene and sandboxing: using containers, ephemeral credentials, and limiting blast radius
• Addressing supply-chain risks associated with third-party MCP servers and shared skills
Lab: Observe an agent being compromised by a poisoned repository, then implement hardening measures to prevent such incidents
Rolling this out to a team
• Defining a staged adoption path; deciding what to standardise and what to leave to individual discretion
• Identifying metrics that reflect genuine value versus those that are misleading
Requirements
• Proficiency in Python, Git, and command-line operations
• Basic familiarity with an AI coding assistant
• NobleProg will provision Dadesktop VMs for participants, pre-installed with Docker, VS Code, and Python 3.11 or newer
• A functioning AI coding assistant of the participant's preference, such as Claude Code, GitHub Copilot, Cursor, Codex CLI, or Gemini CLI. The labs are tool-agnostic, with instructions provided for each supported tool
Audience
• Software engineers, tech leads, and architects who use AI coding assistants but struggle to achieve reliable outcomes
• Platform and developer-experience engineers implementing AI tooling across multiple teams
• Engineering managers responsible for defining standards, guardrails, and success metrics
Testimonials (3)
The trainer is patient and very helpful. He knows the topic well.
CLIFFORD TABARES - Universal Leaf Philippines, Inc.
Course - Agentic AI for Business Automation: Use Cases & Integration
Good mixvof knowledge and practice
Ion Mironescu - Facultatea S.A.I.A.P.M.
Course - Agentic AI for Enterprise Applications
The mix of theory and practice and of high level and low level perspectives
