Your team already uses coding agents, but you are not satisfied with the resulting performance and/or quality. This course helps your team to become proficient at AI-assisted development.

Duration

1-2 days

Technologies

claude

Audience

Software development teams that already use AI coding agents day-to-day, but are in need of improvement

Goal

Learn how to make your team's existing use of agentic AI more consistent, safe, and effective.

Structure

In-classroom or virtual. The entire course is hands-on and based on real-world team workflows

Need this for your dev team?

Outline

This course is for teams that already rely on AI coding agents in their daily work, but need guidance on how to do it well. It focuses on finding where the current way of working breaks down, and establishing shared standards, quality gates, and workflows to fix it.

Agent

This course is taught with Claude Code and Claude CLI by default. But it can be taught with any programming language, development environment and agentic agents.

Prerequisites

Participants should already use a coding agent as part of their day-to-day work, even if informally or inconsistently.

1: Diagnosing the Current Way of Working

Recap: what agentic AI can and can’t reliably do
Discussion: how the team actually uses agents today, and what to improve
Common symptoms of an ungoverned practice: review fatigue, silent quality drift, duplicated effort
Building a shared vocabulary: prompts, context, agents, tools, MCP, skills
Agreeing on scope: which tasks agents may perform autonomously vs. under supervision
Risk categories: correctness, security, licensing/IP, and compliance

2: Standards and Guardrails

Writing and maintaining shared project memory/rules across repositories
Team-wide prompting and workflow conventions
Defining autonomy levels per task type
Sandboxing, permissions, and tool access control
Handling secrets, credentials, and sensitive data safely
Reviewing and approving agent-introduced dependencies

3: Quality Assurance for AI-Assisted Code

Code review practices tailored to AI-assisted changes
Spec-driven development and verifying agent output
Test coverage expectations for agent-authored code
Preventing hallucinated APIs, packages, and behavior
Integrating agent output checks into CI/CD (linting, security scanning, review gates)
Keeping a human in the loop for critical decisions

4: Team Workflows and Collaboration

Pairing with agents vs. delegating autonomous work
Orchestrating subagents on team-sized tasks
Keeping shared context up to date (docs, ADRs, onboarding material)
Version control hygiene for agent-driven changes: commit size, PR structure
Measuring productivity and quality impact over time
Avoiding overreliance and preventing skill atrophy

5: Governance and Continuous Improvement

Turning informal habits into a documented, agreed team policy
Data privacy and confidentiality considerations with third-party models
Keeping standards alive: onboarding new team members, updating shared rules
Assigning ownership: who maintains prompts, project memory, and guardrails
Feedback loops: retrospectives and continuous improvement

Optional topics

Depending on your needs, we can cover additional topics, or replace some of the topics mentioned with others. Here are some suggestions:

Deep dive on Model Context Protocol (MCP) server development
Token management and minimization (see our dedicated workshop)
Security-focused review of agentic AI tool permissions
Building internal evaluation suites for agent-generated code

Improving AI-assisted Development