Agents in practice.
Practical tools and a collaborative workshop that turn agentic ideas into useful everyday workflows.
- Role
- Workshop lead & hands-on engineer
- When
- Ongoing
- Built with
- Claude Code, Cursor, GitHub Copilot, Codex, MCP, Agent skills
I use agentic tools daily for planning, research, refactoring, feature development, and tests. At ROLI, I also lead a collaborative AI workshop where developers discuss ideas and build tools to improve their workflow.
- Runs as a collaborative ROLI workshop
- Applied across planning, code, and tests
- Built around review and verification
How it works.
Define the task
Planning and research come first, so each agent works on a well-scoped problem.
Build with agents
Claude Code, Cursor, Copilot, and Codex, extended with MCP servers, agent skills, and sub-agents.
Review & verify
Code review, tests, and documentation stay central, so faster work is still readable and secure.
The challenge
Useful automation has to fit real engineering work. Faster implementation is valuable only when the resulting code remains readable, secure, documented, and thoroughly reviewed.
What I did
I work hands-on with Claude Code, Cursor, GitHub Copilot, and Codex, alongside MCP servers, agent skills, and sub-agent workflows. The workshop gives developers a place to explore ideas together and apply them to concrete tasks.
Under the hood
The work includes a command-line tool using Claude and GitHub Copilot APIs for sprint ticket creation, account provisioning, and pull requests. MCP servers, agent skills, and sub-agent workflows are part of my broader practice; I keep code review and verification central to the process.
The result
Workshop results include automatically generated CMS image metadata, generated test criteria for tickets, and the identification of performance and refactoring opportunities. Personal AI projects include an automated expense tracker and an educational learning tracker.