Abstract

AI Coding is one of the most FOMO inducing “recent” (2-3 years) developments affecting software engineering. My goal is to create a compact, hands-on workshop for learning how to use the most recent tools and more importantly - how to slightly reduce that feeling that things are moving at lightspeed without you.

Over the past 3 years I have found myself changing tooling multiple times. So my goal is to not only teach what the “latest and greatest” is - as that can change fast - but rather to adapt people’s workflows into ones that can allow such fast changes.

Practically speaking - in this 1.5-2.5* hour-workshop I will present how I work using Cursor and Claude Code. My goal is to cover initial set-up all the way to advanced tooling such as subagents and skills, explain how they interact, how they can use eachother, and how to create your own.

The workshop is “code-first”: participants are going to be using their own laptops on a shared codebase that has the basics ready for learning and enhancements.

Target Audience

  • Backend, data, and ML engineers who want to level up their AI coding skills.
  • Participants should optimally already have Cursor/VSCode installed but we can also do that part together.
  • A shorter, no-hands-on is also an option.
  • Works well for groups of 5-15; this allows for a “sit down” experience.

Prerequisites

Sent ahead of time: install Cursor + Claude Code, Node, gh, clone the shared repo, set API keys.

Syllabus

Part 1 - Foundations (~30 min)

  • Cursor tour; picking the right model (Haiku/Sonnet/Opus, thinking models)
  • Context engineering: CLAUDE.md, @file references, project vs user memory
  • Claude Code - what it is, what it does, how it differs from Cursor

Part 2 - Tooling That Compounds (~45 min)

  • How I work with Claude Code?: honest review of how I work
  • Planning vs execution: plan mode, interview-style prompting, /handoff between sessions
  • MCPs and CLIs (gh, Linear) as first-class tools
  • SKILLS.md and subagents - tour the provided .claude/, then open the hood on one skill and one subagent
  • Hooks and scripts - automating the boring parts

Part 3 - Problems (~20 min)

  • Git hygiene with agents: worktrees, commit discipline, reviewing diffs before accepting
  • Evaluation and guardrails: catching “LGTM drift” and silent wrong answers
  • Security issues and how to best avoid them

Part 4 - Capstone: the self-improve skill (~30 min)

One exercise that threads everything together. Participants build a skill that, at session end, reviews what they just did and proposes updates to their own CLAUDE.md.

What You Walk Away With

  • A working self-improve skill committed to your own dotfiles
  • A mental model for when to reach for Cursor vs Claude Code vs a subagent vs a hook
  • A workflow that survives the next tooling cycle, not just this one

*Timings are approximate; a precise structure/syllabus will be decided depending on final duration.