AI-Assisted Development Essentials
Your team has the licences. This is where the engineering discipline begins. A half or full day of hands-on exercises in a real codebase — building the habits, commands, and workflows that make Claude Code a serious part of how your team writes software, not just a smarter autocomplete.
Three skill areas, built in a day
Everything is hands-on and practiced on a real codebase. No slides to take home — just new muscle memory.
Work with Claude Code, not around it
The foundational habits that separate engineers who get consistent results from those who get lucky ones.
- How LLMs actually work. Tokens, context windows, probabilistic output — the mental model that lets you reason about when Claude will and won’t be reliable.
- Prompts that hold up. Specificity, few-shot examples, structured output, chain-of-thought. Techniques that produce consistent results, not just a single good run.
- The engineering loop. Slash commands, custom skills, safe refactoring with Claude proposing and you reviewing. How to navigate an unfamiliar codebase in minutes, not hours.
Connect Claude to your actual tools
Claude Code is more useful when it can reach the systems your team already works in.
- MCP in practice. Configure an MCP server so Claude can read PRs, linked issues, and comments — the context it needs to give a real review, not a surface-level one.
- Reusable commands. Build a
/review-prcommand that applies your team’s review logic automatically, every time. - Spec-driven development. Take a feature from first conversation to implementation plan before writing a line of code.
Make it safe and consistent for the whole team
Individual habits are the start. A team harness is what makes AI-assisted development sustainable.
- The AI harness. CLAUDE.md for team standards, permissions to control what Claude can touch, hooks that enforce behaviour automatically — the three components that make Claude Code consistent for everyone, not just whoever configured it first.
- The attack surface. Understand the risks that come with agentic tools and how hooks and permissions work together to keep Claude’s behaviour within safe boundaries.
- Automated enforcement. How to make team standards enforceable rather than just advisable — and how to package the result so any team in your organisation can adopt it.
From first prompt to team-wide discipline
How LLMs process text, why the same prompt doesn’t always produce the same result, and the prompting techniques that hold up in production. The foundation everything else builds on.
Your engineers work through a realistic codebase, using Claude Code to navigate, refactor, write tests, and prep PRs. They build custom commands and skills that encode their own workflows — the habits and tooling that survive contact with a real Monday morning.
MCP servers let Claude call external tools directly — your version control, issue tracker, whatever your team already works in — without you copying and pasting context in manually. This step introduces the concept and has your engineers configure and use one from scratch.
CLAUDE.md, permissions, and hooks working together as a harness — so Claude Code behaves the same way for everyone on the team, and team standards are enforced rather than just suggested.
Engineering discipline, not a chatbot tutorial
Your engineers work in a real TypeScript codebase with real bugs to find and fix. Not a slide deck with screenshots of Claude doing clever things — a session where they build the muscle memory themselves.
Most AI coding workshops are about prompting. This one also introduces MCP, team governance, and the security concerns that come with agentic tools — so your engineers leave with a map of the landscape, not just a starting point.
The workshop ships with its own codebase, but if your team would rather work on something familiar, we can adapt it. Either way, what they build — commands, skills, the team harness — is theirs to keep.
Your team is already paying for the licences.
Whether they’re using Claude Code as a chatbot or not using it at all, this is the fastest way to change that. Book the workshop, or start with a short intro call.