Ado Kukic's Advent of Claude

Table of content

Ado Kukic is a full-stack developer with over 20 years of experience building web software. Before joining Anthropic, he led developer relations at Auth0, MongoDB, DigitalOcean, and Sourcegraph. He’s a Google Developer Expert for Web Technologies.

At Anthropic, Kukic runs developer relations. In December 2025, he shared one Claude Code tip per day on X/Twitter and LinkedIn, then compiled them into Advent of Claude on his blog adocomplete.com.

The 31-Day Series

Kukic’s tips cover the full spectrum of Claude Code usage:

Category	Examples
Getting Started	`/init` for project onboarding, `CLAUDE.md` for memory
Thinking Modes	`think`, `think hard`, `ultrathink` for controlling reasoning depth
Slash Commands	Custom commands, project-specific workflows
Subagents	Delegating tasks to specialized workers
Plugins	Extending Claude Code with community tools
Autonomous Loops	Long-running agents with minimal supervision

Key Tips

/init — Let Claude Onboard Itself

Instead of manually writing documentation for Claude, run /init and let it read your codebase to generate its own onboarding docs:

claude
> /init

Claude creates a CLAUDE.md file with project context, conventions, and commonly used commands.

Thinking Token Control

Kukic clarified how to control Claude’s reasoning depth:

Keyword	Thinking Tokens
`think`	~4,000
`think hard`	~10,000
`ultrathink`	~31,999

Include the keyword anywhere in your prompt: “ultrathink about the best architecture for this feature.”

Custom Slash Commands

Create reusable commands for repetitive workflows:

# .claude/commands/test-and-commit.md

Run the test suite. If all tests pass, commit with a descriptive message.
If tests fail, fix the issues and try again.

Invoke with /project:test-and-commit.

The Ralph Wiggum Plugin

One of the more experimental patterns Kukic documented: autonomous loops where Claude runs for extended periods with minimal human intervention. Named “Ralph Wiggum” by Jesse Vincent, this pattern lets Claude stress-test features or iterate on implementations autonomously.

From Tips to Guide

The compiled guide reorganizes the daily tips into a learning path:

Beginner Essentials — Setup, basic commands, project configuration
Intermediate Patterns — Custom commands, subagents, permissions
Advanced Techniques — Plugins, MCP servers, autonomous workflows

Kukic added context that couldn’t fit in 280-character social posts, including code examples and explanations of when each feature is most useful.

Community Impact

The series spread quickly. A dev.to summary by Damien Gallagher brought the tips to a wider developer audience. The guide was shared across LinkedIn, Twitter/X, and Threads.

Josh’s YouTube video “His Claude Code Workflow Is Insane” referenced Boris Cherny’s workflow (Kukic’s colleague at Anthropic) and many of the same patterns from Advent of Claude.

At Anthropic

As a Developer Relations engineer at Anthropic, Kukic has direct access to Claude Code’s development. His tips reflect both official capabilities and practical patterns discovered through extensive use. He also presents at conferences like the AI Coding Summit, where his talk “Agentic by Default” covered autonomous developer workflows.

His role involves helping developers get value from Anthropic’s tools. Advent of Claude is that job made public.

Key Takeaways

Tip	Why It Matters
Use `/init` early	Claude writes its own context
Control thinking depth	Save tokens or go deep when needed
Build custom commands	Automate your repetitive workflows
Try plugins	Community extensions add capabilities
Explore autonomous patterns	Unlock longer-running agent work

Links

Next: Bas Nijholt’s Agentic Coding Journey