How to Make Claude Code Better Every Time You Use It (Full System) | Kieran Klaassen
A 4-step compound engineering system that teaches Claude Code to never repeat the same mistakes
Dear subscribers,
Today, I want to share a new episode with Kieran Klaassen.
Kieran my favorite Claude Code power user and teacher. In our interview, he walked through his Compound Engineering system that makes Claude Code better every time you use it. This same system has been embraced by the Claude Code team and others. Kieran is like Morpheus introducing me to the matrix, so don’t miss this episode 🙂
Watch now on YouTube, Apple, and Spotify.
Kieran and I talked about:
(00:00) The compound engineering loop: Plan, work, assess, compound
(03:40) Live demo: The /workflows command to plan your app
(13:25) How to use Claude Skills as just-in-time context
(25:27) Why you should always ask Claude to ask you questions first
(30:00) Live demo: Using Playwright MCP as your AI QA team
(35:03) Code reviews by AI agents playing security, architect, and more
(40:33) The LFG command: one prompt to go from 0 to production
(46:39) Slash commands vs sub-agents vs skills — when to use each
I’m proud to partner with Granola
Meeting notes are the most useful context for AI in any company. With Granola MCP, you can now access this context from Claude, ChatGPT, and any AI app:
Ask Cursor to create tickets for bugs discussed in standup
Have Claude update your Linear board directly from a planning meeting
Get ChatGPT to draft a follow-up from a sales call and push it to your CRM
I honestly couldn’t work without Granola and the MCP just makes it more useful.
Top 10 takeaways I learned from this episode
Kieran’s compound engineering system makes Claude Code smarter every time you use it. Here's the 4-step loop:
Plan. Sub-agents research your codebase, best practices, and framework versions in parallel before any code is written.
Work. Claude asks clarifying questions first, then builds the feature and writes tests from the plan.
Assess. Review agents check security, architecture, and code quality, then triage findings by priority.
Compound. Capture learnings into docs that Claude reads next time. Simplest version: “add this to claude.md.”
The best part: Kieran packaged this entire system into a free Claude Code plugin with slash commands for each step. Install it here and try it yourself: