TL;DR: Cora general manager Kieran Klaassen has written prolifically about compound engineering, his philosophy of software engineering for the AI age. In this piece, based on a camp he gave for paid subscribers a few weeks ago, we get an inside look at how exactly Kieran builds with the compound engineering plugin for the first time. He walks through, step by step, the process of going from a single prompt to a working app in under an hour. If you’ve been curious about how to build with compound engineering, this is the piece to read.—Kate Lee

Was this newsletter forwarded to you? Sign up to get it in your inbox.

This time last year, any time Kieran Klaassen opened a new session in Claude Code, he started from scratch. The lessons from his past code reviews, the style preferences he’d painstakingly explained, and the bugs he’d already flagged—Kieran remembered them all, but from the machine’s perspective, it was like it had never happened.

He’d been building Cora, Every’s AI email assistant, and getting tired of copy-pasting the same prompts, correcting the same overengineered tests, and flagging the same bugs. “A human would remember,” Kieran said. “The AI wouldn’t.”

So he decided to create a system that would remember—one that plans before it codes, reviews outputs to enforce his taste, and stores every lesson so the AI applies it next time. The result is what we now know as compound engineering, a signature approach to coding with AI where every bug, fix, and code review makes the system permanently smarter. The official compound engineering plugin has more than 10,000 GitHub stars and is used by a growing community of builders, including engineers at Google and Amazon, who say it changed how they think about software.

At our first Compound Engineering Camp, Kieran walked subscribers through the full loop live, building an app from a one-line prompt to a working product in under an hour. Below is the workflow as Kieran demo-ed it, plus what it means for how software gets built from here.

Key takeaways

1. Brainstorm before you plan. The plugin has a brainstorm step that interviews you collaboratively and fills the gap between your vague idea and a detailed spec.
2. Planning should run without you. Once the requirements of the project are clear, the plugin has a plan step that researches your codebase, checks for existing patterns, surfaces past learnings, and produces an implementation plan with zero additional input needed.
3. Use different models for different steps. Kieran uses faster models—such as Claude Haiku 4.5 or Gemini 2.5 Flash—for brainstorming, Opus for planning, Codex for implementation, and sometimes Gemini for code review.
4. Compound when the context is fresh. The plugin’s compounding step stores lessons as artifacts that future agents can discover, the core of compound engineering. Run it right after something breaks or works—before the AI compacts your conversation and you lose the specifics of what you were talking about.

Write at the speed of thought

That gap between your brain and your fingers kills momentum. Monologue lets you speak naturally and get perfect text three times faster, and your tone, vocabulary, and style are kept intact. It auto-learns proper nouns, handles multilingual code-switching mid-sentence, and edits for accuracy. Free 1,000 words to start.

Download Monologue for Mac

Want to sponsor Every? Click here.

The compound engineering loop

A founder who does everything themselves hits a ceiling, Kieran says. The ones who scale are ...

Become a paid subscriber to Every to unlock this piece and learn about:

1. Why Kieran argues that writing code is only 30 percent of the job
2. The fourth step most AI coding workflows skip entirely
3. How Kieran built a working app in under an hour from one prompt, running 25 agents in parallel