Compounding Engineering with AI

Compounding engineering is the practice of ensuring that every unit of work performed makes the next unit of work easier. Instead of just completing a task, you build a "simulation" of your decision-making process into an AI agent, creating permanent operational leverage for the team.

The Compounding Workflow

1. Identify the "Broken Record" Tasks

Look for any task where a senior team member or manager is repeating themselves. Use the "CEO Benchmark": if you have to say the same thing in a meeting twice, it should be codified into a prompt.

• Common triggers: Headline feedback, copy editing for style, PRD formatting, or code style reviews.

2. Codify Your "Taste"

AI models like Claude 3.5 Sonnet or Opus 4 can now act as "judges" of quality if given the right context.

• Record and Transcribe: Record your verbal feedback to a teammate using tools like Granola or Otter.
• Extract Principles: Feed the transcript into an LLM and ask: "Based on this feedback, what are my core principles for [Subject]?"
• Create the Style Guide: Turn those principles into a specific "System Prompt" that the AI can use to evaluate future work.