HomeGuides › Compound Engineering

Guides · General

Compound Engineering: your method improves on its own


The principle that makes your AI usage improve over time on its own: every lesson is captured and reused, and your system starts each session from a higher baseline.

By Hugo Lahutte· ·~5 min read
  • 1 30-second take
  • 2 The body, visual
  • 3 Go deeper

1. The problem without it

Without a method, every session starts from scratch: you re-explain, you make the same mistakes, you rediscover the same pitfalls. You might go faster today, but you're not building anything.

2. The principle

The term was popularized by Kieran Klaassen and Dan Shipper at Every, while building their email assistant. The idea fits in one sentence: every unit of work should make the next one easier. You do a task twice → you turn it into a tool or a rule. The loop: do → review → capture the lesson → repeat, from a higher baseline.

The lesson enters the context (CLAUDE.md, briefs): next time starts above where this one did.

3. How I apply it

My concrete examples: a deployment that breaks everything → a hard rule engraved in my dev file. An agent that burns its budget → a budget guardrail that's now standard. Each time, the incident doesn't just get fixed: it hardens the system so it never happens again.

4. The cumulative effect

After a few dozen sessions, your context is worth a team that knows you: your tone, your tools, your pitfalls, your past decisions. That's exactly what this Knowledge Hub is for — building in public.

← time / sessionsstarting baseline ↑
Each step annotated with a captured lesson: the system starts higher and higher.

Let's talk

Want to build up, not start over?

If the compound-interest effect on your method interests you, or you want to see how I capture my lessons, send me a message. I document everything in public — that's the whole point.