Compound Engineering Gets an Upgrade

Join me and Trevin Chow for our third compound engineering camp for paid subscribers next Friday, June 5. We’ll show how planning and building are collapsing into one flow—where you hand your AI a goal and it runs with it. RSVP.

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

In its early days, compound engineering was mostly about the code. I wanted to see if I could get an AI model to make a plan, do the work the way I wanted it done, review the results against my standards, and incorporate lessons from my feedback so it wouldn’t make the same mistake next time. The loop looked like this:

Brainstorm → work → review → compound → repeat

That loop is still the core of how I build Cora. But almost a year after we first coined the term compound engineering, the work phase has become boring—in the best way. If the plan is good and the agent has the right context, it usually does the work right. It writes the code and runs the tests. It fixes the obvious issues. The question now is: “Where do I fit in?”

The answer is at both ends of the process. An analogy my collaborator on the compound engineering plugin, Trevin Chow, came up with is a sandwich. AI is the stuff in the middle. Humans are the bread on either end, holding it together.

At the beginning, I need to decide what is worth building. I need to understand the user, the product, the weird edge cases, and the thing that feels exciting enough to spend time on. Then I can hand the middle to the agent. At the end, I come back in. I click around and look at the design. I read the copy. I ask whether the experience feels right. Sometimes everything technically works, but the product is still not good. So I make it better.

As the models have grown more capable, the original compound engineering loop started to feel incomplete. Plan, work, review, and compound still describes the core engineering cycle, but it leaves out the two places where I now spend most of my attention: before there is a plan, and after the work technically passes review.

So I expanded the loop:

Ideate → brainstorm → plan → work → review → polish → compound → repeat

Ideate and brainstorm are the new front of the process. Polish is the new end. Compound is still the most important step, because the whole point is that every feature should make the next feature easier.

I updated the compound engineering guide to explain the full system. The guide is about engineering, but I think the pattern applies to knowledge work much more broadly. The middle of a lot of work will get automated. But if you want the work to be good, and if you want it to feel like yours, you still need to be there at the beginning and the end.

Kieran Klaassen is the general manager of Cora, Every’s email product. Follow him on X at @kieranklaassen or on LinkedIn.

To read more essays like this, subscribe to Every, and follow us on X at @every and on LinkedIn.

We build AI tools for readers like you. Write brilliantly with Spiral. Organize files automatically with Sparkle. Deliver yourself from email with Cora. Dictate effortlessly with Monologue. Collaborate with agents on documents with Proof.

For sponsorship opportunities, reach out to [email protected].