Compound Engineering vs Loop Engineering

Compound engineering is a way of working with coding agents where every finished task leaves behind lessons that make the next task easier. Loop engineering is designing the loops that run those agents without you prompting each step. Compound engineering is about what the work teaches; loop engineering is about what keeps the work running. The best setups do both.

What compound engineering is

The term comes from Every. Kieran Klaassen started writing about "compounding engineering" in 2025 while building Cora with agents, and he and Dan Shipper laid it out in "Compound Engineering: How Every Codes With Agents": building software with agents so that each completed task makes future tasks easier.

The practice runs in four steps:

1. Plan — agents research and write the implementation plan.
2. Work — agents write the code and tests.
3. Review — the engineer reviews the output and what went wrong.
4. Compound — the lessons go back into the system (prompts, skills, project notes) so the next cycle starts smarter.

Most of the effort sits in planning and review, not in typing code. Every ships it as an open-source Compound Engineering plugin for Claude Code and other agents.

What loop engineering is

Loop engineering took off in June 2026: stop prompting your coding agent by hand and design the loop that does it — a trigger, a job, a verifier, a stop condition, and a point where a human is pulled in.

Side by side

	Compound engineering	Loop engineering
Core question	How does each task make the next one easier?	How does the agent keep working without me starting each run?
Unit	A plan → work → review → compound cycle, usually started by a person	A triggered run with a stop condition, started by a schedule or an event
What compounds	Knowledge: lessons, conventions, review notes written back into the system	Throughput: more runs, in parallel, while you're away
Where the human sits	Planning and review, every cycle	At the escalation point you designed
Typical failure	The compound step gets skipped when you're busy, so nothing compounds	The loop has no real verifier, so it runs a lot and ships little

They need each other

A loop without compounding repeats the same mistakes at machine speed. Compounding without a loop depends on a person remembering to do the review and the write-back, every time. The combination is a loop whose output feeds its own next run: what you decided last time changes what it proposes next time.

The catch: both are methodologies

Compound engineering and loop engineering are both things you practise. You install the plugins, write the prompts, keep the notes current, tune the triggers, and maintain all of it as the codebase changes. That's powerful, and it's a second job.

Where Tekk fits

Tekk is the productized version of both, pointed at one problem: making your product better continuously.

• The loops are already engineered. Twelve of them — Security, Reliability, Backend, Payments, Performance, Testing, React, Code quality, AI engineering, Observability, Alerts, Product analytics — wake on your code changes or Sentry alerts and bring you one proposal per run.
• The compounding is built in. When you decline a proposal, your reason goes into the next run's brief, so that kind of proposal stops coming back. Each loop records which of its methods actually produced proposals you accepted, and leans on those. Each merge is the next change the loops read.
• You keep the review step. Nothing reaches your codebase without your greenlight, and Tekk never writes code — your coding agent builds the approved spec through MCP.

If you already run compound engineering, Tekk's loops are a source of well-scoped work for it. If you don't want to maintain either methodology, it's both, without the upkeep.