# As agents flood the codebase, one good pattern can improve thousands of changes

Original title: Shaping the Loops, Closing the Loops

> Hacker News story 50553223 · language: en · submitted 2026-11-18
> Web version: https://zeli.app/story/50553223
> About this feed: https://zeli.app/llms.txt · Agent guide: https://zeli.app/skill.md

▲ 16 · by [based0x10](<https://news.ycombinator.com/user?id=based0x10>) · 💬 5
[original](<https://revyl.com/blog/pattern-engineering/>) · [HN discussion](<https://news.ycombinator.com/item?id=50553223>)

## Summary

Hayden White argues that engineering leverage is shifting toward two disciplines: verification engineering, which closes the loop by proving whether a change meets its requirements, and pattern engineering, which shapes the loops by curating the conventions, abstractions, and context agents learn from. Because agents repeatedly extend whatever patterns already exist, a good one compounds across thousands of changes while a bad one keeps steering work wrong. His example: revup, the CLI that sets up every engineer's and agent's machine at Revyl from a single service definition.

## Key quote

> A good pattern can improve thousands of subsequent changes, while a bad one can repeatedly steer work in the wrong direction.

## HN discussion (5 comments)

### [bombela](<https://news.ycombinator.com/item?id=50595735>)

> In summary: Removing friction for an increase in efficiency.

### [anamhira](<https://news.ycombinator.com/item?id=50553297>)

> Huge move to working at higher level of abstractions to have the ability to have more leverage.

### [cootsnuck](<https://news.ycombinator.com/item?id=50596019>)

> I keep seeing people rediscovering tried and true software engineering principles and saying AI makes them "matter more"... And I'm a bit skeptical of that framing.
>
> Things like platform engineering, DevOps, code quality, linting, CI/CD, QA, etc. were just as foundational before coding agents as they are now (regardless of never being called "pattern engineering"). I'd argue what's changed is less their importance than the consequences of neglecting them.
>
> I think a lot of this is how we as an industry are coping with the realization that the narrative we've been sold over the past few years, that AI was going to render software engineering a "solved problem", isn't actually true.
>
> So instead of saying "Hm, it looks like being able to arbitrarily generate massive amounts of plausible code has exposed just how much of software engineering was never about writing code in the first place"... We say "Wow, these amazing AI coding agents are so powerful we're having to formalize new levels of abstraction for engaging with software engineering!"
>
> Idk, maybe I'm being a yum yucker. People can call stuff whatever they want. Just irks me when there's plenty of prior art, knowledge, instruction, etc. that has run into these problems years and years ago that may be more useful to center than coining new shibboleths.

---

## Cite

- Page: https://zeli.app/story/50553223
- Date: 2026-11-18
- Source: revyl.com (https://revyl.com/blog/pattern-engineering/)
- Hacker News: https://news.ycombinator.com/item?id=50553223

## Other languages

- zh: https://zeli.app/zh/story/50553223 · markdown: https://zeli.app/zh/story/50553223.md
- ja: https://zeli.app/ja/story/50553223 · markdown: https://zeli.app/ja/story/50553223.md
- ko: https://zeli.app/ko/story/50553223 · markdown: https://zeli.app/ko/story/50553223.md
- de: https://zeli.app/de/story/50553223 · markdown: https://zeli.app/de/story/50553223.md
- es: https://zeli.app/es/story/50553223 · markdown: https://zeli.app/es/story/50553223.md
- ru: https://zeli.app/ru/story/50553223 · markdown: https://zeli.app/ru/story/50553223.md
