Change to "generating code is cheap", reserve "writing" for the manually written code for the pre-AI era. I think this is a good wording separation.
2. Agentic Engineering Patterns
I must ask to add at least a chapter to be read by agent.
I.e., patterns just to tell agents how human might be working when working with them. Without this, I believe the book's content will be less relevant in 3 months, but with that, it feels a agentic-native book to me.
(this is not try to be cute, we have to write for agents now)
Personally I do think that people with huge amounts of karma contribute to a bit of a groupthink atmosphere, but Hacker News is set up to be more in their favor (which I think makes sense, to get a lot of karma they've been around a while and also understand what gets upvoted around here).
Can we rate limit them a bit?
This submission was not made by simonw themselves, so I'm not sure what the ask is here. I think people ought to be submit to post whatever websites they want? It would be interesting to block simonw's own votes from influencing simonwillison.net submissions, but that would be hard to impose fairly across all webmasters.
It would be interesting to block simonw's own votes from influencing simonwillison.net submissions, but that would be hard to impose fairly across all webmasters.
I think that would be undesirable. Let people vote on themselves as long as they are using their own account and not a second one. If their own vote is all they get, HN will filter them out anyway.
Simon, if you're reading this, I'm a huge fan of so much of your work! Rodney CLI has been a game changer for local development. I have a few open PRs that I've contributed but it doesn't look active. Are you planning on maintaining it or should I fork it?
Long-time reader of simonwillison.net here.
It's a nice guide, but I think it's missing a few points.
Point 1: Software architecture is still a differentiator in the agent era. It is the single measure that separates absolute bu..sh..t from gold. Point 2: LLM/Agent does not suffer from the maintenance burden; the human owner does. So you must somehow "make it *feel* the same suffering" when the codebase starts to become a sh.t pile. Point 3: Agent by default does not align with your goals - they pretend to be - so you need an alignment system to ensure it would do everything to achieve *your goal* (not its own goal).
Comments
A few nits:
1. Writing code is cheap now
Change to "generating code is cheap", reserve "writing" for the manually written code for the pre-AI era. I think this is a good wording separation.
2. Agentic Engineering Patterns
I must ask to add at least a chapter to be read by agent. I.e., patterns just to tell agents how human might be working when working with them. Without this, I believe the book's content will be less relevant in 3 months, but with that, it feels a agentic-native book to me. (this is not try to be cute, we have to write for agents now)
I kinda feel that this person makes a lot of noise in this website. Can we rate limit them a bit?
This person pretty much lives on HN:
https://news.ycombinator.com/user?id=simonw
Personally I do think that people with huge amounts of karma contribute to a bit of a groupthink atmosphere, but Hacker News is set up to be more in their favor (which I think makes sense, to get a lot of karma they've been around a while and also understand what gets upvoted around here).
This submission was not made by simonw themselves, so I'm not sure what the ask is here. I think people ought to be submit to post whatever websites they want? It would be interesting to block simonw's own votes from influencing simonwillison.net submissions, but that would be hard to impose fairly across all webmasters.
I think that would be undesirable. Let people vote on themselves as long as they are using their own account and not a second one. If their own vote is all they get, HN will filter them out anyway.
Simon, if you're reading this, I'm a huge fan of so much of your work! Rodney CLI has been a game changer for local development. I have a few open PRs that I've contributed but it doesn't look active. Are you planning on maintaining it or should I fork it?
summon Simon with a hand drawn SVG of a pelican riding a bicycle
one theory I am thinking through is that people say AI is good at greenfield and has a harder time editing code in a legacy code base.
So maybe you just treat projects as greenfield instead of editing.
Like if you want to make changes to a page you generate a new version of the page instead of editing the old one in place.
Then you keep the old version as fallback if problems come up with the new version.
When code gets complex its much harder to treat each request as greenfield. You would need to rebuild dozens of features into the new page every time.
Post from February;
Large related discussions on Agentic Engineering Patterns:
https://news.ycombinator.com/item?id=47125374
https://news.ycombinator.com/item?id=47243272
Long-time reader of simonwillison.net here. It's a nice guide, but I think it's missing a few points. Point 1: Software architecture is still a differentiator in the agent era. It is the single measure that separates absolute bu..sh..t from gold. Point 2: LLM/Agent does not suffer from the maintenance burden; the human owner does. So you must somehow "make it *feel* the same suffering" when the codebase starts to become a sh.t pile. Point 3: Agent by default does not align with your goals - they pretend to be - so you need an alignment system to ensure it would do everything to achieve *your goal* (not its own goal).
(February 2026)