"Our findings suggest that traditional maintainability principles remain highly relevant in the era of AI-driven development, shaping the computational cost and navigational efficiency of coding agents."
"Our findings confirm that human-friendly code is also more compatible with AI tooling."
"Investing in maintainability not only helps humans; it also prepares for large-scale AI adoption."
> If the code is ugly, but defects are low—does it matter?
> If the code is hard to read, but clients are happy—should I care?
For you who wrote it and are the sole developer, maybe not. But if you want to have other contributors or hand it off, then it can be a problem. New devs joining the project may not want to work with it and will push for a rewrite that will cost money or need to spend extra time on working with code that is hard to work with for them which will cost money. And from the papers above it seems this also affects LLMs as well as they seem to work more efficiently with "cleaner" code.
I didn't see in the post, but have you considered also Btrfs if you are on Linux?
Currently, I use Btrfs on Synology, mainly because snapshots and checksums are nice, but I never really did the work to do benchmarks against other FS. In the past my choice was Ext4 or XFS with LUKS/LVM, then Btrfs matured and since then I moved to it. So far I had no issues, but YMMV, some people consider it still not entirely reliable, but I do backups. I'm setting up a FreeBSD server where ZFS was the natural choice, again snapshots, boot environments, native encryption and just maturity/reliability. I haven't considered OpenZFS yet, mainly because my Synology does not support it, but your benchmarks have some interesting numbers, when I'll have again enough money for some new HDDs I may give it a try.
If a NAS has a 1G or even 2.5G NIC, improving the filesystem performance is a waste of time... the network is the bottleneck. My N100 two-disk btrfs-raid1 NAS hits line rate on a 2.5G NIC, that's all that matters to me.
I use ZFS on my laptops and desktop but kind of considered Btrfs for this new NAS. What made me decide to use ZFS instead is that Btrfs does not have [yet] transparent encryption, I'd need to use dm-crypt/LUKS. ZFS has that natively, so :shrug: When transparent encryption lands on Btrfs, I may consider it for new machines (or Bcachefs if it gets back in the Linux Kernel).
But you can't just hire one, you have to hire functional consultants (who tell you your flow is wrong and you have to adjust that to how SAP does things) and then implementation consultants who don't know how the process works, but can actually implement that integration. And then again after the next release because the integration broke.
And the customer being cheap doesn't pay for the proper modules and thus everything gets mapped to PSP elements -- to keep the same old garbage piles that get pushed around.
I often click accept, at least for site banners that get through ublock. But my browser blocks 3rd party cookies and then it clears all cookies on close (except for trusted sites). I also use private browsing for random sites where I dont bother rejecting cookies usually.
I really think this cookie consent concept should have been a browser functionality, so I can make my default choice for all sites and be done with it.
I also ditched Google years ago for DuckDuckGo, but its not without problems for sure. Often times still full of obviously fake sites in results, that I try to report them. Many times it just returns nothing where Google still manages to give results. And I still have to scroll through their ads when I am on a machine without an adblocker (like Firefox Focus/Klar on Android). But I still rather use them than Google, if I don't find something it is usually not the end of the world and I just move on. Recently, I switched all my browsers to their noai site, on some I still have the lite version, I think.
I was kind of interested in the content, but I am so overloaded with AI slop by now, that reading this generated text gives me nausea.
I was looking to see why they landed on this stack, but there are no alternatives or evaluation criteria listed - given the generated article, I wonder how much of the infra was selected by an LLM.
Claude helped write the article. It is 2026. I proof read it though and yes, giving an LLM a list of specific criteria of what you are looking for in a product is actually a pretty good experience.
If it works for you, it works. I just see the same phrases used repeatedly so frequently nowdays - including my own LLM conversations.
Regarding the use of LLM for picking infra. The issue I usually have with such task is that they frequently omit things - either from the list of options or the features compared. And depending on my familiarity with the topic, I might never notice, which might steer my decision making into a different direction. Basically a certain bias. Sometimes prompting it to repeat reveals more, but ultimately I end up hitting the search and doing my own research, then I might use the LLM again with now more knolwedge and data. Did you run into this too? What was your process?
I do understand what you mean with bias.. some models where quite stubbornly ignoring things like "I want made in EU - not GDRP compliant - not one office or data center in the EU". I remember this being especially painful for TEM and market email providers. Usually they suck at finding the right pricing data at first try.. so I ended up throwing screenshots of pricing pages. Now that I am writing this up, in some instances manually comparing them would have been faster :D ... The bias might come from the huge amount of US dominance in training data and might not even be intentional. In some niches you don't have many options, that's what I tried leaning on in the article.
If that's the case, why do we have to suffer through an AI-generated article? Just give us the prompt.
This topic interests me but I stopped reading as soon as I noticed the slop. I'd much rather read a couple of human-written paragraphs with your personal experience.
Hah, what a coincidence, just started to look into yesterday how do I setup LDAP/OIDC on FreeBSD and today I was going to try FreeIPA or Keycloak. Thanks for sharing.
Awesome, it definitely helps. I realized I have your blog already bookmarked, I subscribed to the RSS feed now as well :) I am new to FreeBSD and these kind of practical articles are really helpful. Thank you very much for sharing your knowledge with others.
Thanks for mentioning this, I am just beginning my FreeBSD journey and wanted to setup a small pre-boot env with mfsBSD[1], didn't know FreeBSD has a tool already to do something like that.
I've seen a few papers recently on the topic in terms of LLM, here is two I found by quick googling
* "Does Code Cleanliness Affect Coding Agents? A Controlled Minimal-Pair Study" https://arxiv.org/abs/2605.20049
"Our findings suggest that traditional maintainability principles remain highly relevant in the era of AI-driven development, shaping the computational cost and navigational efficiency of coding agents."
* "Code for Machines, Not Just Humans: Quantifying AI-Friendliness with Code Health Metrics" https://arxiv.org/abs/2601.02200
"Our findings confirm that human-friendly code is also more compatible with AI tooling."
"Investing in maintainability not only helps humans; it also prepares for large-scale AI adoption."
> If the code is ugly, but defects are low—does it matter? > If the code is hard to read, but clients are happy—should I care?
For you who wrote it and are the sole developer, maybe not. But if you want to have other contributors or hand it off, then it can be a problem. New devs joining the project may not want to work with it and will push for a rewrite that will cost money or need to spend extra time on working with code that is hard to work with for them which will cost money. And from the papers above it seems this also affects LLMs as well as they seem to work more efficiently with "cleaner" code.