Feedback on using AI/LLMs for Xymon development and reviews
list Bruno Manzoni
Hi everyone, Here is my feedback on working with AI/LLMs for Xymon: - We no longer code by ourselves; LLMs do it for us. The real skill now is guiding the AI, designing the architecture, and validating the result. LLMs write code better than even experienced C developers, though skilled individuals—including good developers—are still essential to direct them. - We still need to write a specification, like an RFC. - We still need to do reviews to catch problems that a good specification or tests might miss. All this changes the meaning of being a "code author" and shifts how a team should be built. Let me know what you think! Cheers, Bruno
list Kris Springer
I agree that the days of hand coding without AI assistance is quickly fading into the past. Due to this and the years-long waiting for Xymon to get upgraded, I have used AI to build my own modern monitoring server from scratch (including agents), and fully abandoned Xymon. Please remove me from the Xymon mailing list, and also I can be removed from the Github contributor list. Thank You, Kris Springer Systems Admin I/O Network Administration
▸
On 9/10/26 3:02 AM, Bruno Manzoni via Xymon wrote:Hi everyone, Here is my feedback on working with AI/LLMs for Xymon: - We no longer code by ourselves; LLMs do it for us. The real skill now is guiding the AI, designing the architecture, and validating the result. LLMs write code better than even experienced C developers, though skilled individuals—including good developers—are still essential to direct them. - We still need to write a specification, like an RFC. - We still need to do reviews to catch problems that a good specification or tests might miss. All this changes the meaning of being a "code author" and shifts how a team should be built. Let me know what you think! Cheers, Bruno
list Jaime Kikpole
Kris, the unsubscribe directions are at the bottom of every message sent through the list.
Jaime Kikpole Director of Technology |
▸
On Thu, Sep 10, 2026 at 10:55 AM IO Support via Xymon <xymon@xymon.com> wrote:
I agree that the days of hand coding without AI assistance is quickly
fading into the past.
Due to this and the years-long waiting for Xymon to get upgraded, I have
used AI to build my own modern monitoring server from scratch (including
agents), and fully abandoned Xymon.
Please remove me from the Xymon mailing list, and also I can be removed
from the Github contributor list.
Thank You,
Kris Springer
Systems Admin
I/O Network Administration
On 9/10/26 3:02 AM, Bruno Manzoni via Xymon wrote:
> Hi everyone,
>
> Here is my feedback on working with AI/LLMs for Xymon:
>
> - We no longer code by ourselves; LLMs do it for us. The real skill
> now is guiding the AI, designing the architecture, and validating the
> result. LLMs write code better than even experienced C developers,
> though skilled individuals—including good developers—are still
> essential to direct them.
> - We still need to write a specification, like an RFC.
> - We still need to do reviews to catch problems that a good
> specification or tests might miss.
>
> All this changes the meaning of being a "code author" and shifts how a
> team should be built.
>
> Let me know what you think!
>
> Cheers,
> Bruno
>
> xymon@xymon.com
> To unsubscribe send an email to xymon-leave@xymon.com
xymon@xymon.com
To unsubscribe send an email to xymon-leave@xymon.com
list Scot Kreienkamp
Agreed, I think the days of writing original code are mostly at an end, unless you are one of the lucky few working on the AI itself. And that may be short lived as well.
RFC/specifications are excellent as they convey the intent behind the code, which often isn't available or understandable by looking solely at the code. AI written code definitely needs tested for bugs or unintended pathways. One of the prompts I try to give AI is to document the code with comments. That seems to work pretty well for the most part, but my AI did recently put some comments in places that made the code fail to work, which is a rookie mistake.
RFC/specifications are excellent as they convey the intent behind the code, which often isn't available or understandable by looking solely at the code. AI written code definitely needs tested for bugs or unintended pathways. One of the prompts I try to give AI is to document the code with comments. That seems to work pretty well for the most part, but my AI did recently put some comments in places that made the code fail to work, which is a rookie mistake.
Scot Kreienkamp | Principal Applications Infrastructure Architect | 1 La-Z-Boy Incorporated
(XXX) XXX-XXXX | X-XXX-XXX-XXXX | user-9678697f1438@xymon.invalid
One La-Z-Boy Drive | Monroe, Michigan 48162 | la-z-boy.com
facebook.com/lazboy | instagram.com/lazboy | youtube.com/lazboy
▸
-----Original Message-----
From: Bruno Manzoni via Xymon <xymon@xymon.com>
Sent: Thursday, September 10, 2026 5:03 AM
To: xymon@xymon.com
Cc: Bruno Manzoni <user-81df5df13c04@xymon.invalid>
Subject: [Xymon] Feedback on using AI/LLMs for Xymon development and reviews
Hi everyone,
Here is my feedback on working with AI/LLMs for Xymon:
- We no longer code by ourselves; LLMs do it for us. The real skill now
is guiding the AI, designing the architecture, and validating the
result. LLMs write code better than even experienced C developers,
though skilled individuals���including good developers���are still essential
to direct them.
- We still need to write a specification, like an RFC.
- We still need to do reviews to catch problems that a good
specification or tests might miss.
All this changes the meaning of being a "code author" and shifts how a
team should be built.
Let me know what you think!
Cheers,
Bruno
To unsubscribe send an email to xymon-leave@xymon.com
From: Bruno Manzoni via Xymon <xymon@xymon.com>
Sent: Thursday, September 10, 2026 5:03 AM
To: xymon@xymon.com
Cc: Bruno Manzoni <user-81df5df13c04@xymon.invalid>
Subject: [Xymon] Feedback on using AI/LLMs for Xymon development and reviews
Hi everyone,
Here is my feedback on working with AI/LLMs for Xymon:
- We no longer code by ourselves; LLMs do it for us. The real skill now
is guiding the AI, designing the architecture, and validating the
result. LLMs write code better than even experienced C developers,
though skilled individuals���including good developers���are still essential
to direct them.
- We still need to write a specification, like an RFC.
- We still need to do reviews to catch problems that a good
specification or tests might miss.
All this changes the meaning of being a "code author" and shifts how a
team should be built.
Let me know what you think!
Cheers,
Bruno
To unsubscribe send an email to xymon-leave@xymon.com
This message is intended only for the individual or entity to which it is addressed. It may contain privileged, confidential information which is exempt from disclosure under applicable laws. If you are not the intended recipient, you are strictly prohibited from disseminating or distributing this information (other than to the intended recipient) or copying this information. If you have received this communication in error, please notify us immediately by e-mail or by telephone at the above number. Thank you.
lazboy_2024_inc_navy_4a4d68ec-613a-4141-a2aa-d73a2ae749f6.pnglist Roland Rosenfeld
Hi Bruno! On Do, 10 Sep 2026, Bruno Manzoni via Xymon wrote:
Let me know what you think!
Maybe I'm too old and pessimistic, but I don't like the new workflow, where IA generates tons of code and comments, that I have to read and review. I understand that this may be the future, but the frequency that it generates code doesn't match my performance of code review. So I'm completely ambivalent: On the one side I'm very happy, that xymon is alive and updated, on the other side I have not much fun reviewing the AI generated stuff. Maybe that will change at some point in time, but at the moment I personally didn't reach this point. Maybe it's also a problem, that I'm not a C developer. I can read and understand C (and write small fixes), but this usually takes much time. So it looks that the AI outclasses me, which doesn't feel good. That said, the project needs more C developers that are able to review the AI generated stuff... Except this, I still ask myself what all this AI generated code means to the LICENSE/Copyright of the project. COPYING file is GPL-2, most source files are Copyright Henrik Storner <user-e180c691dbbd@xymon.invalid>, but what about the code generated by AI? At least the tests reference GPL-2+, but no copyright or author is mentioned. So can this really be GPL-2+ without a copyright owner? I'm not sure about all this... Greetings Roland
list Michael T Pins
▸
Roland Rosenfeld writes:
Maybe I'm too old and pessimistic, but I don't like the new workflow, where IA generates tons of code and comments, that I have to read and review. I understand that this may be the future, but the frequency that it generates code doesn't match my performance of code review.
Maybe it's also a problem, that I'm not a C developer. I can read and understand C (and write small fixes), but this usually takes much time. So it looks that the AI outclasses me, which doesn't feel good.
That said, the project needs more C developers that are able to review the AI generated stuff...
I did another round of testing chatbot generated code a few months ago. It was all still, at best, junior level programmer grade code. Not only was the code non-portable (it wouldn't even compile on FreeBSD), but it had numerous use-after-free errors, and a variety of other bugs. All in about 500 lines of code. Personally, I find I'm still better off writing the code myself rather than asking a chatbot then having to audit the code and clean up the mess. This isn't actually surprising - despite what far too many people believe, there is no intelligence in an "AI" chatbot. The code a chatbot generates will never be any better than the code it was trained on, and they were all trained primarily on what they scraped from GitHub. Considering how bad much of the GitHub code is, expecting an LLM trained on it to produce quality code is almost laughable.
▸
Except this, I still ask myself what all this AI generated code means to the LICENSE/Copyright of the project. COPYING file is GPL-2, most source files are Copyright Henrik Storner <user-e180c691dbbd@xymon.invalid>, but what about the code generated by AI? At least the tests reference GPL-2+, but no copyright or author is mentioned. So can this really be GPL-2+ without a copyright owner?
I'm not sure about all this...
Considering how an LLM works, there is a non-trivial chance it will generate code at times that is line for line identical to what it was trained on. Who owns the copyright on that code? All that said, if someone finds using a chatbot to generate code helpful, and someone is willing and able to fully audit the resulting code, go for it.
list Bruno Manzoni
Nice
to see some of us are still around!
I
think there’s definitely truth in what you’re saying, but that’s
not the whole story. To me, LLMs are tools—"stupid" tools in the
sense that they don't actually understand what they're doing, they
just predict the next token. They will never replace us humans!
But
they do pack an enormous amount of knowledge, which makes them
awesome partners. You don’t need to be an absolute genius; you
just pair up with an LLM, and boom—you've got a winning combo. Of
course, we still need to learn how to tame the machine, understand
its limits, and figure out how to team up with it. That’s actually
one of my goals while playing around with Xymon.
Xymon’s
challenges are complex. Solving these puzzles is way harder than
building something small from scratch. You quickly run into the
"context" problem: your prompt or the LLM doesn’t have the whole
Xymon codebase under its hood, and you can’t just dump all of it
in there at once. At some point, the AI loses the big picture. A
human can keep track of it (even if we tend to forget things too
sometimes!).
Still,
this is my playground, so come along and play with me and the
Xymon code! Don’t think for a second that past developers never
made mistakes—trust me, they made plenty! And they didn't have the
full picture either. Some features are redundant, and some code
breaks other things. That’s just the reality of development, so
come join the fun!
Per
our policy, we always say it takes two people to agree to push
some code. I’m still fully convinced this rule is super powerful.
Unfortunately, I’ve been pushing a ton of PRs lately simply
because there’s no one around to review them. Because of that, I
produce a lot of code and have gotten pretty good at making it
clean, but I'm honestly not as sharp when it comes to reviewing
others' work, just because I rarely get to practice it.
Let’s
be real: porting existing features isn’t always the most thrilling
task, and doing code reviews to test things isn't exactly the most
exciting adventure of your life. But I’m sticking to the plan for
now! We started implementing Terabithia, and I really want to see
this through. All the improvements in that branch are great ideas
that solve real-world problems. It’s pure gold.
I’ve
already explored IPv6, TLS, APIs, CMake builds, better probes,
improved graphing, custom tests, and even hooking things up with
LLMs. So yes, there is plenty of cool stuff to do—come join the
ride!
Bruno
▸
Le 10.09.2026 à 22:08, Michael T Pins
via Xymon a écrit :
Roland Rosenfeld writes:Maybe I'm too old and pessimistic, but I don't like the new workflow, where IA generates tons of code and comments, that I have to read and review. I understand that this may be the future, but the frequency that it generates code doesn't match my performance of code review.Maybe it's also a problem, that I'm not a C developer. I can read and understand C (and write small fixes), but this usually takes much time. So it looks that the AI outclasses me, which doesn't feel good.That said, the project needs more C developers that are able to review the AI generated stuff...I did another round of testing chatbot generated code a few months ago. It was all still, at best, junior level programmer grade code. Not only was the code non-portable (it wouldn't even compile on FreeBSD), but it had numerous use-after-free errors, and a variety of other bugs. All in about 500 lines of code. Personally, I find I'm still better off writing the code myself rather than asking a chatbot then having to audit the code and clean up the mess. This isn't actually surprising - despite what far too many people believe, there is no intelligence in an "AI" chatbot. The code a chatbot generates will never be any better than the code it was trained on, and they were all trained primarily on what they scraped from GitHub. Considering how bad much of the GitHub code is, expecting an LLM trained on it to produce quality code is almost laughable.Except this, I still ask myself what all this AI generated code means to the LICENSE/Copyright of the project. COPYING file is GPL-2, most source files are Copyright Henrik Storner <user-e180c691dbbd@xymon.invalid>, but what about the code generated by AI? At least the tests reference GPL-2+, but no copyright or author is mentioned. So can this really be GPL-2+ without a copyright owner?I'm not sure about all this...Considering how an LLM works, there is a non-trivial chance it will generate code at times that is line for line identical to what it was trained on. Who owns the copyright on that code? All that said, if someone finds using a chatbot to generate code helpful, and someone is willing and able to fully audit the resulting code, go for it. xymon@xymon.com To unsubscribe send an email to xymon-leave@xymon.com
list Bruno Manzoni
Hi Kris, Thirteen years, and you were one of the fourteen owners of this organisation - you could have pushed whatever you wanted, whenever you wanted. I wish you had. The wait was real, and it is on all of us who held the keys and waited for someone else to do it. You were one of us - which is why I am sorry to see you go. Your seat is removed. The door stays open. Cheers, Bruno
▸
Le 10.09.2026 à 16:55, IO Support a écrit :I agree that the days of hand coding without AI assistance is quickly fading into the past. Due to this and the years-long waiting for Xymon to get upgraded, I have used AI to build my own modern monitoring server from scratch (including agents), and fully abandoned Xymon. Please remove me from the Xymon mailing list, and also I can be removed from the Github contributor list. Thank You, Kris Springer Systems Admin I/O Network Administration On 9/10/26 3:02 AM, Bruno Manzoni via Xymon wrote:Hi everyone, Here is my feedback on working with AI/LLMs for Xymon: - We no longer code by ourselves; LLMs do it for us. The real skill now is guiding the AI, designing the architecture, and validating the result. LLMs write code better than even experienced C developers, though skilled individuals—including good developers—are still essential to direct them. - We still need to write a specification, like an RFC. - We still need to do reviews to catch problems that a good specification or tests might miss. All this changes the meaning of being a "code author" and shifts how a team should be built. Let me know what you think! Cheers, Bruno