Appreciation Post to Old School Sysadmins
Posted by StrikingPeace@reddit | sysadmin | View on Reddit | 83 comments
This is an appreciation post to the old-school sysadmins.
You did incredible work… without AI.
You built systems, solved problems, and kept everything running using nothing but manuals, forums, documentation, and raw experience. No copilots. No GPT's. No instant answers. Just skill and persistence.
I consider myself part of the AI-generation sysadmins. I went through college without AI, then it showed up 4 years into my first job—and everything changed.
With AI, I’ve been able to produce scripts, build applications, analyze things, and make decisions in hours that would’ve taken weeks or months before. Things that solve real word or workplace problems. It’s a massive force multiplier. It’s accelerated my impact in ways that are honestly insane.
But that’s exactly why this respect matters and is coming through
Because you did all of that… without AI
You figured it out the hard way. You read, tested, failed, fixed, and mastered your craft from the ground up. And because of that, you built the systems that the world still runs on today.
That’s not normal. That’s elite.
Respect.
Crazy-Rest5026@reddit
Still had Google. Just had to work harder back then lol.
flecom@reddit
I mean, books were a thing, and work labs/ homelabbing was pretty critical to learning... Remember many friends having piles of cisco 2500 routers to learn ccna... Or a bunch.of old Pentium workstations to study for MCSE... Good times
Lonely_Rip_131@reddit
I agree! I’m the youngest in my department and I echo the same sentiments with AI in my current level of productivity. My colleagues actually despise AI and feel they can find the solution faster on their own. I don’t doubt their ability since they all have 20+ years of experience. Nonetheless, AI is allowing me to keep up
lrosa@reddit
When I started there wasn't even Google or Internet: a problem with NetWare 3.12 was your problem :-)
wyrdough@reddit
DejaNews was very helpful when it came around. Before that, I dedicated an hour or two a day to skim relevant newsgroups and hopefully commit some problems/solutions/helpful knowledge to memory. It might not give you the answers, but it could narrow down the problem space dramatically.
Really, though, my particular form of madness was mainly setting up systems and poking things until they broke so I'd have an idea of how to fix them.
Directed fucking around based on vague memories and experience solved so many problems...
Tenshigure@reddit
Back when Slashdot was my homepage and first thing I checked at the start of every work day…
Bob54548@reddit
I miss Novell.
VA_Network_Nerd@reddit
Novel NetWare 3.12 on an Intel 486DX/2-100MHz with 8MB of RAM and a string of 512MB SCSI disks on an Adaptec 2900-series controller.
Go ahead, ask it to manage a domain of 12,000 user accounts spread across 400 departments and 30 file shares and 400 printers.
NetWare is a literal Honey Badger. It will not give a single F.
It will serve up authentication sessions and file data just as fast as that 3Com Fast Ethernet NIC can move packets across the EISA bus.
lankyleper@reddit
I never experienced it myself, but the sysadmin who hired me talks about how awesome NetWare was all the time.
Bob54548@reddit
I remember running my own 3.12 server on a Compaq 386/33 Deskpro with 2 MB of RAM and a 500 MB HDD. I thought I was the Shizz when I upgraded to 4 MB of RAM.
Big-Preparation-1109@reddit
Netware 3.12 OS/2 and compuserve and manuals for support.
lrosa@reddit
Manuals for sure, I had a CI$ account when shared knowledge was already available via FidoNet BBS network.
I am Italian, I stated in FidoNet in 1987, got the possibility to have a gateway in Italy to CI$ in 1992 and it was a long distance call (within Italy)
Big-Preparation-1109@reddit
I also had a fidonet account, never used it for support, but certainly spent many hours on the BBS systems in the 80s.
lrosa@reddit
I was a "point" since my first years in FidoNet and moderator of Italian MSDOS echoser
CharcoalGreyWolf@reddit
Remember AltaVista? I miss it.
Loudergood@reddit
The results were so raw.
wyrdough@reddit
So much better than what came before, but rapidly eclipsed by Google.
DaftPump@reddit
They were the top engine at their time.
Dolapevich@reddit
The before the internet times were interesting indeed.
pdp10@reddit
There was WAIS if you had access.
Searches were quite slow by modern standards, so I had a system that ran them as background jobs on a local workstation or server, and queued up results. Results were usually on anonymous FTP, so it was helpful to have another system to non-interactively fetch human-selected resources.
In the case of Netware, I think Novell had the CNE program by then. It was still the days when doing to a reference book's index was expected to be faster than trawling across millions of douments in the WAIS index, hoping to get lucky. I was typically using WAIS to find papers.
msalerno1965@reddit
What, no grepping newsgroup articles directly on your own UUCP node? Bah.
CNE = Certified-Non-Expert. /s /S /S I SAID!
Anyway, cut my teeth on pdp10's (KA and then KS w/TOPS-10) in the early 80's - high school students across an entire county shared code and bullshit stories. R MAIL[565,11
pdp10@reddit
I didn't have my own NNTP node until later, but good point.
Entire country indeed.
DaftPump@reddit
Many of you probably know of the DR-DOS sabotage Microsoft pulled decades ago. Our shop spent over a day wasting time with that. If the googles were around it would've been a different outcome.
lumpynose@reddit
Forums on Compuserve.
TheGooOnTheFloor@reddit
Back when vendors provided well written documentation that solved about half of the problems you ran into!
gumbrilla@reddit
Yeah, Google turning up was great. Actually the Internet turning up was pretty useful as well come to think of it. Gopher wasn't quite cutting it.
User__234@reddit
“That’s not normal. That’s elite.”
No. Believe it or not it was once normal to work to understand things.
Encrypt-Keeper@reddit
Don’t read too much into that phrase given that it’s the biggest giveaway that this post was written by AI. Not only can OP not do his own job, he can’t even make a reddit post on his own lmao.
Fallingdamage@reddit
Gen Z is the generation of connect-the-dots. They're good at paint by numbers but the world they grew up in didnt press them to think outside the written instructions.
According_Ad1940@reddit
Google wasn't a thing when I started, shit, dial up internet was barely a thing back then
analoghumanoid@reddit
Even when Google did become the predominant search engine, we still had to give it programmatic instructions. what must be included, what to exclude, which sites we're searching specifically. it was a very different prompt back then... there were books on the subject
achristian103@reddit
If you're not still doing all those "old school" things on some level, the AI will replace you eventually.
Yes, you're using AI as a tool, but if you're not also cross checking whatever it spits out as an instant answer by reading documentation, searching the web, etc., then you're doing a disservice to yourself.
SelfImproveAcct@reddit
I mean obviously? It’s a tool lol that’s like assuming people who google something blindly trust the first result
terminal-admin@reddit
If you aren’t cross checking what AI gives you you’re crazy.
I understand Powershell, bicep, az cli, and python enough to look at something that AI gave me and to know if it’s bullshit or not. I could probably write a lot of what I have AI assist me with but it would take me triple the amount of time.
genscathe@reddit
Huh . You use made some assumptions there dude. Thats a bad sysadmin
walks-beneath-treees@reddit
He wasn't making any assumptions, just giving an advice. AI is a tool that makes mistakes all the time.
achristian103@reddit
What assumptions did I make?
lerun@reddit
If you stop practicing/working a skill it will atrofeed to nothing. AI is a dual edged sword and as with everything humanity does, some will use it to grow. But most will just become something less as they will lose the growth applying yourself to learning would have given. Why bother when you can just get AI to do it for you
vogelke@reddit
None. You said if, and your conclusion was right on the money from what I've seen.
WretchedMisteak@reddit
It was a lot of fun, I've still got all my Linux and Windows books.
Trawling through logs is vital no matter the system.
When I was a inix/Linux admin the man command was my go to.
It helped when I moved to Windows management, command line was far more powerful. By that stage we had some decent online documentation.
doglar_666@reddit
In my own personal experience, AI multiplies whatever it is prompted with. Crap in still equals crap out, "force multiplied". Anything of meaningful complexity usually requires more tokens than the LLM offers behind the scenes. And by the time you've deduced how to chunk it to avoid hallucinations, you likely could've Googled, read the docs an implemented the solution in less time. You get the illusion of productivity because LLMs force you to constantly iterate but that action isn't always equivalent to progress.
Ok-Double-7982@reddit
This post reads like AI.
There were so many bad OG sysadmins that AI is now helping those cut from the same cloth.
Tech advances aren't always a bad thing.
Responsible_Ad5216@reddit
I second the OP and as for being replaced by AI...
LLMs are a tool, you still need to read the output, cross check, read the docs, test in dev, then deploy granularily.
But I wouldn't be able to write thousands lines of code a day on multiple problems.
I am able to read or parse them though.
BigCarRetread@reddit
Cross checking is so important. An old-school admin armed with AI is formidabl;e. We can spot the hallucianations and AI will help us with rarely used command line parameter formatting :)
donith913@reddit
I’ve been recently trying to figure out some whack Modern Standby behaviors, and was using Copilot to compare things and I was very up front with my team about what was and wasn’t from Copilot.
Turns out if you give it some data + a hypothesis, it’s just a sycophantic coworker who always tells you your ideas are right and makes up reasons why. Unsurprisingly.
It was decent at helping me analyze CBS.log recently though.
BigCarRetread@reddit
I hear this a lot from Copilot users. I've been using ChatGPT 5.4 with reasoning and it's been good for us. Yes, it can be sycophantic (I suspect this is why it appeals to the PHB's), but it's also been quite useful. I see it as a keen assistant.
ErikTheEngineer@reddit
This is what it's great for...coding and pattern recognition. I've fed it log files, network traces, all that stuff, and the stuff that used to take ne an hour takes a couple minutes to at least get the log distilled down to the area I need to hunt around in.
I honestly think that's the real secret sauce that makes the AI bubble so appealing to everyone. Experienced people with this tool in the toolbelt have a massive productivity improvement. Stupid people see it as a way to tell the computer to do my homework for me and can't tell whether the output is good or bad. Lazy people see "agentic" and see an army of robots to do their tasks for them. I do worry about people leaning too heavily on it and the removal of opportunities to build experience; not having junior engineers who can become senior ones is going to be a painful realization that we'll need to adjust later on.
Responsible_Ad5216@reddit
You are missing out on better models. I have been told I am wrong (my hypothesis is wrong), by Claude Opus and even by ChaGPT this year (the change started around December 2025). The point is, you must use the corporate models and the correct "mode." Each provider (Antrophic, OpenAI) nerfes their models heavily for end users to save money.
I am using Cursor IDE with Claude Opus for most on prem debugging and it is incredible. Of course you still have to manage the chain of thought and make the correct decisions.
donith913@reddit
Yeah, at a certain point I’m bound by the tools and policies of my employer. But Copilot does have access to some reasoning models including Opus as I understand it, but it explicitly forbids use of agentic workflows.
For workflows where you can set an objective outcome and let the LLM(s) riff on that and check their output against unit testing, they seem like they’ve come a long way. But it’s still very easy to get inconsistent output and hallucinations that require iteration on prompts or testing against systems to resolve.
For products where logs are well documented and are easily found online, LLM troubleshooting is possible. For more proprietary tools, unless you include what the log messages mean and tell it exactly what you want it to pull out, it’s gonna just make shit up. Still can be useful if you include “hey if you see this, it means this. Now go find that, timeline it out and tell me what it means and what needs done”, but it’s been hit or miss on whether that’s actually faster than doing it myself, and it’s certainly a LOT of expensive (heavily subsidized for now) compute to do something I can do on my own nearly as quickly.
Fallingdamage@reddit
LLMs are trained on our hard work.
Admins dont do that kind of hard work.
What will future models be trained on? I guess we'll be fine as long as technology never progresses anymore.
meatballwrangler@reddit
wtf is this linkedin, chatgpt ass post. disgusting
coyote1897@reddit
All I can say is thank you!
Mountain-eagle-xray@reddit
AI is cruise control for cool, but you still have to steer.
BearcatPyramid@reddit
Autocomplete on steroids is a "massive force multiplier" that's "accelerated" your impact. LOL.
Who let the MBA into the sysadmin reddit?
SysAdmin127001@reddit
Lol that was written with GPT
Darkchamber292@reddit
100%
boli99@reddit
With AI I've been able to make niche mistakes in scripts faster than ever before, and it's only due to my without-AI experience that I've been able to spot them.
Fallingdamage@reddit
And we documented it and spent time in discussion about problems, what we did to solve them and why. We built wiki's, created repos and discovered new ways to do things.
Then AI consumed it all. Now the next generation just asks for the answers without knowing why.
Now that nobody is actually building good, consistent public documentation anymore, fewer and fewer are having conversations about what they're doing or getting into the weeds of problems anymore. They're just following the recipes without wondering why it calls for a specific kind of flour. The practice of sysadmin continues but the science of sysadmin if drying up.
Instead of today's sysadmins building the documentation and vocally troubleshooting their problems, they ride on the backs of AI trained on our work (that's fine, whatever) but when you're in your 50's, what will your Jr Admins have to work with? You created nothing for them to stand on.
terminal-admin@reddit
I would also consider myself an AI admin. I’m only several years into being a sys admin/cloud engineer, but I have no idea how I would do this job without AI. I think I could but my output would drop severely. When I first started in desktop support there was crappy gpt-3 and I rarely used it. Now AI is used in some fashion every single day as part of my job. I view it as another tool. Almost like a progression of Google search, intellisense, etc.
Suaveman01@reddit
I work with a junior Engineer who is completely dependant on AI. At least 75% of the time he recommends something or tries help, I have to tell him what he’s suggesting is completely false.
DaftPump@reddit
If I could go back to 1980 and learn 8-bit ASM with AI now. :D
justaguyonthebus@reddit
Thank you. It's been an interesting journey. We would always joke about "kids these days" never having to deal with IRQ conflicts or token ring disconnects or high and low memory assignments. But the next generation did just fine.
The technology changed and we changed with it. We did what we had to do. I was always an early adopter and that served me well. I am now an AI admin too and so the journey continues.
weaver_of_cloth@reddit
Big breathing Netscape N.
bobs143@reddit
Yes. Novell Netware. Solving issues Groupwise email. Migration for XP to Windows 7.
ErikTheEngineer@reddit
Also remember that systems were MUCH simpler back then. I started in the late 90s and that was about the end of the well-documented systems with the Wall of Manuals. If it wasn't in the WoM, it couldn't be done according to the manufacturer and a lot of sysadmin stuff was solving stuff that ventured outside that. IBM and DEC were famous for meticulously kept documentation...you could pick up the manuals and learn the system from scratch.
I just cleaned out my basement last weekend and trashed a copy of the Windows 2000 Server Resource Kit, among others. That was the last time Microsoft tried to write a comprehensive man-thousand-page set of product manuals...by that time the internet was in full swing. Wanted to hang onto it but that just sounds like a few steps away from hoarding.
sodiumbromium@reddit
I think so long as you realize LLMs are a tool and not a magical answer box, you'll be okay.
For a lot of problems, they can give good, accurate answers. Always cross vet them, tho.
The problem comes when you're attempting to combine things in a weird way, don't know how to phrase your question correctly, or are off in the weeds on a really strange issue.
Like any tool, it has it has its uses and it's "off label" uses. Think of using a butter knife as a flathead screwdriver. The key to using it well is to know it's limitations and liabilities and being aware of that when you use them.
For instance, i'm attempting to deploy something really silly for testing in an attempt to do some analysis, but for the most part the "in a box" solutions the LLM's are giving me don't work.
So I'm using it as a tool to get an idea of where I need to go, what I need to continue searching for and even if what I want is possible.
For ref, I started, unofficially, back before google, but post www. The tricks that we learned back then are still good ones for sorting information or verification and will never go away. Source validation, test labs, not accepting the first answer you find as the correct one, etc.
I like that the new Gen of sysadmins are using the tools that they have available to the best of their ability! It's great and don't ever let somebody tell you that it's not. You have a tool available that can help? Fucking use it. Don't fight with a hand behind your back!
What I will say is that people need to try other things first....That sounds bad, lemme explain.
Learn to figure out the scope of what you are trying to accomplish, think about it critically and plan it a bit. Exercise your critical thinking and problem solving skills before going to the oracle. Learn the limits of the tools in your toolbox and look to add more.
Those are the real tools of old skool system administration. Being able to piece together the problem, organize it into steps and work on the solution either as a whole or in parts.
I say this because I've worked with people that if Google/LLMs don't have an answer, they're stuck.
_30Harsh_@reddit
It was so nostalgic back then we made our own college's infrastructure
kshot@reddit
When I started, I used to LEARN thru mentoring. There was this guy with 20years expérience, passionnate and probably 160iq, I would learn a lot by checking his configs, setups he did and talking with him. Damn he was good. Working with him helped a lot my career.
Tex-Rob@reddit
48 year old sysadmin who has been a sysadmin more or less since I was 9 years old, that's when I started fixing computers for family and friends. Started building computers and servers at 11.
I'm here to tell you to get rid of that AI slop and actually learn your job. Force multiplier my butt, prove it, with examples. The problem with AI is you don't know what you don't know because it does it all for you. You might flounder on a task I could do in an hour by doing it smarter, but then you find an AI way to do it a convoluted way that takes you 4 hours, and in your mind you are saving 3 hours, when it should have only ever taken an hour. Again, you will never know what you don't know if you choose to lean on AI so heavily.
1stUserEver@reddit
Thanks! I’ll take any appreciation I can get in this place. It’s been a hell of a ride. 25 years. We also had decent vendor support and backing before AI. There is little of that today and I feel for those coming into the sysadmin space with only AI to guide them. AI is making the job more bearable. I find answers in minutes instead of days or weeks. I don’t have to deal with a language barrier on the phone. I get a few hours of downtime instead of constantly grinding to fix problems. Is it worth the toll on the planet? No. Definitely not but I think that will be solved in short order. Those loosing jobs from it really sucks and that is Corp greed. You didn’t want to be with that company. Ai is an Aid not a replacement.
We are in a constant battle against hackers trying to infiltrate our workplace. Is endless and tiring. It’s so different from the virus and malware of decades ago. We can’t get ahead of the next wave of emails that will inevitably trick an end user. there is no lack of job security in what we do. We are valued and essential. Keep up the good work.
sveenom@reddit
E sem YouTube, porque quando comecei YouTube estava no início. Não tinha um vídeo de um Indiano com 200 de QI pra te ajudar
weasel286@reddit
It’s funny to think about the period from the mid ‘80s to the mid ‘90s and how IT worked. How there was no Google, but there were books and Archie, and user groups, and bulletin boards where people would discuss the technical hurdles they faced and come up with solutions. A time when the vendors actually knew their own products and would actually help to implement and provide substantive support. How the system admins oftentimes spent their own money for their own labs at home to tinker with and apply their knowledge at work. Today, some still do that but the knowledge is more accessible and the outcomes are taken even more for granted than ever before by the execs and end users. It’s almost like we had a time of magic, and now that magic has been forgotten or is viewed as folklore…
CharcoalGreyWolf@reddit
These days make me miss DOS.
blissadmin@reddit
I remember physically going to a Microsoft event on a mission from my boss to ask the MS employees questions about the then-brand-new Server 2003 firewall. There was literally no other way to have the discussion we needed to have.
GremlinNZ@reddit
On the other hand, more frequent and much faster updates break things and introduce more bugs requiring more updates.
dpf81nz@reddit
i dunno, i mean i definitely 'googled my way out of problems' right from when i started in the mid 00's until now so i wouldnt say i had no help, AI is next level though
dubl1nThunder@reddit
i started as linux admin rhel 2.1 in 2002. installing rpm's was a real pain, you'd install one and it'd fail because of some missing dependency and this was before google was much of a use. not a lot of discussion boards (dejanews and such) to get help so a lot or the work was trial and error. you cat get that kind of experience very easily now, all the answers are so easy to get for any problem at all. it's crazy how different things are.
Tall-Introduction414@reddit
Ah, yes. Calling the datacenter and asking them to put the redhat CD in the server to resolve dependencies.
graph_worlok@reddit
Dependency Hell - The reason those Debian fanboys were always so smug… Kind of funny yet appropriate that the RPM / Redhat equivalent was created by Apple Mac users 🤣
shimoheihei2@reddit
Its scary to think that people will be growing up now without knowing a world without AI. So many people will be using it as a clutch, not learning how to do things, just relying on the chatbot to do everything for them. We see this happening right now. My one advice for anyone just getting into IT is to learn how systems work properly. There's nothing wrong with using AI to help accelerate your work, but you have to become a subject matter expert first before you can trust the models to give you all the answers.
riders_of_rohan@reddit
AI is only as good as the info fed to it. Blindly trusting AI to write scripts is a fools errand. Use it as a basis and do your due diligence.
PERFECT_7613@reddit
The framing of old-school sysadmins being elite because they built systems without AI is romantic but misses the point. The tools change the leverage, not the underlying skill. A sysadmin today with AI amplifiers can operate at a scale and velocity that was impossible in the manual era. The elite part was always the judgment and the systems thinking, not the manual execution.
No_Kitchen9630@reddit
🥱🥱🥱