So I know this post is going to be somewhat controversial, but I don’t really care because I think we all need to start realizing this. I think it is time people started accepting that Pandora’s box has been opened, and we are now stuck with the reality of living with A.I. (Artificial Intelligence). While I think we are starting to get past the hype, and many industries are starting to realize it isn’t useful for their industry, there are tons of industries it has and will now most likely forever be a part of. The one I will be covering in particular is the one that affects my industry: general information technology.
Impacts in Operations
So this is where I currently make my living. I am, by title, the UNIX/Linux Systems Engineer Technical Lead. However, I do way more than that, helping out on other teams and such. In this role, at my workplace, we are encouraged to leverage internally approved AI tooling to assist us, with guardrails ensuring the tooling cannot touch the systems or know their specifics. I think that is a fair trade-off, and from what I have seen, there have been mixed experiences among folks on my team using the tools.
With the exception of a couple of people on the team not wanting to use A.I. at all (which, I cannot blame them, I am on the fence myself), we have about what I would consider three groups of users that I will define as follows:
- Hesitant Skeptic: Very skeptical of its usefulness in doing Systems Administration and Engineering tasks. Will work with it reluctantly, and take excessive caution in how well they trust the responses. Almost always, double-check and research responses using “old-fashioned” methods like web searches, blogs, and forums.
- Researcher: Leans on using A.I. for doing research on new topics they are asked to work on that they haven’t done much work with before. They will take the data they get and add it to their own research. Basically, using it as a teaching aid to get better help in understanding poorly written documentation and things of that nature.
- Slightly Trustworthy Oracles: This group tends to use A.I. not just for research but also to guide them in building out their plans, structuring their work based on feedback, and leaning heavily on its suggestions. However, due to the good forced education my work instilled in A.I. usage, they will tend to take their formed plan and double-check it against other AIs to see if there is consensus. It does lead to someone’s mass hallucination, as some A.I. tools are prone to agreeing when still wrong.
So personally, I started out as the Hesitant Skeptic, and in many ways, I still tend to lean on doing the research myself. However, over time and trying different tooling and building out stuff on my own to better understand how A.I. works, I have to say I am now closer to using it as what I defined as Researcher… but still take a lot of time to double check my research on my own but it has been good because some of the tooling I leverage now makes sure it sights all sources so I can quickly review what it read to come to conclusions. This has significantly sped up the completion of certain work tasks, on or before plan.
This, however, leads me to my concerns with those on the team who have migrated to the Slightly Trustworthy Oracles… because they end up setting up lots of calls and meetings with me and others on the team to help them double-check their different A.I. responses when auditing each other. It has made me see that using it that way is causing the harm I have been reading about a lot: we are offloading most of our critical thinking and self-researching skills because it just answers what we ask (right or wrong) and takes the critical thinking out of the mix. So I think this type of use case should be limited or even just kinda removed from the equation for at least now. I cannot mandate that people not do it this way; I can only encourage them to reach out to me less for critical thinking and provide them with resources they can use to make more headway on their workloads.
Impacts on Development
This one opinion I have, I believe, will be the most controversial of all my thoughts in this post. That is because, I think it is a great tool for helping developers do more in way less time, but before you start ranting at me, please hear me out. I am not saying that just prompting all your code to get it written for you by A.I., I am suggesting that you use it as a more complex auto-complete, debugging helper to find that stupid one line issue you keep glazing over, and finally just asking how to do a thing that you just cannot find a clear answer on to help you understand how to impliment it yourself.
So that is a lot to say in my first paragraph of this section, but I stand by pretty much everything I said in that post. Because, in honest reality, I have tried, for the sake of job security, to leverage it and see how using the different levels of it in my programming work would affect me in day-to-day things. With that said, I think I have come away with some key takeaways from my own experiences.
True Vibe Coding: Good, Bad, and Ugly…
I think it isn’t a horrible idea to fully vibe-code some personal projects that will never see public deployment, but rather for your own use cases. Personal quick tools and scripts are a great example of needing a quick-and-dirty tool or script that can parse your downloads folder the way you like. Proof of concepts for yourself, like a simple, quick version of an app that lets you organize your favorite written fan-fiction for easy search and reading later. Those are what I think are fine, as long as you never use them for an actual deployment where you put stuff you really care behind it.
The reason I say ugly is that… while it can be fine for those personal tasks and proofs of concept stuff… it will be bad because you didn’t write any part of it. So if you take that base yourself and have A.I. vibe the rest of it out… when you start needing to dissect it and understand how or why certain things happen… You are now at the mercy of the tooling you used to build it, unless you are really skilled at reverse-engineering what it built for you. This is why I think it can get ugly if you rely on it for anything more than quick scratch work.
Assisted Development
The way I have found myself using it is less for writing code and more for helping me understand and deal with the roadblocks I hit. As an example, I was struggling to get Better-Auth working in a project I was trying to set up, and it was because I wasn’t as familiar with SvelteKit’s hooks as I was with just using straight-up calls to code from the React world. It helped me understand the nuanced differences. I think leveraging it in ways like this is super helpful, and saves time in trying to bang your head against a wall repeatedly when you just are not able to get an explanation that works for you.
It has also been helpful for auto-completing larger sections of code for me, since it knows my codebase when I am quickly writing queries and things like that refencing other parts of the code base to talk to each other, it is able to quickly link and know what I am trying to do allowing me to auto-complete a whole database query in my code as an example in a one shot.
Personal Life Usage
I am not trying to be one of the A.I. Bros, just trying to keep myself educated, and also, if I can live out a little fake dream world of what the sci-fi movies told us for generations, of having a computer we can just talk to get answers to things. So on that front, I have tried picking up agentic computing, but I’m taking lots of precautions to ensure I don’t accidentally cause a major incident involving sensitive information about my personal life. So taking a moment to cover my personal journey down the rabbit hole of running my own agentic A.I. stuff.
Starting with OpenClaw
So, everyone and their brother on the internet who even remotely follows A.I. has probably heard of OpenClaw by now. So I figured it would be best to review first and take many of the lessons learned from others’ failures. So I started out with a VM on my home network, which was isolated in my DMZ virtual network so it couldn’t go crazy. I then configured it with very limited integrations with anything personal; instead, I created an email account just for it, gave it read-only access to my calendar by sharing it to its account, and limited its ability to check out and download things on my behalf.
Thankfully, with all the precautions in place, it was a pretty smooth experience for the most part… until I tried to give it more responsibility and access. Now it wasn’t like it destroyed things for me, just that it became unreliable and I had to start taking tons of extra time to remind it how to do things or ask why it didn’t follow up on things I asked it to. The worst part was when skills would just randomly break, causing me all sorts of issues. I ultimately just got kinda tired of wasting money on token usage just to get half-useful stuff.
Enter Hermes Agent
I started seeing and hearing from people I respect in the I.T. community talk about Hermes Agent and how it behaves differently. That said, it is also way better at remembering things without it costing a fortune to re-send context all the time. So I decided to start fresh and build out another agent for myself. This time, things started out much more reliably. I noticed myself having to remind it a lot less often, and loved how it communicates way better, telling me at a glance what it is doing in the background rather than just “typing…” all the time. It tells me what terminal commands it is running, what stuff it is researching, etc.
What use does it serve for me?
Well, that was the big question: what could I do with it to justify keeping it around? So I started experimenting with different things and decided to treat it as if I had a personal assistant who nags me and goes out to do the things I just don’t give myself time for, since they are too time-consuming.
- Help me keep up with my appointments and tracking tasks, because as someone who suffers from ADHD, it is a major struggle, and I am constantly forgetting to write tasks down, appointments, and snoozing alarms. Having my agent poke me and ask regularly about whats on my ToDo list for the day(synced to my TickTick) or whats on my Calendar of if there was any appointments I forgot to add (yep, forget frequently keeping a stack of appointment cards in my wallet) so I can just feed them to the agent and it will add them to my Calendar(NextCloud Calendar).
- Keeping me on task after work, as I tend to just spiral when I get home with personal task overload. It looks at my to-do list and pokes me when I get home about what it thinks I should focus on, based on the energy level I tell it I have for the evening. Then it nudges every little bit to make sure I haven’t gotten distracted. This has honestly been a game-changer for me.
- Summarizing the newsletters I subscribe to, because I subscribe to many, and now technically it is because I have them go to both my inbox and theirs. This way, it can be based on the preferences I have given it, pull out the stories it thinks I would most likely want to read. I don’t have it summarized because I still like to go to the website the article is from and read it there myself.
- Budgeting is something I struggle with, and no, I am not linking it to any of my banking or transaction details. I just give it my total income, my bills, subscriptions, and other categories I budget lump sums for. Then it also has a running tally of my debts, mixed with any debts that have interest-free clocks. It is really helpful at counting down and presenting it as a quick single-look dashboard, giving me insights into my debt, how long it is taking me based on minimum payments, and those interest-free clocks in my face help me prioritize those.
In Conclusion
I am not by any means saying that A.I. is this amazing thing. However, I am saying it is here and most likely isn’t going anywhere. I think we have already started seeing the bubble start to pop, as several companies are starting to rehire people, ditch certain A.I. tools for basic tooling, and, heck, even regular consumers are just being burnt out and tired of it impacting our economy, making things more expensive for us.
With all that said, I think it is a big deal to remember that there is no such thing as ethical consumption under capitalism. Does that excuse it and make it right? No… I still avoid using A.I. Image Generation; I just think it is kinda gross and immoral. I still spend tons of money commissioning artists for texture work on my avatars, learning how to texture things myself, and paying for art of my sona’s. That said, I do know that it does little because big businesses still continue to use it for their stuff, but it is why I think it is more awesome to support artists even more now, because it’s like paying for that “handcrafted” type label for things we buy at the store these days.
This was just my personal opinion, and I know people will disagree with me. However, I know that if I don’t adapt, I will be left behind and lose my job. Sadly, I still have debts to pay and artists I still love to commission. So I have to accept that I must change with the tides and keep myself educated on how to leverage the tooling that is now being thrust upon me and my industry. I just felt like it would be a good post to share my thoughts on how I think it is a better way, I feel less dirty using it, and how it has helped me in other ways with the personal agent stuff.
