LLM Agents have actively improved my life

Browsing posts online or talking to friends often leaves me with a feeling that many of my peers are disgruntled with the idea of using LLMs to code, and specifically around the concept of vibe-coding. While I don’t particularly love the name, I have to disagree about the value add I’ve gotten from it. Vibe coding has actively improved my life, particularly around my personal interests. In short, it’s enabled me to do the things I’m too exhausted from work to take on.

I think part of the issue is around the mental image of a vibe-coder. For most, I think it’s just some generic techbro with more knowledge of buzzwords and fads than actual experience. I have been hacking and engineering for nearly 30 years. That means that I have a wide breadth of experience, concepts, and understanding, but I’ve also reached a point in my career where my job keeps me so busy that continuing my interests at home feels like too much work. My homelab has suffered for a long time because of this. Add in family and social commitments, and it means I’m basically just engineering for money and not for fun.

But LLM agents, and claude in particular, have changed that for me.

The big value add

I’ll give you a practical example. I run a Unifi network stack at home, and several home servers that manage my media collections, streaming, home automation, etc. I have over a hundred wireless devices connected throughout my home, and several family members that are technically proficient but not engineers. And because I’ve setup so many niceties, they have become expected to “just work”, which we know is rarely true. When there’s a major outage I jump into action and troubleshoot, but when we experience things like “huh, my phone isn’t working as expected sometimes” or “I wonder why the network cameras all seem to be locked up” I take a quick look, but rarely have the time or energy to do a deep dive into finding the culprit.

I realized that setting up a SIEM at home to centralize logging events would be really beneficial to me. But do I really have it in me to go setup a full grafana stack and then customize it? Not really. But you know what is more appealing? Describing my use case to claude, my goals, and brainstorming a solution that sounds appealing.

Focus on the big picture strategy, and oversee the results

The LLM agent is just your junior engineer. I have delegated projects to people in my career so many times, and this really isn’t any different. Sure, they might have taken an approach I wouldn’t have, and I can give them advice on what to do differently, but sometimes they surprise me and I learn an approach I wasn’t as familiar with. My job is to review the results, see if they fit the expectations, and look for any obvious problems or concerns.

That’s exactly what I do with my claude agents too.

Stop doing the boring work

I’ve setup grafana stacks plenty of times, and the idea of doing it again isn’t entertaining to me, it’s just a chore. But working with an LLM to develop a strategy and a plan, and then overseeing its work is slightly more compelling. So that’s what I did. And in this particular case, I’m actually trying to take it a step farther and add in some additional intelligence.

After brainstorming with claude a bit, I decided that I wanted to create a SIEM, and then begin working on a way to feed recent events to my local 2.5-32B model for active analysis. The idea (that is still very much in flight) is to collect all the important events from my homelab network, feed it into a centralized location, and then setup some triggers that suggest something might be awry. When this is determined, we’ll then work on some prompt engineering to feed that into the local LLM for analysis to look for correlating events and once a certainty threshold has been hit, log the event in a newly created WebUI for future analysis. Or in some cases, possibly send a push alert to my phone so I have more awareness and can look for other events that might not be so obvious.

That is fun.

And that is also something I wouldn’t have time for in my busy life these days, but I now can tackle thanks to having an army of assistants to help me with the basics.

Validation along the way

I’ve been working with claude code to fix all of the syslog ingestion from different devices, as well as setting up transforms for Loki. This is great because it’s boring work and the perfect thing to delegate. Since claude knows what my overarching goals are, it can provide me insights along the way. For example, after we got a Unifi dashboard built out to try to highlight some random traffic loss I sometimes experience, it was checking the ingestion changes we’d just made and shared this (sanitized):

● Data is flowing to Loki now. The dashboard should be showing real data. Since you mentioned the Grafana dashboard is now
  populated, that's great!

  The raw logs you're now receiving are incredibly valuable for diagnosing your intermittent issues. Looking at those logs, I       
  can already see:

  Potential issue identified: Device ff:ff:ff:ff:ff:ff on "<home network ssid>>" is repeatedly being rejected due to low RSSI
  (signal strength around -80 to -89 dBm, but the MinRSSI threshold is set to 23 which translates to about -72 dBm). This device    
   is trying to connect to the U6-LR AP but keeps getting rejected.

  Similarly, a client ff:ff:ff:ff:ff:fe is being kicked from the "Outside" AP due to Low RSSI (15).

  This is exactly the kind of thing that would cause intermittent app loading issues - devices hovering at the edge of AP
  coverage, getting rejected or kicked, then trying again.

  Would you like me to update the dashboard queries to better highlight these RSSI/signal-related issues?

Just jump in

There’s likely a good balance you can find for yourself as well. And honestly, learning how to use the tool itself presents a new and interesting challenge that is worth exploring. This is a greenfield opportunity to be at the starting point of something new and amazing, and there are so many good ideas floating in your head that haven’t been attempted yet. Get out there and hack.

And for those interested, I’ll share more about my journey with the SIEM creation as I get a bit farther along with it.