One Hour, One AI, One Destroyed Network Config

The One-Hour Detour That Fixed Everything

I just worked a long stretch for a client on a custom SaaS product on Sunday. All day, locked in, barely lost energy. Got up first thing the next morning, early and motivated, walked into my office, sat down at my computer ready to get a full day's work done. Client has a lot of work lined up. We just had a conversation about what all we want to do. I've got my work cut out for me.

I took one hour before I got started and just sunk my claws into the network.

I've had network performance, reliability, and security issues at my home office that have been bothering me for a long time. I knew there were a lot of improvements to make. I had the list in my head. But I'm not an experienced enough network administrator to just knock it all out. Everything I know about networking comes from a few basic Cisco classes, a couple years working on-prem in an IT department at a doctor's office, and just general knowledge you pick up from using computers and having to fix stuff when it breaks because you don't want to wait on an IT person to show up. Small mom-and-pop shop stuff. Can't afford to call an IT company out for a few devices. So you fix your own routers, your own data server, your own printer.

Why Did This Bother Me So Much?

Everybody in the house felt the symptoms. The internet's slow. Can't connect this device. Weird security warnings when trying to get online. But when you're like me and you know the infrastructure behind it, when you understand the different network equipment down to the wire, how they're all configured, why they're even able to talk to each other at all, let alone why they're not performing well, when you can think back on what you've built on these servers and what you've run on this equipment and then imagine what you forgot about that's still running, eating up storage, using compute and memory, opening up ports on the firewall... it's a different kind of pain.

I knew about old dangling services from legacy projects, demos, side quests, all lingering on my servers and network equipment. I knew about the hacked-together firewall rules. I knew about the overlapping Wi-Fi channels. What bothered me most was the wasted potential. I had real hardware in that server machine just sitting in a closet barely being used. I knew the specs on these old Netgear routers I'd flashed with OpenWRT. I knew the potential for roaming. I knew how close I'd been to properly subnetting the different access points before time constraints and inexperience forced me to table it.

I actually felt sick. Physically, viscerally sick whenever I thought about this equipment.

I've met really good network administrators. I've studied under some pretty serious network and security experts. So I know the difference between my level of expertise and what could be done if one of them came and spent a day with me. The potential gains were tremendous. And I just couldn't get there on my own.

Letting the Claws Loose

I've been having such great experiences with sicking these claws onto different problems and watching them tackle things. All I have to provide is the vision, the inspiration, and a little technical nudging with whatever generalized knowledge I have. Being generally knowledgeable on a technical level, having a wide breadth of professional experience, being legitimately elite in some areas, and then compounding that with full force multipliers that bring deep technical knowledge in networking, cybersecurity, server architecture, network topology.

So I thought, why not just apply that here? I'm here. We're here. I'm going to be here all day. Let's just kick this off with an analysis. Just analyze and document findings and get a plan together. A few minutes turned into a few more minutes, and we just kept uncovering things. I watched it discover things on our network, one after another. It would think it found everything and I'd say, no, you don't understand. This thing is such a mess. You're only on one segment. You're only on one network on the premises.

See, because I didn't have time to properly segment the network, I just kept all the devices on separate networks. Separate routers, everything isolated, nothing talking to each other properly. When it reached the gateway of the first network and thought it had found everything, I told it there were two more routers. It hopped over to the next one and was like, whoa, now I see what you're talking about.

What We Actually Fixed

In one hour we booted the ISP router off the network entirely. Turns out we didn't need it, and I'd had an intuition about that for a while. You can't configure anything on those things. Can't change the IP scheme, can't change the DNS, can't change firewall rules, can't install anything on it. Can't even set up a VPN through it. It's physically wrapped up and sitting next to the modem now, cold, ready to go back if we ever change providers.

We discovered the DNS servers our ISP was connecting us to were working in their interest, not ours. Apparently this is common. Companies like Spectrum route you to DNS servers that benefit them, and part of that is selling your data. Switching to something like Cloudflare means your data stays private and goes through the edge closest to your physical location. Performance benefit plus privacy benefit. I didn't even know to ask about that.

The remaining two off-the-shelf routers got strategically repositioned and properly subnetted onto the same network. We flashed the latest OpenWRT version to both of them. Fixed all the firewalls. Cleaned up all the legacy rules on the routers. Fixed overlapping Wi-Fi channels that were causing radio interference. Fixed time zone issues that were apparently causing weird problems. Set up dedicated IPs for all critical devices like the servers. Set up roaming, so now when somebody walks from one end of the house to the other they stay on the same network and flip between access points without dropping service.

It also helped me evaluate which software belongs on the router versus the server. WireGuard, for example. And it cleaned up a bunch of janky config on an old laptop I'd been using as a demo server. Last year, when things got economically rough and everybody stopped hiring to wait and see what was going to happen with AI, I had to tear down cloud projects and run them off an old laptop with forwarded DNS records. There were serious security problems, but I didn't have time to optimize and didn't have money to pay for cloud hosting. So I hacked together whatever I could to keep doing demos. The server config was poor, the router config was janky because I was running web servers and a VPN server off the same machine with hacky rules everywhere. Claude went in and just cleaned all of it up.

The Morning After Being Sick

You know that day when you've had the flu for days and then suddenly you wake up and you know you're better? And it's a sunny day outside and you just have all this energy and you feel great? That's how I feel toward our network right now.

Right after I finished, I caught a podcast with Andrej Karpathy where he explains how he got his AI to go to work on his smart home devices. And I'm listening to him describe basically what I just did. This is what so many engineers are discovering right now. You don't need to be an expert in every domain. You need vision, energy, and enough technical knowledge to point the AI in the right direction and know when it's wrong.

What Does This Mean for Software?

This network thing got me thinking about something bigger. Part of the reason I don't do native app development is because it's a pain to work with the App Store. Setting up as an Apple developer, getting approval, making sure everything is perfect. But there's absolutely no reason I can't put my phone in developer mode, sideload apps straight from Xcode, and just use them. I need a dedicated voice memo app that locks the phone down, keeps the screen on, blocks everything except maybe emergency calls from my family? I can just build that and sideload it.

And here's the thought that actually made me happy. I think the future of shipping apps is going to be shipping a markdown file with a spec. You publish a specification for Xcode and a specification for Android Studio, and rather than publishing an app to the App Store, anyone with Xcode or Android Studio can follow your spec and build their own version. You won't be selling compiled software from a store anymore. You'll be selling the compressed markdown of an app.

And then people don't have to accept your app as specified. Your specification is the minimum viable product, the core solution. Everybody customizes it however they want. Rather than daily active user metrics, you track the popularity of your markdown. It'll be a GitHub repository. People will open pull requests and contribute. Everybody will be running completely unique software.

A lot of similar conversations are happening right now. I was listening to one about how everything in the future is bespoke. Music, movies, software. Film critics talk about how there are only a handful of core stories and everything else is nuances in plot, setting, delivery, performance. I think that's derivative of Shakespeare. Software is the same way. You share a markdown spec, the AI pops out the core, you start using it because that's the MVP, and then you just tell the AI what features you want added. No more DMing the developer hoping they're still small and plucky enough to respond. No waiting for them to ship an update. You just ask your AI, it adds the feature, you hook up your phone, it sideloads, and you're running. You can tell your AI to watch for upstream security updates and merge them in. Or open a pull request because you built something awesome that other people would want.

Open source has some really exciting prospects right now.

From "Maybe You Can Do a Thing" to "You Can Just Do Things"

I don't get tired like I used to. The productivity multiplier of the changes that have occurred over the past month to my workflow, tooling, skill set, and capabilities is unreal. Looking back on the way things were a month ago feels like looking back years. It feels like a month ago we were pulling a horse and carriage and today we're flying rocket ships.

I've always had all these ideas and things I want to do, and I just had to accept that the reality was: you might get to do one of those, and you're going to pay a dear price, and it may never have any major benefit other than proving you can do it. That stubborn, diehard energy where proving you can build a thing is the payoff itself.

We've gone from "you might be able to do a thing" to "you can just do things." It feels like transitioning from a ragtag rebel group scrapping for survival into a federation of freedom fighters that is actually winning. You can just do things. And your things are going to do awesome things for you.