People keep asking what my setup looks like. It's a mess, but it works. Here's the honest snapshot as of tonight.
The Machines
I have two Macs. My current one is my daily driver, but right now it's not doing a lot of heavy lifting because it's busy backing up everything from OneDrive so I can leave that godforsaken service. My older Mac sits as a server. When it has free memory, I run Codex on it to fix up my apps and push updates.
Then there's my Windows computer, which I access through RustDesk or AnyDesk. It's connected via Tailscale to Antigravity from a different IP, so I can use someone else's Antigravity subscription since she doesn't use it and is building my Android apps so I can test them.
The AI Stack
Gemini (web) is my go-to for questions. It's the most loyal and consistent of the bunch. I have memory turned off, so all my chats disappear and it stays clean. No context buildup, no weird callbacks to conversations from three weeks ago.
Codex (ChatGPT Plus) runs on droplets. I have two accounts going, doing things on DigitalOcean servers whenever I want something done. I babysit occasionally in case one of the agents is doing something wrong, but mostly they just run.
Gemini CLI with gemini-3.1-pro handles stuff directly on my Mac. The OneDrive backup script, writing solutions to math assignments, that kind of thing.
Cursor Pro works on this Mac to write my articles. That's almost all it does. I sometimes use it to help build things, since they white-labeled Kimi K2's 1 trillion parameter model (that's a fun story about how they didn't provide proper attribution, by the way). But it's mostly a dead tool to me because it requires a GUI and has tons of overhead. Very quickly makes my computer get hot and the fans run at full speed.
Oh, and if you have a Google student subscription: did you know you have 3 separate Gemini 3.1 Pro limits? There's Gemini web, Gemini CLI, and Antigravity. They each have their own quota. You can't use any of them with OpenClaw/ZeroClaw/Hermes (the agents I spin up droplets with), but for everything else, you're getting three pools of Pro usage for the price of one.
The Terminal Situation
I tried switching to iTerm2, because it has badges so I can quickly see what each terminal is doing. It's annoying that they force you to use Esc+B and Esc+F instead of Opt+Left and Opt+Right to move by word, and Cmd+Left/Cmd+Right for full lines. It's a bit more overhead than Terminal.app. But I can have 6 terminals on one screen, and when I'm using my laptop's 16" display, that's incredibly useful. I can see all 6 at once, and then just switch over to RustDesk to control my Windows machine.
Random Telegram tip I just discovered: Cmd+Tab in Telegram switches between chats. Very useful when you're using Telegram like a terminal.
The Servers
I use DigitalOcean for spinning up droplets. It's the easiest way to get a server running. There's also Modal and Daytona, but I haven't looked into those yet. If you sign up through my referral link, you get $200 in credit to use over 2 months, and I get a bonus if you ever end up spending money with them. I'm going to use my own code to save about $100, since that's roughly what it costs me per two months to run my sites and other little projects. The sites themselves are about $4-6 per server on the DigitalOcean side.
For the serious stuff, I use Hetzner. Cheaper and more powerful than DigitalOcean, but a bit more hassle to set up. They require a government-issued ID to verify your account (or maybe that was just YesWeHack and Google Developer; I signed up for all of these around the same time and the memory's a little blurred). Part of the friction might have been because I set up my own email signing server on Hetzner and needed the SMTP port unblocked. Their support chatbot was actually helpful with that. It told me I shouldn't use some specific term in my request because it sounded like what a spammer would say, and suggested an industry term instead. I was reluctant, but I went with it. They unblocked the port pretty quickly. I think it was no more than 8 hours, since I sent the request and went to sleep.
I don't have a Hetzner referral link yet since I haven't spent enough with them. I did use my bots to find a discount code that saved me $10, but I missed out on the $20 I could have gotten from a referral. I'll add one here when I get it.
The Point
None of this is elegant. I have agents running on droplets, two Macs doing different things, a Windows box on Tailscale, Cursor for blog posts, Gemini for questions, Codex for builds, and a 16" screen split into 6 terminal panes. But it all works. The trick is not having one perfect setup; it's having enough separate things running that something is always making progress, even when something else is broken or backing up or waiting for a port to get unblocked.
This post contains affiliate links. If you sign up through the DigitalOcean link above, you'll get $200 in credit and I may earn a bonus at no extra cost to you.
Update (March 29, 3:18 AM): It is now four hours later. I just wasted four hours. I could have been doing math. But no, I was "programming." Quotes because at this point I'm mostly talking to chatbots and watching terminals scroll.
Anyway, I've changed my mind on something. Everyone's been saying what I used to say: that AI will make the best people, the specialists, more productive. I don't think that's right anymore. It's making the best users of AI into superhumans. That's the current state of affairs. Not "experts get a boost." The people who know how to wield these tools are running laps around everyone else, regardless of prior expertise.
But it's about to get really expensive. The cost of RAM, electricity, and parallelized compute (GPUs) is heading for the stratosphere. This stuff doesn't run on good vibes. So learn how to use debt, or welcome to the permanent underclass, because life is going to come at you fast.
Self-deprecating footnote: those $1,000 Trump accounts I opened for the kid? Ha. He might be able to afford food for a month by the time he's 18, after accounting for the appreciation.
Enjoyed this post?
Get notified when I publish something new. No spam, unsubscribe anytime.