The Myth of Going Offline When Your Job Never Sleeps

A few days ago, I published a post about glass houses, magnets, and the friction of trying to coexist with people and systems that don't quite align. It was personal. It was meant to be.

What I didn't say in that post: I wrote it during one of those weeks where I genuinely tried to step back. Reduce screen time. Be less digitally present. Focus on the tangible, the physical, the human.

It lasted about six hours.

The Backup Window

Here is something most people outside of IT don't understand: our real work starts when everyone else goes home.

Not metaphorically. Literally. The daily backup window opens after business hours. Maintenance windows are scheduled for nights and weekends. Monitoring alerts don't care that it's Sunday. And if something breaks at 2 AM, there is no "I'll look at it Monday" option, because by Monday the damage is done.

I can't talk about specifics here, and I won't. Anyone in this field knows why. But the principle is universal: if you are responsible for the systems that keep a business running, your availability is not a 9-to-5 arrangement. It never was.

LinkedIn at Midnight

So when I say I'm "only on LinkedIn away right now," what I actually mean is: I closed the social feeds, I stopped scrolling, I redirected my attention. But the terminals are still open. The monitoring dashboards are still running. The phone is still on, with notifications from systems that don't know what a weekend is.

The irony of writing a post about stepping away from digital noise while simultaneously tuning language models at 11 PM is not lost on me.

What I've Actually Been Doing

Instead of doom-scrolling, I've been building. Over the past few weeks, a few things came together that I think are worth sharing, not as announcements, but as observations from the trenches.

Model failover for AI agents. When you run AI-assisted workflows in production, the models go down. Not sometimes. Regularly. Rate limits, auth failures, provider outages. I built a failover system that automatically detects when a provider is throttling or rejecting requests and switches to the next available model in a configured order. The interesting part is not the code. It's the realization that running AI in production is less about picking the "best" model and more about ensuring you always have a working one.

Multi-model orchestration. Related to the above: I now run over 40 language models across five providers, 23 of them effectively free through a single subscription. The economics of AI have shifted dramatically. A year ago, running multiple frontier models in parallel was expensive. Today, the bottleneck is not cost. It's reliability and knowing which model to send which task to.

Self-healing infrastructure. This one is still in dry-run mode, but the idea is simple: when something breaks in the agent pipeline, instead of just logging an error and waiting for a human, the system opens a GitHub issue with diagnostic context and attempts a fix. It's the kind of thing that sounds obvious in hindsight but requires careful guardrails. You don't want an automated system "fixing" things you didn't ask it to fix.

All of this is open source. All of it grew out of the same impulse: I can't be offline, so I might as well make the time I spend online more productive.

The Honest Part

I know the "disconnect to reconnect" narrative. I've read the books. I've tried the apps that lock your phone. I've done the digital detox weekends.

None of it sticks when your phone buzzes at 3 AM because a certificate is about to expire, or because a backup job failed silently, or because a model provider decided to change their API scopes without notice.

The magnet problem from my last post applies here too. I can try to push myself away from the screen, but the systems pull me back. Not because I'm addicted. Because I'm responsible. And responsibility, unlike a social media feed, doesn't have an off switch.

What I've learned to do instead is redirect. Not less time online, but different time online. Building instead of consuming. Shipping instead of scrolling. Writing instead of reading threads that go nowhere.

It's not balance. It's triage. And in IT, triage is the only honest form of time management.

The Connection

The glass house post was about friction between people. This one is about friction between intention and reality. I intended to go offline. Reality reminded me that my job, my projects, and the systems I maintain don't recognize that intention.

So I'm here. At midnight. Tuning models, writing code, and apparently also writing blog posts about why I can't stop writing blog posts.

If you're in IT and you've ever tried to "unplug," you know exactly what I mean. The rest of the world gets to log off. We get to choose which screens to look at.

That's not a complaint. It's just the job.