Abstract data visualisation in dark blue

The rebuild was never about the servers

This series opened with an AI agent wiping all my servers and me, briefly, wanting to walk away from IT for good. Several posts later I run more agents, with more freedom, than I did before any of it went wrong.

That still surprises me. So let me say plainly what actually changed, because it wasn’t the hardware.

The servers were never the point

I rebuilt the infrastructure twice. Hardware is replaceable — and mine kept proving it. If this series had been about servers, it would have been a dull one: a thing broke, I bought another thing, I plugged it in.

What actually got rebuilt was how I think about working with a machine that can act on its own. That part doesn’t fit on a shelf, and it’s the only part worth writing down.

What I’m keeping

A handful of convictions came out of it that I now hold harder than I did before:

  • The information you give a system matters more than the rules you wrap around it. Most “AI is unreliable” problems are really “I described the world badly” problems.
  • Freedom and safety aren’t opposites. You get both by drawing the blast radius first, then letting go inside it.
  • The boring layers are the ones that save you. Not the clever new system — the old cron job, the decade of documentation, the backup nobody bragged about.
  • A partner is something you design, not something you buy.

None of those are products. They’re ways of thinking, and they cost nothing to take with you.

What I actually do

I build things for a living — software, infrastructure, the occasional agent whose only job is to listen. But the thing underneath all of it, the thing this whole disaster clarified for me, is older and simpler: I like understanding a problem all the way to the bottom, and then making it smaller. Optimising. Tightening. Removing the part that never needed to be there.

That’s what I have, and it’s what I wanted this series to be — not a pitch, not a funnel. Just the knowledge, told honestly, including the parts where I nearly gave up.

If any of it was useful to you, that was the entire point.


Part of The 2026 Rebuild.