I was a principal engineer. This morning I applied to pack perfume boxes.

This morning I applied for a perfume packing job in Al Qusais. The listing says exactly what you’d expect: pack boxes, industrial unit, shifts. Until recently I was a Principal Engineer at Property Finder: fifteen years in software, the last stretch as the person other engineers escalated to. I left voluntarily. Nobody pushed me out and nothing dramatic happened. My final year there was spent on AI labour automation, pointing models at knowledge work and measuring what fell over. Less fell over every quarter. I looked at where that curve was going, decided I was on the wrong side of my own conclusions, and resigned to take a break and work with my hands.

This is the first in a series about what I’m doing instead, and why the reasoning isn’t specific to me.

What fifteen years of software actually taught me

Not much of it was typing. The durable skills turn out to be portable in ways I didn’t expect.

Procedure discipline. A production deploy and a safety-critical melt have the same shape: a written sequence, preconditions checked before you start, no improvisation past the point of no return. My pewter melts at about 230°C and only fumes if you badly overheat it; the runbook says melt just above the melting point and never let it smoke. That is a deploy checklist. I already know how to follow one. More useful: I know how to notice when reality has drifted from the document.

Error-catching. Fifteen years of code review trains one specific reflex: scan a hundred correct things and lock onto the one that’s wrong. That reflex is quality control on a packing line, more or less verbatim. It’s also what keeps a hobby metallurgist healthy. The one hard rule at my casting bench is “if I can’t identify the metal by where it came from, it doesn’t go in the pot”, because unknown scrap is where the lead, zinc and cadmium fumes live. Spotting the single wrong item in a tray of right ones is a trained skill, and I have it.

Learning systems fast. Every warehouse is a system: throughput, bottlenecks, failure modes, the informal workarounds that keep it running despite the official process. I’ve spent a career walking into unfamiliar systems and building a working model of them within days. I expect the skill to hold when the system is made of conveyors and people instead of services and queues.

What doesn’t transfer: my hands are junior. That’s part of why I’m going.

The inversion

Here’s the argument, and my own project is the only proof I can offer.

The career ladder I climbed assumed a hierarchy: hands at the bottom, knowledge work in the middle, judgment at the top. You start by doing, graduate to writing code, then to reviewing it, then to deciding what gets built at all. AI is climbing that ladder from the middle outwards. Writing code, drafting messages, doing research, keeping a project plan honest. The middle rungs are exactly what models do first and best.

What’s left is the two ends: judgment about the real world, and physical presence in it.

So I inverted my own stack. I run a metal-casting project: lead-free pewter for detailed pieces, thirty kilos of clean aluminium offcuts waiting for a propane furnace day, brass ruled out because zinc fumes and I like my lungs. My AI assistant does roughly 99% of the knowledge work. This morning, while I slept, it messaged fourteen scrap dealers and sellers across Dubai, Sharjah and Ajman, every message opening with a disclosure that it’s my assistant. It offered AED 70 on a lot of old trophies (mostly plated junk, probably, but one stamped pewter piece would pay for the lot) and AED 35 for a used hotplate. Mid-run it found and fixed three bugs in its own messaging bridge. Earlier this week it pulled a tin quote of about AED 220 a kilo from a dealer in Ajman, built a full starter cart with live stock levels (“one silicone kit left, order today”), and wrote the first-pour guide, the patina recipes and the lead-free verification checklist. I read the morning brief over coffee.

My 1%: show up with cash, look the seller in the eye, check the metal is what the stamp claims, carry it to the car, pour it. Every item on that list requires a body in the Emirates. None of it can be done from a data centre. The operation has hard rules. Anything over AED 300 or any binding commitment comes to me, and money is cash, in person. Inside those rules it runs itself.

That’s the inversion in miniature. The knowledge work became the cheap part. The scarce input is me, physically, at a scrap yard in Al Sajaa at eight in the morning.

The next few months

Hands-on jobs, starting with whichever listing takes me. Perfume packing is this morning’s application, not a destination. A first pewter pour from a 3D-printed master. An aluminium melt day on a rented farm plot in Ajman, because it turns out no makerspace in Dubai, Sharjah or Ajman does hot metal, a fact my assistant established overnight so I didn’t have to. Flea markets at weekends; there’s one at Festival Plaza today. The assistant keeps running sourcing, negotiation, guides and logistics. I supply hands, judgment and presence.

There’s also a plainer reason for the packing bench. My automation year taught me that you can’t reason well about work you’ve never done. I got the knowledge-work half of that education over fifteen years. I’d like the other half from the inside, at shift pace, with my own back and hands doing the learning.

What it costs and what it buys

Costs: income, for now. Status, too. The feeds don’t have a reaction button for this one. The title was doing more for my ego than I’d admitted, and watching it go was informative. Some dinner-party conversations now contain a pause.

What it buys: calibration about the real economy that no dashboard was ever going to give me. A skill stack, hands plus AI, whose floor rises as the models improve. And objects. At the end of a pour there’s a piece of metal with weight in it, and weight is the one premium signal software can’t fake.

I’m not telling anyone to quit. I’m telling you what the ladder looks like once you step off it: the middle rungs are dissolving, and the two ends, judgment and hands, were the valuable part all along. I spent fifteen years on the middle rungs. This week I find out what the bottom one actually feels like. My working hypothesis, to be updated here weekly, is that it isn’t the bottom any more.