My hand is still vibrating from the impact of the old sneaker against the floorboards, a dusty smudge marking the spot where a spider used to be. It was a primitive solution, honestly. No sensors, no cloud-connectivity, no firmware updates required. Just a physical object meeting a biological problem at high velocity. It worked with 98% efficiency if you count the adrenaline spike as a success metric. My monitor, however, is currently displaying a much more ‘sophisticated’ failure. There is a provisioning script blinking at me, 408 lines of PowerShell that I wrote to save myself from the repetitive task of onboarding new users. It was supposed to be the holy grail of my department. Instead, it has become a hungry pet that eats my afternoons.

This script, with its 28 external dependencies and its habit of failing silently about 18% of the time, requires more daily health checks than the manual process ever did. I spent 8 hours yesterday debugging why it couldn’t find a specific container that definitely existed. If I had just manually typed the commands, it would have taken me 48 minutes. Yet here I am, tending to the machine that was supposed to serve me.

The AI Babysitter Paradox

I am not the only one trapped in this loop. My friend Reese T., an AI training data curator, lives in the heart of this paradox. She manages a team of 148 people whose entire professional existence is dedicated to pretending that software is smarter than it actually is. They spend their days labeling images of crosswalks and traffic lights so that an ‘autonomous’ driving system can eventually fail to recognize a cyclist in a rainstorm. The system is marketed as a miracle of machine learning, but it is actually powered by the sweat of Reese’s team, who have to manually override the ‘intelligence’ every 8 minutes because it decided a shadow was a solid wall.

The Exhaustion of Broken Promises

There is a specific kind of exhaustion that comes from maintaining a broken promise. When you do a manual job, you know when it’s finished. You can feel the weight of the task leaving your shoulders. But when you maintain an automated system that is prone to ‘edge cases’ (a fancy term for things the developer didn’t think about), you are never truly off the clock. You are always waiting for the 3:08 AM alert that tells you a dependency has updated and broken your fragile web of logic. It is a psychological tax that we don’t account for when we calculate the ROI of a new tool.

Compare that to a system designed with the understanding that reliability is the only metric that matters. When you are setting up a Remote Desktop environment, for example, the last thing you want is a licensing server that requires a weekly séance to keep running. You need something like a windows server 2019 rds user cal that doesn’t demand you build a 408-line script just to stay compliant. There is an elegance in things that just work without requiring you to become their full-time mechanic. We’ve lost sight of that. We’ve been taught to believe that if it isn’t complex, it isn’t ‘enterprise-grade.’

Lying to Ourselves in Spreadsheets

I once tried to explain this to a manager who was obsessed with ‘eliminating human touchpoints.’ I showed him that his new automated reporting tool was actually requiring two interns to spend 28 hours a week cleaning the data before it could be fed into the ‘auto-generator.’ He didn’t care. To him, the fact that no one was manually typing the final report meant it was automated. The interns were just ‘overhead.’ This is how we lie to ourselves. We bury the human cost in the ‘Operations’ budget and keep the ‘Innovation’ budget clean and shiny.

It’s a bit like the spider I just killed… Instead, I used the shoe. It was a manual intervention. It was ‘un-automated.’ And yet, I am now free to do other things, rather than spending the next hour troubleshooting my spider-trapping infrastructure.

The Human 80%

We are building a world of fragile dependencies. Every time we add a layer of ‘convenience,’ we add three layers of hidden maintenance. I see it in Reese T.’s eyes when she talks about the ‘automated’ content moderation tools that require her team to look at 8888 flagged posts a day because the AI can’t tell the difference between a Renaissance painting and a violation of terms of service. The machine does the easy 20%, and the humans are left with the 80% that is exhausting, nuanced, and soul-crushing. But on the company’s annual report, it says: ‘98% of content is moderated by AI.’

True Efficiency

True efficiency isn’t about removing the human; it’s about respecting the human’s time. Sometimes that means writing a script. More often, it means choosing tools and processes that don’t break the moment you look away. It means admitting that some things are better done by hand, or by systems so stable they don’t need a curator. I look at the smudge on the floor and then back at the blinking cursor. The cursor is demanding an update. The shoe is just sitting there, ready for the next spider. I know which one is more ‘automated’ in the long run.