On Friday I sat on a panel for the Massillon WestStark Chamber of Commerce discussing automation decisions. “What should you automate? and what should you never automate?” was the title of the panel and a question that came up directly.
I gave a safe answer. My response was that I automate the rules-based and repeatable admin work that robs me of time doing more valuable things. My monthly accounting entries, setting up the calendar and to-dos for the week, and agenda preparation.
I said I would not automate delivery or anything client-facing that was not rote follow-up, like appointment reminders or subscription confirmations. But my response was exactly what can happen when we grab the first available answer.
Then this morning I read an essay by Giorgia Lupi in the New York Times. She’s a designer at Pentagram. Her piece is about using AI in the creative process, but it turned out to be about my work too.
And probably yours.
Her team was hired to make an installation about the New York subway system. They spent weeks studying records and open data. They even read through the missed connections board on Craigslist. What came out of that work treats each subway line as a living character. How some see the sun and others never do. Or, some trains wait for each other at junctions, while others never cross paths.
She’s clear that no prompt would have gotten her to the same place. A tool could have drawn the map, but it wouldn’t have thought to read the missed connections.
Her team does use AI. They use it after the thinking is done, to take something a person already defined and carry it out at scale. Where they won’t use it is the beginning where it can result in what she referred to as “visual elevator music.”
One of A.I.’s greatest selling points is that it removes the challenges and barriers for discovery. You type in a prompt, and A.I. gives you an answer. But sometimes discovery is as much about interrogating the questions.
The risk is not that AI will do bad work. It is that it will do good enough work, instantly, on the wrong question.
Figuring out the right question is the hard work. Looking for shortcuts isn’t new. We were trying to skip the slow part long before AI showed up. We subscribe before we identify the problem. We write the procedure before we ask the people who do the work. We hire before we know what the role is for.
The tool just makes skipping quicker.
Lupi writes about detours and weeks of wandering. How what looks like a dead end can turn into an idea. In a business, the wandering looks different. It’s the hour with the person on the front line who tells you why the new system isn’t being used. A meeting where nothing is decided, but someone finally says the real problem out loud.
The impulse to optimize for the obvious and to stop at the simplest answer is what I believe we have to resist.
That doesn’t seem productive. But, it’s where we’re going to find the real answer.
I’m not against AI tools. I use them. And, the line she draws makes sense to me. The machine can do the carrying out. But the people must do the finding out.
In the end, that panel question wasn’t really about which tasks should be automated automate. It was about how much of the thinking we are willing to give up before we’ve done it ourselves.
If I got that question again, I’d talk about just that.