A few weeks ago, there was a pretty significant announcement sitting at the intersection of AI and commerce.
If you’ve been in the industry, you know how this goes. Clients started asking questions, internal teams started looking for a company POV, we needed to produce something so people could understand what just happened.
Ideally yesterday.
So I defaulted to speed. I sent my initial thoughts to an AI agent I had already built for this exact type of thing, let the team know I could have it blow out the POV and kept moving.
At least that was my instinct.
Then Ben Hovaness stopped me. His response was:
Mike - think it would be helpful for the group if you could articulate your personal thesis on the significance of the news in a single paragraph. Once we are aligned on that THEN we can have an agent blow it out. But we need to start from human expertise and insight.
I really liked that.
I liked it because he wasn’t saying it was the wrong process, it was more about the order of operations. I took what he was saying as:
Give me the human thought first. Then go off, king!
AI makes it really easy to skip a step
I am a pretty aggressive user of AI.
It’s really twofold for me. First, I have always gravitated toward being a practitioner. I like to get under the hood and really learn something. When AI was getting popular, I decided to go deep on ChatGPT and I have never really looked back.
I also tend to be in roles where people ask me about the latest trends and how we should take advantage of them, so going deep is basically required learning for the job.
The second is obvious, productivity. Move faster (like yesterday), synthesize more information (and then compare it to a whole other data set), pressure test an idea (and then automate the insights into action).
With some basic knowledge of the platform and only having a few opinions of my own (to put it lightly), AI makes it possible to go from zero to sixty instantly.
If I have an idea for an app? I can start vibe coding it. I (with no actual skills) made a Family Feud style game to play with my two boys the other day. It’s that easy.
Need a POV for this big announcement that everyone is asking about? Great, I built an AI agent for that specific purpose.
I love that for me! So much time for other activities.
It also makes it incredibly easy to skip the moment where you stop and ask, what do I actually think?
That was the part of Ben’s note that stuck with me.
Before we go do the thing, he wanted the actual thesis. Something a human being was willing to put their name behind.
The order of operations matters
I don’t think the lesson is that humans should be the only ones doing the thinking and AI should only do the production.
I don’t know about you but AI helps me think all the time. I use it to challenge assumptions, organize messy thoughts and analyze ridiculous amounts of data to try and identify patterns. I even ask it to tell me I’m wrong just to hear the other argument.
Sometimes it helps me better understand the things I thought I already knew.
But there is an important gate before we accelerate. Particularly a human gate.
What are we trying to say? What do we believe? What is the actual judgment? If a group of humans cannot align around that, generating 100 other docs filled with slop probably doesn’t solve the problem.
It just makes your uncertainty look more polished.
I learned this one the hard way with Field Notes
When I first started experimenting with these Field Notes, I had a similar approach.
I recorded this really long voice note with all my thoughts on the topic. I gave AI a ridiculous amount of context on the idea, what I wanted to say, the flow, etc. We went back and forth and eventually I had a beautiful first Field Note that I loaded into Substack.
If you are not aware, Substack recently rolled out a partnership with Pangram where you can scan any article for AI text. I ran my very humanly concepted but very AI produced newsletter through it.
100% AI.
I literally went to my wife and asked if I should just not do the newsletter anymore.
Not because I think an AI detector gets to decide whether an idea is mine. I knew the stories were uniquely mine. But after the score and when I read through the piece again, I understood the value of the score.
My writing was too clean, too complete, the prose was too predictable and it just felt too far away from how I naturally speak human to human.
I made the decision that if I am going to publish something called The Modern Commerce Field Guide under my own name, it needs to sound like me.
So after that, I completely changed my process. The Field Notes are now a whole lot messier but they are a whole lot more like me (and frankly, I get more out of it when I write them now).
Turns out Ben was pushing me toward something I had already learned the hard way with the Field Guide. Before you go full speed with the machine, there better be something real underneath it.
Speed makes the stop moments more important
The faster the machine gets, the more valuable the gates become. This is going to sound like third grade all over again, but maybe we need to get really good at knowing when to stop, think and ask a question before we keep going.
Because imagine this situation applied beyond writing a quick POV.
In our industry, we are moving toward a world where more and more media buying could (will) become agentic.
That means you are going to be able to tell the system what you want. Give it a budget and then it’s going to go nuts. Buy, optimize, shift budgets based on certain KPIs and keep going.
But before the rocket launches, somebody still has to establish the human intent.
What business outcome are we trying to drive? Who are we trying to reach? What are we willing to spend? What are we not willing to do? What tradeoffs are acceptable? What is the machine allowed to decide without asking us? What does good actually look like? How do we even know it’s good?
If anything, the easier execution becomes, the more important those questions get.
Give me the human sentence first
I have to admit it, Ben was absolutely right.
I was optimizing for speed but I was one step too early. My mindset was the news happened, clients were asking, teams needed something…AI agent.
There will absolutely be moments where that is good enough.
But my learning was that before I do that, I need to identify the human gates first. What do I need to not just articulate, but actually understand myself before I move to production?
In this case it was a paragraph. Sometimes it may just be a single sentence. It’s really about taking the time to work through one question, what do I actually believe here?
So before we go from zero to sixty:
Give me the human sentence first.