Being the Friction

A guy I’d worked with called me a few months back. His team had spent the better part of a year automating a workflow that wasn’t ready to be automated. There was one question that would have surfaced that in week one, and I’d asked it in week one. It went the way you’d guess. And on the call, sorting through the wreckage, I said the same sentence I’d said back in month one, close to word for word. This time it landed. He repeated it back to me like it was news. Nothing about the sentence had changed. The listener had.

Which leaves a strange question sitting on the table. If the advice only works after the fall, what exactly is advice for?

I want to be fair to him, because I’ve been him. Somebody warned me once, early and specifically, about a project I was excited about. I heard the warning about eighteen months later, which is to say I heard it in my own voice, standing in my own crater. So none of what follows is the smart guy complaining that nobody listens. Nobody listened to me either. Including me.

Here’s what I think is actually going on, and it’s less about psychology than about a promise. Most AI conversations right now hold some mix of the same crowd. Early adopters riding the momentum and loving it. People who’ve been told to use AI, by a boss or a mandate or a quarterly goal with their name on it. People being nudged into it with links and lunch-and-learns. Plus the quietly unconvinced, the burned-once, and at least one person who just wants the meeting to end. Different roads into the room. One expectation most of them share, installed by a thousand demos and keynotes: this is supposed to be easy.

And the truth underneath any real decision process is that doing this well takes work. Questions before tools. Counting before buying. A test before a contract. So when you show up carrying that truth, you’re not offering advice at a bad moment. You’re re-pricing the product after the room already ordered. You’re the waiter coming back to mention that the lobster was market price. The early adopter hears betrayal. The mandated hears more burden stacked on the burden. The encouraged hears the trap being sprung, because deep down they suspected there was one. Nobody in that room ordered difficulty.

And prevention was already a hard sell, since the disaster you prevent never happens, and the crater you don’t make never gets a parade.

Which brings us to the part where you’re the one who sees it. Seeing it turns out to be the cheap half. Saying it is the expensive half, and not because anyone’s shy. Being the friction takes confidence, and it takes something like leadership, because you are introducing friction and a little chaos into work that is already moving, and nobody wants either one. Ask “are we trying to automate something that doesn’t work yet?” out loud and you haven’t just raised a question. You’ve become, in the flesh, the difficulty this room was promised wouldn’t exist.

And there’s a version of this trap we build for ourselves. At the start of any project or any client relationship, you provide no friction on purpose. It’s how you get the work, get the project moving, get invited back. I know this move intimately. I have sat in first meetings being spectacularly easy to work with, nodding along to a plan I could already hear creaking, my inside voice saying “that step doesn’t exist yet” while my outside voice said “love this energy.” Sometimes both voices in the same sentence. That agreeableness isn’t weakness. It’s strategy, and mostly it’s good strategy. The trap is that it compounds. Every nod raises the price of your first real question, until month four arrives, you finally ask one, and the room looks at you the way you’d look at a golden retriever that suddenly said, “Actually, I have concerns.” Same question. Wrong month.

There’s one more wall, and it deserves to be quoted in the room’s own voice: nobody has time for tests. All these steps, the counting, the two-week run. People don’t have time for this, or believe they don’t, and every step you propose reads as bureaucracy bolted onto a product whose entire pitch is speed. This isn’t the AI dream. That sentence, said or silent, is what every upstream question eventually hits.

So here’s the reframe I’ve landed on. Knowing the right process was never the hard part. The six questions I run AI decisions through fit on an index card, and plenty of people can recite something like them. (If you want mine, they live at topfiveaitech.com, where a short free course walks you through one a day.) The hard part, the actual skill, is getting a room full of people who were promised easy to run them anyway, without you becoming the person everyone dreads inviting. That’s learnable. Three moves for the person in the room. One for whoever runs the room.

The first move is timing, because timing sets the price of the question. In a first meeting, nothing has momentum yet and nothing is threatened, so describing how you work just reads as knowing what you’re doing. Say it then, before anyone needs you to, in one breath, delivered like a coffee order and not like terms and conditions: “One quick note about how I work. I’ll ask what problem we’re solving in plain words, I’ll count how often it actually happens, and I’ll want a short by-hand run before we build anything. Fast, painless, and it’s why my stuff ships.” The by-hand run, if the phrase is new to you: do the job yourself for two weeks, AI open in a window, before anybody builds anything. Automation multiplies whatever is already there, so two cheap weeks tell you whether there’s something worth multiplying. That’s the whole move. Already months into agreeable? Don’t announce a personality change. Attach it to the next fresh decision: “For this next one, I want to try something.” The project changed, that’s the whole story, and nobody has to acknowledge the last four months. One honest footnote, though. Knowing this move and making it are different sports. The pull to just be easy doesn’t retire because you own a script. It costs a two-second flash of “am I being difficult?” every single time. You’re not. You’re showing up early enough to be the person they’ll actually listen to, instead of the one they wish they had.

The second move is sizing. Don’t bring six questions to a room that will tolerate one. Bring the one that matters most for this decision, and shape it as a measurement instead of a verdict. “Can we run twenty of our real cases through it for two weeks before we sign?” A measurement can’t be argued with. It can only be run or refused, and refusing a measurement out loud is its own answer, one the whole room gets to hear. Minimum viable friction is still friction. It’s just friction that fits through the door.

The third move is language. Never defend the test as being careful, because careful sounds slow, and slow is the enemy in every AI room. Sell it as the faster route, because it is. Two weeks and twenty real cases now, against the alternative half that room has personally watched: build the wrong thing, discover it in month three in a meeting where people study the table, then rebuild. The sentence is: “I’m not slowing this down. I’m making sure we only build it once.” And bring the labor with you: “Give me the two weeks and I’ll come back with the answer.” Friction that arrives carrying its own workload reads as a gift, not homework.

And if you run the room, there’s a bigger move available: stop requiring bravery at all. As long as the hard question depends on somebody choosing to be the difficult one in a meeting, it mostly won’t get asked. Bravery is scarce, and it bills the same person’s social capital every time. So build the question into the process instead. Before money moves on any project, a short kickoff checklist gets filled out, and one of the questions printed on it is: “Is this running in production anywhere?” Now it gets asked every time, by nobody in particular, and it is remarkably hard to be mad at a piece of paper.

The checklist takes the heat a person would have taken, which protects the company’s money and the asker’s relationships in the same stroke.

All of which still leaves the people who need to fall down once, because some of them do. Some conviction is only purchasable with a person’s own money. Secondhand pain doesn’t transfer, and that’s not a defect in people. It’s how conviction forms. When you’re standing next to one of them, the job changes. Sort the coming failure by cost and reversibility. Cheap and undoable? That’s tuition. Let it ride. Expensive or permanent? That’s where you spend your credibility, loudly. Shrink the fall where you can. Keep your question quietly on the record, with zero I-told-you-so banked, because banked I-told-you-sos earn negative interest. And be first there with the map when it happens, because that’s the moment the month-one sentence finally gets heard.

The dream most people were sold was ease. I understand the appeal. Some days I’d like it to be true myself. But the dream worth having is that it works, and small, well-placed friction is the shortest road there. Which makes the person carrying it the most pro-AI person in the room, whatever the room thinks that week.

Right now, this looks like being difficult. That’s the fringe stage, and it’s temporary, because most teams are wiring AI into their work this year, and each of them is about to find out which of its people can ask, “are we trying to automate something that doesn’t work yet?”, and still get invited back. There’s a version of the post-mortem coming where the first question is who asked, and when. The people who end up mattering in this era won’t be the ones who were right. Plenty of people were right. They’ll be the ones who were heard.

The New Typing continues — the series lives at drjeffwurfel.com/writing.