Embrace the Mess

SOPs. Manuals. Policies and procedures. Process maps. These all give me the heebie-jeebies. I’m not someone who embraces chaos, but I’ve always rankled at the idea that business problems can be solved with simple yes/no answers, or a flow chart. I understand the desire. Every executive I’ve met wants to do their best to reduce uncertainty, and wants that from their teams. And every business article I read seems to want AI to work the same way.

But here’s what’s funny about that. Businesses are already messy in the middle. Getting clarity across a team of people who all have their own mental models, their own frames, their own assumptions requires constant translation and bridging. We’ve been doing it forever and we’re honestly not great at it. Now we’ve added AI, which is messy in its own ways (probabilistic outputs, hallucinations, shortcuts you didn’t ask for, different answers tomorrow with the same inputs). You’ve got messy meeting messy, and the instinct is to at least make the technology side clean (to treat AI the way we treat other software, with rules and procedures that force a normal computer to behave).

I originally had the same impulse when I started working with AI every day. But now I think that approach is working against what makes both humans and AI useful. AI isn’t a deterministic technology, and professional judgment isn’t deterministic either.

Pulling the Plug

I was talking with a small business owner recently who’d turned off the AI scheduling assistant bundled with her main software. It had booked her for an in-person meeting fifteen minutes out, with a teammate who was actually in another state. She (rightly) shut it down. She wants to have AI help with coordination and scheduling but she needs to be able to trust it first.

When something fails on you in a way you wouldn’t have expected, the immediate response is to remove it before it does worse damage. But the AI didn’t fail because it was AI. It failed because it didn’t have the context it needed to do the job well. It didn’t know which teammates were where, what the business hours actually were, or how she books her time. With regular software, missing context feels like a settings problem; with AI, it looks like bad judgment. And this wasn’t the business owner’s fault either. Often it’s not clear how much input AI needs from us to be successful at a task. The things AI does well out of the box can feel magical, but they are not reliable enough for real work without more context from you. And since they don’t come with an instruction manual, we just turn it off.

I had the same reaction when AI burned me a few times. I tried to tighten the rules around it and sometimes just took it out of the workflow entirely. We all need technology to behave more predictably, and AI needs more from you to make the right choices. It’s easy to think the AI is just broken or won’t work for you, but you’re not addressing the underlying problem (which was almost always that the AI didn’t know enough to begin with).

We’re a Mess Too

Our judgment works the same way. We’ve built it over years of bad decisions, stumbles, and breaks we couldn’t write into a set of rules even if we tried. For me, that shows up as pattern recognition I can’t always articulate, or gut calls that turn out to be informed by something only half-remembered from an experience three years ago. Most of our day-to-day work (reading a client situation, deciding when to push and when to back off, knowing what’s worth flagging and what to let slide) isn’t deterministic. It’s just so familiar that you stopped noticing how messy it is.

We have all been navigating other people’s messiness our whole lives. Translating between people who use the same words to mean different things, bridging the gap between one person’s mental model and another’s, knowing both will hear the same information differently. AI is now another party in that mix. The brand of mess is different (AI brings probabilistic outputs and pattern-matching shortcuts you didn’t ask for, where humans bring their own unspoken assumptions and competing priorities), but the work is recognizably similar. You’re learning what it does well and where it goes off the rails, when to lean on it and when to redirect. That’s the same skill you’ve been practicing with people, just applied to a partner that operates differently than they do.

Past the First Answer

For a long time I was paying GoDaddy thirty bucks a month to host four small websites. They were simple HTML and CSS, no databases, nothing dynamic, just a few pages each. I’d been hosting my own sites since the late nineties (back when you’d FTP files into a directory called public_html and that was the whole deal), so paying for hosting just felt like the cost of having a website. At some point I figured I’d ask AI whether the thirty dollars still made sense.

The first answer I got was the careful one: keep what I have, maybe try an AI app or web service as a side project to see how it feels, no need to disrupt anything that’s already working. Reasonable advice, but it didn’t feel right. It felt like the kind of answer that protects you from making a mistake but doesn’t actually look at your situation.

So I pushed back. Asked again, more directly: do I really need to be paying for this? The next answer swung the other direction completely (cancel everything, move it all to a free service, but here’s how to set it up properly). And then it kept going. Next thing I saw was a whole stack of new tools to set up, configure, connect, and automate the whole thing. I had to stop the conversation and say “Isn’t this overkill for my simple websites? Why are we building a software factory to host them?”

We worked backward from there. Pulling apart the over-engineered version, keeping the parts that made sense for what I had and dropping the parts that didn’t. What I ended up with was a system that works for me (text files I could edit from anywhere, automatic deploys whenever I made changes, free hosting), sized appropriately for four small sites. That’s not where I started but I’m glad I ended up there. We swung from a conservative answer saying “don’t change anything,” to an aggressive answer telling me I needed a whole new technical stack. But like most great working conversations, we found the right solution in the middle.

My new “right-sized” setup turned out to be much more capable than what I’d originally asked about. I moved my blog off a paid platform onto the same setup, rebuilt my consulting site on it, and it was my first foray into building with AI so I could manage my website with a conversation.

I’m not sure I’d have gotten there if my conversation with AI had gone cleanly. The first answer was safe, the second was overkill, and the right one only showed up after I pushed back a few times and we worked to figure it out together. Which is pretty much how every good working conversation I’ve ever had has gone.

And now I don’t have to pay thirty bucks a month to GoDaddy anymore.