Getting Comfortable with Uncertainty

A few months ago I built an agent to handle a chunk of my workload automatically. It worked. And the reward for that was more work landing on my desk, because now I clearly had the capacity for it. What I didn't write about was the part before it worked, the stretch where I genuinely didn't know if it would work out. That not knowing is the part I keep coming back to.

The uncertainty is not one thing

When people talk about AI transformation, they usually mean it at the organisational level. New tools, new processes, a plan nobody's quite sure of yet. But the same uncertainty shows up at every scale I'm operating at right now.

There's the career-level version: what does my work look like in two years, when the skills I'm building today might be table stakes or might be irrelevant. There's the process-level version: how does a workflow change once AI is genuinely part of it, and nobody has run that experiment before, at least not in a way you can just copy. And there's the small, daily version: will this specific agent, this specific automation, actually do what I want it to do.

I used to look at these as different problems, but now I see it as the same discomfort showing up at different sizes.

We have everything except the answer

Here's what's strange about this moment. It's not that we're flying blind. I, together with my colleagues, have years of experience, a decent amount of technical knowledge, and genuinely good intentions about how this should go. Most people I know building with AI right now are in the same position. We are not underprepared.

We just don't have a tried and tested playbook yet, we're the ones writing it and evolving with it, in real time, by doing the thing before we know if it works.

That's a strange place to sit. Usually competence comes with some certainty attached. You know your field, so you can predict outcomes reasonably well. Right now that link is loosening. You can be genuinely skilled and still have no real idea whether the workflow you're testing this week will be the way anyone does it next year.

Doing it anyway is the actual skill

The part I've come to enjoy, if I'm honest, is testing something without knowing the outcome and then watching it either work or fall apart. There's a specific kind of satisfaction in building an agent, having no strong prediction either way, and then seeing it do exactly what you hoped. It doesn't happen every time. But when it does, it's a different feeling than shipping something you were already confident about.

I don't think this discomfort resolves into certainty at some fixed point, the way learning a new skill eventually does. I used to expect that: get through the uncertain phase, arrive at competence, feel settled. What I'm noticing instead is that the ground keeps shifting under the thing you just got comfortable with, and you adjust again. Not because you failed to learn it properly, but because the thing itself hasn't finished changing.

Maybe that's just what working with AI is right now, for anyone actually building with it rather than watching from the side. The comfort isn't in reaching solid ground. It's in getting used to how it feels to not have it, and building anyway.

I don't know exactly when, or if, that feeling settles into something steadier. I'm still in it, same as everyone else doing this work. But I'd rather be honest about that than pretend there's a map I'm quietly following.

Thought pondered by Sarah exploring the intersection of AI, creativity, and human wellbeing