I Did a Thing: The Friction of Letting an Agentic Workflow Do the Work

Building with agents feels like farming badly. The agentic workflow isn’t broken. I’m leading something I don’t understand yet, and that’s a leadership problem.

If you watch Clarkson’s Farm, you’ve heard the catchphrase “I did a thing!” when Jeremy has performed a miracle by successfully executing a relatively basic farmer’s task. That’s how I feel right now.

I’m building my own app using AI and I feel pretty lost and under-equipped to do this properly. Like in Clarkson’s Farm, I know that if I handed the same work to someone like Kaleb, who actually knows what he’s doing, it would be in a better state.

I’m at a crossroads. On one hand, I want to go back to the way I’m familiar with. On the other, I know I need to continue transforming the way I work.

The Pull to Go Back

I want to go back to drawing. Back to sketching things out myself, by hand, where I have control over the outcome and the output with my own skill set.

It feels easier. Give it to me, I’ll do it.

But I want to be careful here, because this is the part where a lot of people stop and conclude that the agentic workflow doesn’t work. I don’t believe that. If I said, “I can’t make decisions, so I’m going back to the way it was,” I’d be dismissing the whole value proposition, and that would be wrong.

The problem isn’t that the agentic workflow is broken. The problem is that I lack the experience and the understanding to build products this way. I can choose to retreat or to redesign the way I work.

The value I bring needs a new way to be released.

I Have Been in This Transition Before

What makes this recognizable to me is that I’ve experienced this role shift before.

When I moved from being an individual contributor to being a design leader, it was a similar shift. Not the same shift, but a similar one. I had full control over outcomes with my own hands. Then I needed to let that go.

I remember the exact pull I feel now. Just give it to me and let me do it.

What I had to do instead was unlearn. And in place of that, I had to reinvent myself. I had to train and educate myself into being a better leader, a servant leader. I had to articulate and transfer my know-how to the people I was now handing my work over to. And I had to put systems in place to set them up for success, so that the whole way of working improved.

That’s exactly what I’m facing right now, except the people are agents.

The Part I’m Actually Losing

The biggest lesson I took from moving into agile development was three words: transparency, inspection and adaptation. The empirical pillars. You make the work visible, you look at it thoroughly and you change course based on what you saw. It’s how you lead something you’re not personally executing.

All three are breaking for me.

How the agents got to the decisions they made is pretty hard to see. You throw an assignment into a black box and something comes back. You lose control of the overview. You lose the continuity. You lose the clarity about how we’re building this and why we’re building this.

Because I don’t understand a lot of the code, I hand off work to something I can’t properly judge. It does take away the risk of micromanagement, which is a positive side-effect, but it also means I have very little to inspect.

What surprised me is that it happens in my design too. Design is my own domain. I know this work. And still I find myself over-explaining, and still I can’t rely on the same level of output. I haven’t properly instructed the agents with the skills, prompts and libraries to really work and think like me.

This results in constant adaptation and reattempts, but without transparency and inspection. And without those two, adaptation isn’t really adaptation. I’m not changing course based on what came back. I’m just asking for it again.

I’m losing agility.

Left or Right

Being agile isn’t about being fast. With agentic workflows your time is compressed, your own thinking gets rushed. You’re constantly facing decisions and outputs from agents that come back so fast. The time to think in between is just gone.

And it feels like I’m moving fast. But when I look at my actual progress, I find myself looping and losing control. Agility is about delivering value sustainably. Definitely not about burning tokens rapidly.

The hardest part is that I know I’d actually be faster if I stepped away from the generation for a moment. Sketch things out. Give a proper drawing as input. Create a better frame for the agents to operate in.

But that’s in direct conflict with the time I have available.

There’s this constant intersection. I can go right and generate more, or I can go left, stop doing and think a little. My brain says go left. And then right there is the dopamine shot, the next generation, something new, something potentially better, maybe this one is the fix.

Going right feels like it takes me exactly where I want to be. It usually doesn’t.

It’s almost like a casino. I’m constantly choosing red or black. I’m gambling my time and my tokens with a lot of hoping.

What This Is Really About

I’ve generated a lot of iterations. I keep looping. And underneath all of it is something simpler than a tooling problem.

You can’t lead, challenge or guide people and agents in the right direction if the direction itself is ambiguous.

That’s what I’m struggling with. I know that I’m responsible and need to guide something I don’t comprehend enough myself. So the question becomes: do you double-down on the workforce you’ve paid for, in this case agents, and just trust them and give them more? Or do you stop and build your own clarity first?

I’m not going to hand you a technical answer here, because I’m too early in this process to advise on that. Many of you reading this are further along than I am. Some are not.

But I do know the leadership version of this answer, because I’ve had to learn it before.

As a leader you have to develop the skill and the character to accept that you don’t know. And then to tell the people you need some time to formulate your thoughts, your vision, your direction, so you can properly guide them and send them the right way. If you don’t develop that part, and if you’re not creating room for it, you end up directing people in directions without the honest answer about what it should be.

That applies to agents too. Maybe even more.

Closure

Maybe the thing to do isn’t to take advice from me. Maybe it’s just to reassess your own point of view.

Are you gambling, or are you creating clarity?

Take a break. Create the room to think, to not know, to doubt, to figure things out. And then treat developing your leadership skills as part of your agentic experience, not as something separate from it.

Create that clarity, build the systems and make room for the thinking that lets you guide both people and agents. Guide development in a direction you can actually see, inspect and adapt.

I’m still going to be the guy on the tractor shouting that “I did a thing.” But there’s more work to be done to make this field yield.