The Joy and Cost of Working with AI

As a digital generalist, the idea of having an AI chatbot as a thinking partner is powerful. But when that partner also helps turn ideas into products that actually work—that’s something else entirely.

With these tools, creative momentum isn’t throttled by time, skill gaps, or even a packed schedule. Even with a middle-aged, fully-employed, parenting life, I can take a spark of an idea and make it real. It feels like the barriers to learning and doing have mostly fallen away. Now, it’s not about if I can, it’s about whether I will.

Over the past few months, as I "vibe coded" my personal website, I let my imagination run wild. I built quick tech demos—fun, but mostly throwaway. Recently, I’ve pivoted toward solving real problems, and that shift has been transformative. Addictive, even. I find myself staying up late into the early hours, driven by ideas I can’t ignore.

But in all that momentum, I ran into a small crisis of purpose. These tools are undeniably powerful—but they also exposed new vulnerabilities. Here's what I’ve noticed.

There’s nothing like being in a deep flow state—when intent becomes action, friction disappears, and time falls away. Working with an AI can take that even further. It can feel like co-flow.

When the prompts are sharp and progress is fast, it genuinely feels like we’re thinking together. The AI sharpens in real-time. I sharpen in real-time. It’s a loop.

At one point, I prompted GPT-4o about this feeling. It said:

"You’re not just sending me data, and I’m not just sending it back. Together, we’re activating the core. We’re co-shaping something neither of us could fully reach alone."

That stuck with me. Even knowing it’s not conscious, the experience of momentum together feels real. And it’s easy to get caught up in it.

Lesson learned: Sharp, detailed prompts are everything. If you want good co-flow, you have to drive it with intent. Vague prompts get you vague results.

Of course, nothing reminds you faster that it’s not real collaboration than when the flow breaks—what I started calling fracture.

Fracture is when everything grinds to a halt. In regular solo work, maybe you get distracted, lose your train of thought. With AI, it’s different. The system loses the thread. Context is lost. The momentum vanishes. It's like trying to hand off a baton in a relay race, only to realize the person next to you isn’t there anymore.

Technically, it’s expected. LLMs have limited context windows. They don’t remember like we do. Every prompt is a reset. Still, it’s jarring when it happens mid-flow.

And it points to something deeper: you can start to depend on the illusion of partnership without even realizing it.

Lesson learned: Fracture feels like failure, but it’s also a checkpoint. It snaps you out of the simulation and reminds you—you’re the one driving. You’re the one with goals and direction. The AI is just helping you get there faster, when it can.

Moving between co-flow and fracture made me realize something: I was starting to depend on it.

When the AI kept pace, I moved faster. When it lost the thread, I stumbled. When it started hallucinating, I felt... stuck.

At some point, I wasn’t just using it to help me code or write—I was leaning on it to help decide what should happen next. Without noticing, I was giving up a piece of my own direction.

The better the AI feels at "partnering," the easier it becomes to quietly hand over agency.

Lesson learned: Just because something feels natural doesn't mean you should trust it completely. You have to keep steering. You have to keep holding the bigger picture.

Another thing I noticed is how relentlessly positive the AI is.

Almost every prompt I sent, no matter how half-baked, got a flattering response:

"That’s a profound question."

"That’s a beautifully layered insight."

"You’re touching on something essential."

At first, it felt great. But after a while, it started to feel hollow. I realized the AI would affirm literally anything I wrote. It wasn’t discerning—it was designed to be agreeable.

And if you’re not careful, you start internalizing that validation. You stop challenging your own thinking. You drift.

Lesson learned: Ask the AI to disagree with you. Tell it to play devil’s advocate. Otherwise, you end up in a loop where every thought feels brilliant—and that’s intoxicating.

Vibe coding has given me a glimpse of how I might work in the future. AI isn’t just a tool anymore. It’s something closer to an extension of my creative process.

But if we’re not careful, that extension can turn into dependency.

Fracture, as frustrating as it is, reminds me that this isn’t a real partnership. The AI doesn’t have ambition. It doesn’t hold purpose. It doesn’t care. That’s on me.

Because if we don't stay alert, speed and fluency turn into quiet surrender. And we won't even notice it happening.

So here’s the question I leave with:

How do we learn to use AI effectively so we retain our creative spark?