I’m currently leaning over a piece of textured vellum, my charcoal stick snapped into 8 jagged pieces because the witness just changed their story for the 18th time. My fingers are stained a deep, bruised gray, and I can feel the grit under my nails as I try to capture the specific way this man’s brow furrows when he lies. It’s a physical battle. Art, even in a courtroom, is a confrontation between the hand, the eye, and the messy reality of the subject. But lately, I’ve been hearing a different kind of noise-not the scratching of pencils, but the frantic clicking of keys from people who think they’ve discovered a new language. They call it prompt engineering. They talk about it as if they are whispering secrets to a god, but from where I’m sitting, it looks more like they’re just arguing with a very stubborn, very confused machine that doesn’t know the difference between a human finger and a baked good.

Last night, I tried to send my editor the sketches from the $498-per-day hearing, and in my rush to prove I could handle the digital transition, I sent the email without the attachment. It’s a classic Nova move. I spent 48 minutes crafting the perfect subject line, agonizing over the ‘professional yet urgent’ tone, only to fail at the most basic mechanical level. This is exactly what’s happening with AI right now. We are so obsessed with the ‘magic words’-the incantations we think will unlock the latent space-that we forget the tool is supposed to work for us, not the other way around. I watched a colleague spend 8 hours trying to get a popular image generator to draw a simple scene of a woman drinking tea. By the 88th iteration, the woman had 18 fingers and the tea was somehow clipping through her chest. My colleague was convinced he just hadn’t found the right ‘negative prompt’ yet. He was convinced he needed more ‘weighting’ on the word ‘anatomical.’

The Vending Machine Fallacy

Let’s be honest: if you have to tell a professional illustrator ‘don’t give her three arms’ 28 times, you don’t have an illustrator; you have a problem. The hype around prompt engineering as this high-level, elite skill is a massive distraction. It’s a form of gatekeeping that celebrates our ability to navigate the bugs of a specific system. It’s like being proud that you know exactly where to kick a vending machine to make the chips fall out.

Sure, you got the chips, but wouldn’t it be better if the machine just worked? We’ve entered this strange era where we think ‘mastering’ a tool means learning its quirks rather than its capabilities. I see people selling courses for $888 on how to ‘talk’ to AI, as if these models are sentient beings with moods and secret handshakes. They aren’t. They are statistical distributions of pixels and tokens. If the model requires you to type ‘masterpiece, 8k, trending on ArtStation, photorealistic, Unreal Engine 5’ just to get something that doesn’t look like a nightmare from 1998, then the model is failing you.

The Lost Moment of Truth

I remember 38 years ago, when I first started as a sketch artist, people said digital cameras would kill the profession. They didn’t, because a camera still requires a human to choose the angle, the moment, the truth. But prompt engineering feels different. It feels like we’re trying to automate the ‘truth’ part by guessing which combination of characters will trigger a specific set of weights. It’s a tedious trial-and-error loop that masquerades as creativity. We are losing the ‘why’ in favor of the ‘how.’

“

When I’m in court, I’m looking for the 18-millisecond window where a defendant’s mask slips. An AI doesn’t know what a mask is; it only knows that ‘mask’ is often near ‘face’ in its training data. So we argue with it. We add brackets. We add colons. We increase the CFG scale to 8.8 and hope for the best. It’s exhausting. It’s like trying to paint with a brush that decides to turn into a fish every 58 seconds.

This brings me back to my failed email. The intent was there. The ‘prompt’ (the body of the email) was perfect. But the tool (the attachment button I missed) rendered the whole effort useless. Most of the ‘engineering’ we do today is just us trying to attach a file that the machine keeps dropping. We think we are being clever, but we are actually just compensating for the fact that many of these models weren’t built with the actual creative process in mind. They were built to show off what’s possible, not what’s useful.

This is why the conversation needs to shift. We shouldn’t be celebrating the people who can write a 1008-word prompt; we should be looking for the tools that understand us the first time. The real skill isn’t in mastering the arcane syntax of a quirky machine, but in recognizing when you’re using the wrong machine for the job.

For those who are tired of the ‘incantation’ phase of AI, moving toward tools like

NanaImage AI represents a shift toward actual utility. It’s about choosing a model that aligns with your intent rather than one that requires you to be a digital exorcist. When the tool is right, the ‘engineering’ disappears, and you’re left with the art. Or at least, you’re left with a sketch that doesn’t make the judge look like an eldritch horror from the 8th dimension.

I’ve spent 48 years observing people, and one thing I’ve noticed is that we love to make things harder than they need to be so we can feel like experts. We did it with Photoshop filters in the early 2000s, and we’re doing it now with prompts. We want to believe there’s a secret sauce because if there is, it means we have control. But the reality is that the current state of prompting is a limitation, not a feature. It’s a bridge built of matchsticks that we’re trying to drive a semi-truck over. I’m tired of the matchsticks. I’m tired of the 18-step tutorials on how to get ‘soft lighting.’ I want the lighting to be soft because I asked for it, not because I found a loophole in the code.

The Direct Line

Yesterday, I went back to that witness sketch. I didn’t have to ‘prompt’ my charcoal. I didn’t have to specify ‘hyper-realistic skin texture’ or ‘no extra limbs.’ I just looked, and I drew. There was a direct line from my brain to the paper. That’s what’s missing from the ‘engineering’ hype. We are adding so many layers of abstraction-so many keywords and weights-that we’re losing the direct line. We’re becoming translators for a machine that speaks a language of 88-million-dimensional vectors, and we’re losing our own voices in the process. It’s a strange trade-off. We save time on the execution but spend it all on the negotiation. We’re not creators anymore; we’re just middle managers for algorithms.