Fidelity
Over the last few months, I’ve heard a dozen variations of the statement: ‘AI output feels like it’s polished and ready to go but feels a little off.’
The statement stuck with me because they echo something I noticed years ago.
The fidelity of your output needs to match the maturity of your thinking.
It isn't because the AI skipped the work.
It's because you asked it to.
Said another way, your design, written draft, or whatever you’re making shouldn’t look more ‘done’ than where your current thinking lives.
What do I mean by fidelity? I’m speaking about how finished something appears. If it’s design, I mean the aesthetics or the polish of the draft. If it’s writing, it’s the rhythm and grammar of the piece. In architecture it may be a rough exterior sketch instead of a 3D rendered video flythrough. Low fidelity is primitive and gloriously disposable. High fidelity feels expensive and time-consuming and we’re all very resistant to throwing that away.
Until recently, this problem largely solved itself. Polish was expensive and it simply took too long to make unfinished thinking look finished.
When creatives start out, their skills are still developing, and it takes longer for iterations to look or sound any good. Their abilities will meter their progress. If they are diligent and stick with it, the thought process and output will progress in tandem. No client or colleague will mistake the draft for done before it’s done. The shitty first draft will look like a first draft.
Take a design example. I used to design a lot of book covers. On one level they were straight-forward. Once I knew if it was a hardcover, paperback, or spiral bound, the desired dimensions and some other basics, I could knock out a convincing mockup in a couple hours. But I insisted on showing the editor and author a hand sketch whenever possible (or six if I had time). I learned that working through concepts with them always yielded a stronger product. As an author now, I know that capturing the meaning of 20,000 words onto a 9X12 cover takes some time. I wasn't trying to save time. In fact, It took longer. I was trying to delay certainty. The sketches invited conversation. The polished mockups invited approval.
It turns out this isn't unique to graphic design.
Later I moved into User Experience design. That progression typically goes from whiteboard to wireframes to clickable prototype to visual interface design to development. In testing, I figured out quickly that once a user saw a seemingly ‘finished’ prototype, the conversation shifted. Instead of asking, “does this solve my problem” they asked “can that button be green.” I learned the cruder the mockup the more meaningful the feedback and ultimately the design became.
Writing is a deceptive example but no less true.
Maybe it’s my design roots, but I need to make an outline, for myself and for my clients. When presented with an outline, I can move things around, make sure my argument holds up, and that I didn’t lose the throughline. Once I've spent an hour polishing a sentence, I'm much less likely to realize the right edit was deleting it.
So what happens if you’re still following your process, but your output looks finished? It’s why so much AI output looks off. The surface is further along than the thinking underneath it. As writers, designers, architects, etc., our clients hire us to shepherd their ideas. However, that’s not the words they use. The words say they want a communication, a brochure, a logo, plans for a pool house.
Once they see something that looks or sounds like it’s good to go, they stop asking what if.
If we’re honest, so do we. Jumping steps changes which questions get asked.
The creative process has never been about making something look finished. (Wanna see a designer lose it? Ask them to make something look pretty.)
It's about discovering what belongs, what doesn't, and what parts of an idea deserve to survive. Every worthwhile creative endeavor is a process of discovery, refinement, subtraction, and addition. AI is remarkably good at helping us accelerate the process. It's an incredible research assistant, a brainstorming partner, an editor, a proofreader, and yes even a capable polisher if you must.
What it isn't good at is replacing the thinking that happens before the polish. That's not the tool's fault. It will happily do exactly what you ask of it. AI can absolutely work in low fidelity, but we rarely ask it to.
I prefer to smell the graphite in my sketches or chicken-scratch an outline, but do what works for you. The problem comes when we use technology to bypass the beautiful mess in the middle. That's where ideas earn their place.
Because polish has never been the goal. The goal has always been to discover and preserve the meaning that made the idea worth having in the first place.
Want to stay in touch? Join my mailing list.