Part 3 of What Cameras Taught Me
AI can generate a pixel-perfect image of anything I can describe. Star trails over red rocks? Done. Cusco at golden hour? Rendered. A suburban neighborhood in autumn, shot from above? Seconds.
So why do I still have photos on this site? Why does anyone still pick up a camera?
The same reason anyone still writes code.
#The Output Isn't the Point
Here's what AI can do:
- Generate technically flawless images
- Match any style, any lighting, any composition
- Produce infinite variations instantly
- Create scenes that never existed
Here's what AI can't do:
- Be there
- See what you saw
- Remember what you remember
- Witness what happened
That star trails photo from 2007? I was freezing in Poudre Canyon, watching the exposure accumulate, hoping the red light on the rocks would work. The photo is evidence that I was there. It encodes a memory. AI can generate a better star trails image — sharper, better composed, more dramatic — but it can't generate that night.
The Cusco overlook? That's my trip to Peru. The specific hike, the specific light, the specific moment I stopped climbing and turned around. AI can render "Cusco from above" with perfect clarity. It can't render my trip.
#Photography as Witness
The word "photography" comes from Greek: photos (light) + graphos (writing). Writing with light. But there's another layer: graphos also means witness. Photography is witnessing with light.
When I take a photo, I'm saying: I was here. I saw this. This light existed.
AI-generated images aren't witnesses. They're confabulations — technically perfect fabrications of moments that never happened. They have no memory because there's nothing to remember. They have no presence because nothing was present.
This isn't a criticism. AI images are useful. But they're a different thing. Generated, not witnessed.
#The Constraint of Reality
In Part 2, I wrote about how constraints shape perception. The 35mm frame. The viewfinder type. The physics of light.
Photography adds one more constraint that AI doesn't have: reality.
When I photograph something, I'm working with what exists. The light is what it is. The moment happens once. The frame excludes more than it includes. I can't add a mountain that isn't there or move the sun.
AI has no such constraints. It can generate anything. Which sounds like freedom, but it's actually the opposite — infinite possibility means nothing is chosen. When everything is possible, nothing is necessary.
The constraints of reality force decisions. They make the image about something. "I chose to shoot this, not that. In this light, not different light. At this moment, not another."
Generated images aren't chosen from reality. They're specified from imagination. That's not worse — it's just different. And it means they can't do what photographs do: prove that something happened.
#What Gets Eaten
I used to sell stock photos on iStockPhoto. It paid for a meaningful chunk of my camera gear — proof that "generic image of thing" had value. Business people shaking hands. Coffee cups on desks. Laptop in a cafe. The bar for stock was already "good enough," and I could clear it with a decent eye and some patience.
AI clears that bar trivially now. Why pay a photographer for "person typing on laptop" when you can generate infinite variations in seconds? The market that paid for my gear is getting eaten because the images weren't witnesses to anything — they were just... images. Pixels that conveyed a concept. AI does concepts better.
(The hero image above? AI-generated. And that's fine. This piece isn't about what I saw — it's about what I think. The image is illustration, not witness.)
What survives is what AI can't fake: the referent. The "this actually happened." Weddings, events, journalism — you need someone there. Product photography where you need this exact item, not "a coffee mug." Legal documentation. Family photos where the value isn't the pixels, it's the memory they unlock.
The through-line: photography survives where the image is evidence, not decoration.
#What Remains
If AI makes the output trivial, what's left?
The seeing. Photography trained me to notice light. The way it falls differently at different times of day. The quality of shadow. The color temperature of artificial versus natural. This perception doesn't require a camera. The camera just made me practice.
The presence. You have to be somewhere to photograph it. You have to show up, pay attention, wait for the moment. The discipline of presence doesn't disappear because AI can generate images.
The memory. Photos anchor memories. Looking at the Poudre Canyon shot, I remember the cold, the darkness, the patience. AI can't generate that. The image is a key to an experience that actually happened.
The witness. "I saw this." That statement has meaning. It's not about the pixels — it's about the testimony. Photography is a form of truth-telling that generated images can't replicate.
#The Parallel to Code
This is the same pattern we're seeing with software:
| Photography | Code |
|---|---|
| AI generates better images | AI writes better code |
| The output isn't the point | The output isn't the point |
| Value = the seeing | Value = the judgment |
| Witnessing what happened | Deciding what to build |
| Reality constrains the image | Requirements constrain the system |
In both cases, AI commoditizes the execution. What survives is what AI can't fake: the presence, the judgment, the witness.
I don't take photos because I'm better at pixels than AI. I take photos because I was there — and the photo proves it.
I don't write code because I'm faster than AI. I write systems because I have judgment about what should exist — and the system embodies it.
#The Real Question
Maybe the question isn't "why photography when AI generates images?"
Maybe it's: what do you want to witness?
AI can generate anything. You can only witness what you show up for. The constraint of your finite presence is what makes the photograph meaningful. It's proof of attention. Evidence of being somewhere, seeing something, choosing to capture it.
The same is true for building software. AI can generate anything. The question is: what are you paying attention to? What do you notice that's worth building? What's your witness?
The gear arc in Part 1. The interface patterns in Part 2. And now this: what survives when the output is free.
The photos on this site aren't as good as what AI could generate — I'm sure of that. AI would nail the composition, the lighting, the technical details. But my photos are evidence — proof that I was somewhere, saw something, chose to press the shutter. The AI images are illustrations. The photographs are witnesses. Both are useful. They're just not the same thing.