What Can’t the AI Summarize?
If your work is just good enough pages, a machine is coming for the gig. If it's discovery, no AI summary engine can do what you can.
I read Model Behavior, Substack’s new interview series with tech journalist Taylor Lorenz, about how she uses AI, what it’s doing to her business, and how scribblers are pretending not to notice.
The whole conversation is worth reading, but one part resonated.
Lorenz said a New York Times editor once told her the point of journalism was to expose new information. That was the job. Expose what the rest of us didn’t know yet.
That distinction gets useful once you drag it into screen scribbling, game narrative, worldbuilding, comics, podcasts, and whatever strange story machine you’re assembling in the garage.
The Discovery Comes First
AI is very good at rearranging what’s already available.
It can summarize a genre, mimic a three-act outline, spit out twenty character names, bang out a competent scene, scrape a pile of lore, or build you a recommendation queue like Spotify’s. That makes a lot of creative labor cheaper. It also reveals which parts were already close to being commodities.
Ouch. Some of those parts took us years to learn. Oh well.
Of course the machine can scribble a decent scene. The real test: was your value ever just the existence of pages on paper?
For a screen scribbler, the discovery might be a human contradiction nobody else has noticed, time spent with a community that’s only ever appeared onscreen as a cliché, or the dramatic engine hidden inside a real event, family secret, or a strange job.
The screenplay is the delivery system. The discovery came first.
For a narrative designer, the discovery may not be a line of dialogue. It might be a behavior that emerges when players meet a system: a faction rule that creates conflict without a cutscene, a mechanic that forces a player to reveal what they value, a world condition that generates stories the designer never scribbled.
An AI can give you fifty quest ideas. It can’t sit inside your playtest and feel the moment everyone ignores the quest, adopts the disposable creature, and turns your tragedy into a rescue operation. That moment is the new information. The job is noticing and rebuilding around the truth the players just exposed.
For an independent creator, the discovery might be the world itself: a set of people, rules, images, obsessions, and artifacts that could only have come from your peculiar archive.
Taste Must Produce Something
This is where taste enters the picture, though the interview carries a warning here too.
Lorenz described part of User Mag’s value as access to her brain, the stories she personally finds interesting. She admitted readers will probably choose a feed tailored to their own interests over her taste, given the option. A hard admission to make in public.
Taste alone isn’t enough once a machine can manufacture a private recommendation engine for every reader.
Your taste needs to produce something worthwhile. It must connect two ideas that weren’t connected before, lead you somewhere inconvenient, and become a character, a field report, a strange mechanic, a new world, an argument, an image, or a story that didn’t exist until your taste acted on that material.
“Here are ten things I like” is replaceable.
“Here is the thing I made because I could not stop thinking about number seven” is a different proposition.
Bricks vs. Architecture
The interview also raises the transparency question. Lorenz uses AI heavily and talks openly about it, while some scribblers condemn the tools in public and trade prompts in private. I have no interest in the purity Olympics. The tools are here. I use them.
Did the machine sort research, stress-test structure, check continuity, generate options, or produce the thing the audience believes you scribbled? Those are different uses. If we can’t describe the difference, we haven’t decided where the line sits.
The architecture includes intent. It means deciding what the work is for, protecting that purpose through revision, and knowing when a polished answer is wrong. It means taking responsibility when ten plausible choices sit on the table and only one belongs in this story. Judgment work.
The Relationship Is the Work
Lorenz’s advice to young journalists is to build a direct relationship with an audience. Screen scribblers have historically been able to ignore that. We sold upstream. The studio, network, producer, or financier worried about reaching the people.
That bargain is getting wobbly.
The creator with an audience, an archive, and a way to reach people without permission has more options than the creator with a perfect file waiting in an inbox. Nobody needs to become a dancing content monkey. The relationship around the work is becoming part of the work itself. Show people what you notice, let them see what you’re building, and give them a reason to care before the giant finished object arrives.
I’ve started asking a new diagnostic question: what am I discovering here that can’t be obtained by a better LLM summary?
If the answer’s nothing, the project may be fun. But I shouldn’t confuse competent assembly with a defensible creative practice.
Go into the world, notice the things, make the connections, build the artifacts, form the relationships. Then use every tool you can get your dirty little creator hands on.
ABS. Always. Be. Scribbling.