AI's Overlooked Feature is its Most Powerful

Every few weeks, the tech world experiences what can only be described as a collective fainting spell.

A new AI model is announced (in this case, Google’s Nano Banana) and before the demo has even finished buffering on YouTube, a chorus of doom is already in full voice.

“This will kill creativity,” they say.  “Kill the human spirit… Kill Adobe!”

One can practically hear the tolling bells for entire industries.

There’s a ritual to this by now: the launch video lands; the breathless headlines follow; the op-eds sharpen their knives.

The technology is presented as a sort of messiah and executioner rolled into one, destined to both liberate and obliterate in equal measure.

And then… inevitably… people actually try it.

Have you?  Tried doing with Nano Banana what the demo promised?  Yes, so have I.  And no, it doesn’t deliver.  Not yet.

A Stanford CRFM study this year found that 92% of generative AI feature launches underperform their demo claims in early evaluations.  A reminder that Nano Banana isn’t a glitch in the system; it IS the system.

But this is all choreography.  And it’s been going on since the first PowerPoint pitch deck met a venture capitalist’s laser pointer.

CB Insights counted 387 GenAI product feature announcements between January and September 2025.  That’s more than one per day.

At that pace, even the most over-caffeinated futurist could be forgiven for feeling faint.

The Quiet Superpower

While the industry hyperventilates over each new feature, something less theatrical, but far more consequential, is happening beneath the headlines.

From their first day in the world, these models have excelled at a very particular task…

Drumroll please…

Structuring previously un-structurable DATA!

Yeah… I know… it doesn’t sound that exciting…. But it is.

They are prodigious librarians of chaos.  They can take a thicket of language, story, images, or dialogue and turn it into something you can query, map, and build on.

That’s their native skill.  And for the creative industries, it’s TRANSFORMATIVE.

If you’re wondering, creative data is almost entirely unstructured.

Are you seeing where I’m going with this?

IDC estimates that over 90% of enterprise data is unstructured, and that the figure is closer to 95–97% in media and creative industries.  A 2025 MIT Sloan survey found that companies using GenAI primarily to structure data, not to generate it, move from insight to action three times faster than their peers.

A screenplay has a semi-structure, sure, but nobody would confuse Casablanca with Anora.  Yes, both are love stories, but to equate their data structures is to miss the point entirely.

What unites them is not the structure, but the mess.  And AI is finally giving us the tools to turn that mess into something legible.

The Context Window Revolution

Take the recent release of Claude 4.5.  The headlines swooned over the usual laundry list of “smarter, faster, better” claims.

But the real breakthrough was almost buried: the context window finally works.

Claude 4.5 can process up to 1 million tokens, and Anthropic claims 98% accuracy in recall across large documents, compared to roughly 55–65% in previous iterations. Their internal benchmarks show a 70% improvement in mid-document reasoning, effectively closing what I’ve long thought of as the “amnesia gap.”

This is HUGE!

For years, feeding long documents into a model was like whispering to someone with short-term memory loss…  They’d remember the opening, maybe the closing, and little in between.  Now, for the first time, these models can hold and reason across large swaths of information in one go.

Powerful… and nobody’s talking about it.

From Scripts to Systems

At Koobrik, this is exactly where we’ve staked our flag.  While others chase the mirage of AI-generated genius, we’re databasing the data inside scripts for studios, independent producers, and agencies.

Every prop, scene, location, stunt, and character… structured, searchable, and ready.

On the analytical side, this structured creative data allows us to correlate box office performance, actor value by territory, story structures by genre, and surface development insights that once required armies of assistants and months of manual work.

McKinsey estimates that automating metadata extraction in media workflows can cut 30–50% of pre-production time, especially in script analysis, casting prep, and budgeting.

For the first time, creative data is genuinely on the table.

What once took thousands of human hours can now be synthesized in days… sometimes hours.

The Real Shift

So while the industry keeps peering anxiously at the horizon, waiting for some shiny new feature to either destroy or save us, the real story is already happening quietly at ground level.

The models are not yet the auteurs of our age.  They are the archivists.  The cataloguers.  The sexy Librarians.

The unsung data clerks making the creative industries legible to themselves for the first time.

And that’s not something to panic about.  It’s something to build on.

✨ Closing Thought

In time, Nano Banana will probably do everything its demo promised, just as Photoshop eventually did, and the iPhone eventually did, and every other “underwhelming” first draft of the future has done.

(Remember how shit the first iPhone was?  That beer drinking app was the only thing it did well…)

But the creative companies that win will be the ones quietly turning their unstructured creative gold into structured, queryable, actionable intelligence.

Forrester’s 2025 agency benchmark found that firms applying GenAI to data intelligence, not content gimmicks, saw 12–18% margin gains.

It’s not the flash that wins.  It’s the plumbing.

That’s the story worth paying attention to.