Right now, in your phone, there's a photo of someone you love next to a screenshot of a Wi-Fi password from a hotel you'll never visit again. They're stored the same way. They take up the same space. If you search "August 2024," they show up side by side.
Your phone doesn't know one of those is permanent and the other was useful for ninety seconds.
Neither does your wiki.
How we got here
In 1999, a roll of Kodak Gold cost you 24 exposures. Each frame was roughly fifty cents — film, developing, prints. You composed the shot. You waited for the light. You pressed the shutter knowing that frame was spent whether the photo turned out or not.
The constraint was the feature. Fifty cents and a finite roll meant you didn't photograph your parking spot. You didn't take eleven shots of the same angle to pick the best one later. The economics of film were a quality gate disguised as a limitation.
Digital cameras lowered the barrier. Storage was cheap, the LCD preview was instant, and you could take two hundred shots on a vacation instead of twenty. But there was still a natural pause in the system — a speed bump that did more work than anyone appreciated. You had to get the photos off the camera. Find the USB cable, plug into the laptop, open the import dialog, pick a folder. In a pre-cloud world, that ten-minute ritual was a moment of implicit curation. You'd glance at the thumbnails. You'd skip the blurry ones. You'd drop them into a folder called "Beach Trip 2007" and move on. The friction of the transfer step was quietly doing the work of an archivist.
Then the phone ate the camera, and the cloud ate the folder.
Suddenly the marginal cost of a photo was zero and there was no transfer step. The shutter, the storage, and the archive were the same device. Every photo you took was instantly, permanently, undifferentiatedly saved. A photo of your kid's first steps lived in the same infinite scroll as a screenshot of a tracking number you needed for forty-five minutes.
This is happening to everything your company knows
Knowledge management is following the same arc, about fifteen years behind.
The filing cabinet era was film. Creating a document had real friction — typing, printing, routing, filing. That friction was a quality gate. Nobody typed up a memo, walked it to the filing room, and slotted it into a cabinet unless the information mattered enough to justify the effort.
The wiki era was the digital camera. Creating documents became easy, but there was still friction. You had to choose a space, build a page, format it, think about where it lives. The ten minutes of setup wasn't much, but it was enough to make you pause and consider whether the thing you were writing down was worth writing down.
The AI era is the phone.
Meeting transcripts auto-generated. Status updates drafted from chat threads. Documentation written from a prompt. Procedures extracted from conversations. The marginal cost of creating a document is approaching zero, and the transfer step — the moment where a human decides whether this thing deserves to exist in the knowledge base — is disappearing.
The volume is about to explode. And just like your camera roll, the system has no concept of the difference between a compliance policy that will govern your company for the next decade and a meeting summary that nobody will ever read again.
The industry has a name for this
In information management, the acronym is ROT: Redundant, Outdated, Trivial. Analysts estimate that data ROTs at roughly 2% per month — meaning a third of your knowledge base could be noise within a year. Some studies put the share of enterprise data that qualifies as ROT at 33% or higher.
But the acronym, clinical as it is, understates the problem. ROT isn't just clutter. It's an erosion of trust.
Think about the last time you searched your company's wiki for a procedure. You found three pages with similar titles. One was from 2019. One was from 2022. One was a draft that was never finished. Which one is current? Which one reflects actual practice? You don't know. The system doesn't know. So you do what everyone does — you message a colleague and ask them. The wiki becomes a suggestion engine that you verify through human conversation, which defeats the entire purpose of having a wiki.
This is something every engineering lead, every knowledge manager, every new hire who's been burned by a stale runbook recognizes. And it's not any one tool's fault. It's the structural consequence of a system where every document looks the same, is surfaced the same way, and has no concept of whether it's still true.
Now add AI to this system. An AI agent that ingests your knowledge base doesn't check the "last updated" date and squint skeptically the way a human does. It treats the 2019 procedure with the same confidence as the 2025 one. It can't tell the difference between an active policy and an abandoned draft. The ROT doesn't just confuse humans anymore — it poisons the machine.
You could just organize it. But you won't.
Here's the thing your phone already knows about you: you have albums. Your phone has a "Favorites" feature. Google Photos will auto-generate collections from trips. Apple offers "Memories." Every photo app in the world has curation tools.
When's the last time you used them?
If you're like most people, you have a "Favorites" album with eleven photos in it from 2021 when you briefly tried to maintain it. You might have one vacation album that's half-complete. And then there's the undifferentiated stream of 14,000 images that you scroll past every day, occasionally finding what you're looking for, more often giving up.
The tools exist. The discipline doesn't. And it doesn't because curation is a separate chore from creation. Taking the photo is one workflow. Organizing the photo is a different workflow, done at a different time, requiring a different kind of attention. Any task that's separate from the main workflow and doesn't have an immediate consequence for skipping it will not get done at scale. This isn't a character flaw. It's a structural certainty. The behavioral economics are unforgiving: the future cost of disorganization is abstract, but the present effort of organizing is concrete. Each individual photo seems too trivial to bother sorting. And the degradation is so gradual you never hit a crisis point that forces action. You just slowly stop trusting the system.
Your wiki is exactly the same. You could archive old pages. You could set review dates. You could establish a governance process where content owners revisit their pages quarterly. Every wiki has an "archive" button somewhere.
Nobody clicks it. For exactly the same reasons nobody maintains their photo albums. The curation is disconnected from the creation, so the curation doesn't happen, so the ROT compounds quarter over quarter until the knowledge base is so degraded that people stop trusting it — and then stop using it — and then the organization's knowledge reverts to what it was before the wiki existed: tribal, oral, locked in the heads of people who might leave.
The glance layer
Game designers figured this out decades ago.
In a well-designed game, information exists in layers. The HUD — the heads-up display, the transparent overlay on the game screen — shows you what matters right now: your health, your ammo, your next objective. It's designed for a glance. You should never have to stop playing to understand it.
But the game has vastly more information than what's on the HUD. Your full inventory, your quest log, your character stats, your achievement history — all of that exists, all of it is accessible, but it lives behind a deliberate action. You press a button. You open a menu. You go looking for it because you need it.
The critical design principle is that nothing is destroyed. Everything is preserved. But the default view — the thing you see while you're actually playing — only shows what's currently alive.
Stephen Few, the pioneer of dashboard design, formalized this as progressive disclosure: present essential data first, with supporting detail available on demand. The insight isn't about hiding information. It's about respecting the cognitive cost of attention. When everything is presented at the same level of prominence, nothing is prominent. When every document in your wiki looks the same — same font, same placement, same weight in search results — the critical and the trivial compete equally for attention, and attention is finite.
Your camera roll has no glance layer. Every photo has equal visual weight in the scroll. The one-in-a-thousand photo that actually matters is buried in a stream of screenshots, duplicates, and accidental shots of your pocket. You keep scrolling, not because the photo doesn't exist, but because the system has no way to say this one matters more.
Your wiki has the same problem. The compliance policy and the abandoned draft have the same presence in the sidebar, the same weight in search results, the same prominence when an AI agent ingests the knowledge base. There's no signal for "this is permanent and authoritative" versus "this was useful for a week."
What would have to be true
So: the volume is exploding, the signal-to-noise ratio is collapsing, manual curation won't save you because it never has, and the real issue is that every document occupies the glance layer forever. What would a system need to do differently?
Every document needs a concept of durability. Not every piece of knowledge has the same shelf life. An architecture decision record is permanent — it captures a decision and the reasoning behind it, and that remains valuable indefinitely. A quarterly roadmap is seasonal — it's authoritative for a defined period and then it's superseded. A draft design is working — it matters while it's being developed and then either graduates to something permanent or fades away. A meeting summary is ephemeral — it's useful for a week, maybe two, and then it's noise.
These aren't filing categories. They're expectations about the future relevance of information. And right now, no knowledge tool captures them. Every document is treated as equally permanent, equally current, and equally prominent from the moment it's created until someone manually intervenes — which, as we've established, is never.
Critically, durability controls visibility, not existence. A document that fades from the glance layer isn't deleted. It's still in the archive. It's still in the version history. If you need it, you can find it — the same way you can open your game's quest log to review a completed quest, or scroll deep into your camera roll to find that photo from three years ago. But it's not competing for attention in the default view. It's not showing up alongside current documents in search. It's not being ingested by an AI agent as if it's still authoritative.
And durability transitions need to be event-driven, not calendar-driven. This is where it gets interesting. Time-based archiving — "archive everything older than 90 days" — is crude and wrong. A compliance policy doesn't become less relevant after 90 days. A meeting note doesn't become more relevant just because it was written recently.
What actually makes a document stale isn't the passage of time. It's a change in context. A Q1 roadmap becomes outdated when the Q2 roadmap is published. A runbook becomes unreliable when the service it describes is updated. A draft is fulfilled when it's formally approved and promoted. A policy needs review when the regulation it implements changes.
These are events, not dates. And they form dependency chains. A regulation changes at the top → the policy that implements it needs review → the procedures that operationalize the policy need review → the runbooks that execute the procedures need review → the training materials that teach the runbooks need review. One event at the root propagates through the entire knowledge graph.
No knowledge tool today does this. Not because it's technically impossible, but because no tool today models the relationships between documents as first-class objects. Links in a wiki are navigational — they help humans click from one page to another. They don't carry semantic weight. They don't mean "this document depends on that document for its continued validity." But they could.
Finally — and this is the lesson from the camera roll — curation can't be a separate chore. If maintaining knowledge quality requires a distinct, ongoing effort disconnected from the act of creating and using knowledge, it won't happen. The only curation that works at scale is curation that's embedded in the workflow itself.
What if the act of sharing a document also classified it? What if putting a document in a collection automatically set its expected lifespan based on what kind of collection it is? What if publishing a new version of a plan automatically marked the old version as superseded? What if the system noticed — without anyone asking — that a procedure references a policy that's been archived, and flagged it the way a compiler flags a broken import?
That's not curation as maintenance. That's curation as a byproduct of use. The album that builds itself — not because AI magically understands your intent, but because the system's architecture makes curation the path of least resistance rather than an additional chore.
The album your wiki actually needs
Your camera roll doesn't need you to spend a weekend organizing 14,000 photos. It needs a system where the permanent and the ephemeral are distinguished at the point of creation, where documents understand their own lifecycle, and where the act of doing new work automatically curates the old.
The wiki was a good idea. Making organizational knowledge easy to create and share was the right instinct. But ease of creation without lifecycle management is how you get a camera roll with 14,000 photos and no albums — a system where nothing is special anymore because everything is treated the same.
The next generation of knowledge tools won't just make it easy to write things down. They'll give you the glance layer your knowledge base has always been missing — a default view where what matters is what you see, and everything else is a search away, preserved but not competing for your attention.