The big idea: Snapchat for Enterprise
Cheap storage has made us default to saving every small document we create. What if corporate memory had a price?
I’ve always loved Internal Tech Emails, an aggregator of the most interesting text messages, emails, and DMs from tech leaders. These messages shed new light on the stories we only get glimpses of externally, like Facebook’s decision to purchase Instagram.
Recent posts got me thinking about data retention. An abundance of cheap storage has trained the corporate world to live on a retain-by-default basis, where anything created is stored for future evaluation, lawsuits, or eventual publishing (via legal proceedings or otherwise).
But that ethos might be increasingly obsolete. Every day, people generate huge amounts of emails, chats, and documents that never get read. AI is turbocharging this deluge of content, dumping unread AI notes for every meeting and generating endless LLM-written documents with no clear origin or importance.
This isn’t just a waste of storage. Too much content makes it hard to find what’s actually important; simple searches get lost in thousands of low-quality or out-of-date documents. When old documents get dug up and reprocessed, mistakes get made — the horror of accidentally citing from decisions_vFinal.docx and not the much more accurate decisions_vFinalFINAL.docx.
The consumer world doesn’t work like this. Apps like Snapchat have created huge audiences by embracing ephemerality and deleting everything by default. What would happen if the corporate world did the same?
A case for ephemeral data retention
Historically, effort was a brake on content creation. This made it a decent signal; the fact somebody thought it was worth spending a week on a memo indicated that they at least thought the content was worth reading. It wasn’t perfect, but it meant that there was some reason to think that the end output was important.
That signal went away when the cost to create a 40-page deck became $0.30 in tokens; the output no longer carries any indication that someone thought that it mattered. Without that friction, the burden shifts to the reader, who now has to determine whether the output mattered at all — a load that chips away at anybody’s ability to get things done.
This dynamic only persists because storage is treated as unlimited. In the old days, memos without value would eventually be shredded to make room for more physical paper in filing cabinets. While not the intention, this acted as a cleanup process: an automatic validation of whether something is important enough to keep.
In a world where creation stops being a signal, maybe metaphorically shredding everything will prompt better decision-making. But people won’t just start shredding their own documents — the structure of how work gets done needs to change.
There’s one unlikely hero to bring this to life: Snapchat. It’s time for Snap to follow every other tech company and commit to the enterprise pivot.
Introducing: Snap for Enterprise
Snap for Enterprise is a simple concept. The software plugs into the API for every part of a company — every message, DM, email, meeting note, pull request, and memo — and deletes it after it’s read by the person it’s shared with.
Sure, there might be some flexibility — a calendar invite persists until after the meeting is finished, and in-progress documents should probably persist (although perhaps they should start degrading if you don’t work on them regularly). But fundamentally, information goes from permanent to disposable. Only proactively retained content is kept.
What gets retained is governed through Snap Memories: a selection of documents and artifacts that escape deletion. But for retention to be a strong signal, you need companies to be selective in what gets kept.
That’s why Snap Memories charges for each marginal document stored, with prices increasing for older documents over time. Most institutional memory is a depreciating asset; it’s worth a lot for 30 days and almost nothing after a few months. Charging for duration of retention creates an amortization schedule for information, matching the cost of retention to the actual decay curve of value.
Companies already ration attention, roadmap space, headcount, compute, and budget. By putting a price on every piece of retained content, it forces real tradeoffs between what gets saved. Someone, somewhere will have to explain why it’s worth $12 a month to store the meeting notes for an internal check-in.
This might have a few legal obstacles
There is a small problem with building software whose main value proposition is destroying potential evidence.
Retention policies can be legally dicey, and companies are often not allowed to forget (as much as they might like to). Some data needs to be deleted immediately under GDPR; some data needs to be held for years under SOX. The SEC has spent the last few years making it very clear that “the important conversations happened somewhere else” is not an ideal recordkeeping strategy.
Worst of all is litigation. Once legal action is reasonably expected, the company has an obligation to preserve evidence. I don’t know whether it’s illegal to automatically delete everything, but at best you’re annoying people whose job is to be professionally upset about missing records.
Luckily, there’s a pretty simple solution: copy deleted items to long-term storage, only accessible with a subpoena or as part of legal proceedings. The document still exists in a metaphorical legal basement, available for audits, subpoenas, and the kind of discovery process that eventually gets uploaded to Internal Tech Emails.
But for the normal employee, it is gone. It does not show up in search. It does not feed the chatbot. It cannot be casually resurrected by someone who forgot to read the Q3 planning doc.
To maintain the integrity of the system, access has to be genuinely annoying. You can’t have employees DMing Legal to recover an old roadmap file because they didn’t think it was important enough to keep. If the basement becomes just another archive with worse UX, the whole system collapses back into retain-by-default.
The corporate politics of memory
Dynamic pricing for retention raises the question: who gets to decide what gets kept? Somebody has to figure out how to quantify the value of what’s getting produced, compare it against everything else, and either retain it or not. This sounds administrative, but it’s actually political. Corporate memory shapes who gets credit, who gets blame, and which decisions get made.
Without governance, this becomes a budget knife fight — do we need to keep data on everybody’s in-office days if it means trading off against the memory budget to track who’s interviewing for an open role? If old corporate politics was fighting over resources to create something, new corporate politics becomes a fight to not delete the quarterly planning system.
New KPIs will emerge: share of content retained is a pretty sharp signal for how valuable outputs are. Some kind of Chief Memory Officer will need to be established, with a mandate to decide what the company is allowed to remember. This requires a monopoly on memory, with strict enforcement to stop employees from secretly keeping context captured via screenshots.
But the corporate changes go beyond administrative into anthropological. There’s some invisible scaffolding that exists in any information environment — how you got to a decision can be as important as the decision itself. Written cultures can refer back to any document at any time, even as context changes. Oral cultures, by contrast, are homeostatic — they exist in the present, and drop ideas or memories that no longer have relevance for the group.
Snap for Enterprise, through a change in incentives, turns companies homeostatic. Oral cultures need somebody — like a bard or a village elder — to act as a source and transmitter of knowledge, sharing wisdom on why things are the way they are.
In a Snap for Enterprise business, this naturally becomes middle management. Much like a village elder around a fire, the team lead’s job becomes telling the story of why Q2’s priorities shifted and the characters that made it happen.
This structure does more than preserve the story. It naturally invites ritual and culture into the workplace, turning the daily standup into a daily storytime that retains memory and bonds the group together.
The necessary skills go beyond simple storytelling: nobody is going to listen to John talk about how a long-since-lost Excel model said this would make money. Much like the ancient Greeks used dactylic hexameter to turn the Iliad into a catchy and enthralling beat, the great journey to the fiscal year product roadmap will learn to embody the rhythm and energy of oral cultures.
And then there’s the AI pivot
Obviously, once you turn middle management into the keepers of corporate oral tradition, someone will immediately try to automate them. A role that involves absorbing information, remembering context, and explaining it in plain English is almost offensively attractive to an AI startup.
The pitch writes itself: train a model on all of the content, before deleting it forever. Fewer managers, better memory, cheaper storytelling. This creates a new type of corporate haunting: no documents, but you have to run everything through a chatbot that remembers the vibes.
Of course, given that all of the training data is deleted it’s going to be hard to tell whether the outputs are any good. Maybe the future of Internal Tech Emails looks less like leaked communications and more like LLM weights that can tell you the story of why we hired Bill in the tone of a pirate.
Official idea rating
2/5. This is radical enough that only very specific types of companies, like a quirky quant shop, would be able to sustain it. Any business-to-business interaction will be tough if you’ve turned your middle managers into corporate bards. And even if you did try…can two companies doing this actually do business together? What if somebody forgets they had a contract?
That said, pricing data retention might be a 4/5. Every stale memo, abandoned plan, and orphaned meeting note pollutes the information environment. Companies are going to need some way to decide what deserves to survive. And if that means making middle managers recite the product roadmap around the metaphorical campfire, there are worse ways to run a meeting.
Solid post. I've often argued that outdated documentation is worse than none, because (much like Gell-Mann Amnesia) you will falsely believe outdated documentation. More than once, I proposed a plugin for Self-Immolating Confluence. Old doc? Nobody has updated it in 3 years? It's worse than not having it. Burn it down.
With LLMs (like Rovo) it's worse, because the training has no way to weight the validity of anything other than potentially taking in age of the last update and weighing that part of the documentation down in the model. But then as soon as you start turning on document destruction, people will export it to zip files and keep it around outside of the controls track.
I like the alignment tactic of making old documents cost more, in actual dollars. We do not have an accurate (mental) model of how much outdated documentation drags down engineering orgs, for example. Anybody who works in the field knows it's true, but you can't make a metric out of it so the MBA-types pretend it doesn't exist. It would force you to shift from "document and pray" to "build systems that don't require documentation to understand."
Example: a lot of companies use "run books" as a crutch for not investing in proper *systems*; software, automation, education, etc. If you make your runbooks ephemeral.... you will naturally do better systems thinking.
The deletion/archiving time and triggers would need to be intelligent and automatic. The document systems would have a built-in context analyzer that makes educated guesses about when and where a document might be needed again. If it's a person's/team's responsibility, it won't go anywhere, but if it's automatic then it'll just happen in the background and continually improve its judgement of what to save and what to trash.