Prim

The primitive unit of knowledge and memory for AI

Prim

Prim

Prims are files for AI + Human collaboration.

01

  • Excel
  • Word
  • PowerPoint
  • PDF
Prim

Please join us for three minutes.
As we journey into Prims, a brand new file system necessary for the AI age.

To understand why AI and humans need Prims, not PDFs, we have to understand why the printing press needed movable type, not feathers.

02

Before the printing press

Scribes used a feather to copy every book by hand. A Bible took about a year.

A monk at a sloped desk, copying a manuscript by hand.
Medieval illumination of a scriptorium — scribes at desks.

03

After the printing press

Mainz, 1455. A press could create in a day what a monk copied in months. Not a new kind of book. A different rate of making them.

A fifteenth-century printing workshop.

The press didn’t use a feather.

04

In the span of 20 years, book output increased 100x per worker.

By 1500, fifty years of print, Europe had more than nine million books.

One book.
A stack of about a hundred books.

05

They did not scale the feather to a hundred feathers. The instrument itself changed to a press.

A feather was for human hands to write every letter. A press was for human hands to press a page.

A table covered in quills.
A single printing press.

How will AI change how humans interact with data?

Knowledge work is a factory. Raw material in, finished idea out. Since Turing described a machine that could compute anything, a person sat with it and did that conversion. The tools changed. The factory did not.

1950s operator feeding punch cards into a room-sized computer.

1940s–50s

Punch cards

You punched every hole. Overnight, the machine tabulated. You read the listing.

A person at a keyboard, waiting on a printout.

1960s–70s

Command line

You typed every line. A cursor, a command, a wait. One instruction at a time.

1980s office worker at a CRT spreadsheet.

1979–

Mouse and sheet

VisiCalc put a ledger on the glass. Then Word. Then slides. You clicked every cell. The file was the work.

Two people talking. The work is spoken, not typed into a format.

Now

Chat

You ask. It drafts the whole thing.

Every decade, humans improved the tools they used to refine raw data into knowledge.

Now, AI refines raw data into knowledge with near autonomy.

Pre-AI, most files were not the finished product.

They were part of the factory floor.

A spreadsheet became an insight.

A person sat in Excel until a number meant something.

A deck became a board pack.

The slides were how the meeting ran.

A week of both became a firm.

Deloitte. IBM Consulting. People converting raw data into advice, one file at a time.

Knowledge work is becoming idea work and validation work.

AI converts the raw material. People choose the idea. People check the result.

People show up because they want their intent to change the world.

In a positive way. Or to help a customer succeed with an outside perspective. That is the job.

Starting in 2027, we work with our eyes, ears, mouths, and hand signs more than ever.

The interface need not be a computer with a mouse and a keyboard any longer.

The new floor is a table.

Record. Debate. A couple of microphones, and a screen of what you’re all discussing. The files come out the other side.

A table, microphones, a screen of the work.
Eyes. Ears. Mouths. Hand signs.

06

PDFs are the feather. Prims are the press.

07

Those feathers went into a cabinet. SharePoint. Drive. Then we started the next file.

1980s office worker at a CRT, trapped in a spreadsheet.
Spreadsheet
1980s office worker typing a report.
Document
1980s conference room, overhead projector.
Deck

Their work was the file. Then the file went into the cabinet.

08

Their grandchildren sit at the table.

Their grandchildren around a table, talking — the work is the meeting.
They speak. That becomes the next meeting.
Two of them mid-conversation.
  1. Meeting
  2. Voice
  3. Intent
  4. Output
  5. Next meeting

The press didn’t give the monk’s children more feathers. It gave them books. Prim is that, for their grandchildren: a meeting, spoken; intents; outputs that shape the next time they sit down. The instrument changed.

Pre-AI

A human makes every file.

One person. Three formats. Three jobs.

After

Humans and AIs contribute. Files come out.

Same files. The arrow flipped.

1980s office worker at a CRT, trapped in a spreadsheet.
A format for a person at a machine.
Two people working together with a Prim on their screens.
Humans and AIs contribute to the same file.

Those formats were not designed for agents. They were designed before agents existed.

We made each file to represent a single concept. Prim reverses the order: store the concept, generate the files. Same Excel. Same deck. Different century of format.

A factory exists so a person isn’t the machine. Prim is that, for knowledge.

Enterprises love Prims because the work lasts, moves, and belongs to the company.

Time capsules

The work survives the person, the session, and the vendor. Open it in ten years. It’s still the Prim.

Shareable

One file. Desk to desk, agent to agent, counsel to the board. You send the prim — not a zip of conflicting sheets.

Company property

The knowledge is the firm’s. Not trapped in someone’s laptop, a chat log, or a tool that walks out the door.

AIs love Prims because they can finally hold the work.

They can read it

No scraping a deck. No guessing which tab is the source of truth. The file is native — agents reason over it without a translation layer.

They can trust it

Claims carry evidence, hashes, and provenance. Validation is fail-closed. An agent doesn’t have to hope the spreadsheet is right.

They can continue it

A session is a Prim. The next agent picks up the same file. Work doesn’t die when the window closes.

Read the spec Pre-AI

Pre-AI

  1. 80s

    1982–89

    Lotus 1-2-3, WordPerfect, floppy disks

    One program. One file. One disk.

  2. 90s

    1990–99

    Excel, Word, PowerPoint

    The suite becomes the job.

  3. 00s

    2000–09

    PDF, email the attachment

    The file leaves the building.

  4. 10s

    2010–19

    Drive, Dropbox, Slack

    Cloud copies of the same files.

  5. 20s

    2020–

    Copilot, ChatGPT

    Agents arrive. Still opening those files.

We made each file to represent a single concept.

Spreadsheets for structure. Documents for decisions. Decks for the story. Notes for memory. Each file was a projection we had to author by hand — on whatever machine held that format.

1980s office worker at a beige PC, green spreadsheet on a CRT.
Spreadsheetstructure
1980s office worker typing a report on an early word processor.
Documentdecisions
1980s conference room watching an overhead projector.
Slidesthe story
1980s desk with a legal pad, floppy disks, and a CRT.
Notesmemory

Now

Store the concept. Generate the files.

Reverse the order. The Prim holds the idea once. Excel, Word, the deck, the notes — those come out of it. Same four files. The arrow flipped.

One file

The Prim is the concept.

  • Spreadsheet — generated
  • Document — generated
  • Deck — generated
  • Notes — generated
Then: files point into a reconstructed idea. Now: a Prim generates those same files.
Same four files. The arrow flipped.

Portable

Agent to agent. Agent to human. Human to agent.

Like PDF, for AI work

Adobe cracked portable documents. A Prim is a portable package of knowledge — the same file moves between people and agents without translation.

A session is a Prim

An AI session itself is a type of Prim. What people call the model “dreaming” is the raw work. The Prim is what it packages.

Schema first, then autonomy

Define great schemas for Prims up front and agents can finish the work. The spreadsheet is no longer the job — it is an export.

What a Prim is

A self-contained pack of knowledge — not an app, not a document, not a database dump.

AI-native

Agents can read, validate, reason over, and act on them without translation layers.

Memory

Durable, versioned, supersedable knowledge that persists across sessions and agents.

Evidence-first

Claims carry provenance, trust tiers, and hashes.

Generative interfaces

No fixed UX. Views are rendered on demand — json-render style.

Human-readable

Still openable and understandable by people.

Raw data, then package

A Prim is not the dream. It is what the dream becomes when it is packaged so the next agent — or person — can pick it up.

Prim as the source of truth, with arrows out to sheet, deck, interface, and agent views.
The Prim stays put. Sheets, decks, UIs, and agent reads are projections.

How to say it

If it needs a long explanation, it has not landed yet.

  1. 01Send me the prim.
  2. 02Don’t send the spreadsheet — just send the prim.
  3. 03An AI session itself is a type of Prim.
  4. 04Store the concept. Generate the files.

Profile names belong in specs. Everyday speech stays on Prim.

Packaging

A Prim is a pack. Sending one should be one file.

Directory packCanonical source of truth — git, editing, validation.
.prim.zipPrimary interchange. Attach this when you say send me the prim.
.prim.tar.gzAllowed. Unix and agent workflows.
.primReserved branded container. Zip under the hood.
Contents of a .prim.zip: index.md, log.md, structure.json, evidence, and docs.json.
A .prim.zip is a zip whose root is the pack. Attach that file.

The family

Prim is the format. Profiles are the domains.

Under the hood, Prims are additive profiles from Eidos.

Tree from Prim to additive profiles EMF, ORF, OPF, ODFW, OPFF, OMF, and OCSF.
New domains arrive as profiles. Existing Prims keep validating.
EMFHuman intent + durable memory ORFResearch / investigation OPFProduct graph ODFWSpreadsheet → bronze proof OPFFPersonal finance packs OMFMeeting occurrences OCSFCorporate structure & capital

Don’t send the spreadsheet.
Send the prim.

The category specification, family map, and packaging rules live in the open. MIT. Early.

github.com/eidos-agi/prim

Sources

Every number and quote on this page has a citation.

  1. 1Turing / Pilot ACE, 1950

    Science Museum Group. Turing designed ACE in 1946; the smaller Pilot ACE ran in 1950, taking punch cards and paper tape.

    scienceandindustrymuseum.org.uk

  2. 2Drucker, 1959

    Peter F. Drucker, Landmarks of Tomorrow. Coined “knowledge worker”: people who use information as raw material.

    en.wikipedia.org

  3. 3VisiCalc, 1979

    VisiCalc, released 17 October 1979 for the Apple II. First spreadsheet; widely called the first personal-computer killer app.

    en.wikipedia.org

  4. 4Engelbart, 9 Dec 1968

    Douglas Engelbart, “The Mother of All Demos,” SRI. Public debut of the mouse, hypertext, and shared editing.

    invention.si.edu

  5. 5McKinsey Global Institute, 2012

    The Social Economy. Interaction workers spend an estimated 28% of the week managing email and nearly 20% looking for internal information or colleagues.

    mckinsey.com

  6. 6Microsoft Work Trend Index, 17 June 2025

    Breaking Down the Infinite Workday. Microsoft 365 telemetry: PowerPoint edits spike 122% in the last 10 minutes before a meeting; average worker receives 117 emails a day; interruptions every two minutes in core hours.

    microsoft.com

  7. 10Britannica, early printing

    Encyclopaedia Britannica, History of publishing. Before print, European manuscript books numbered in the thousands. By 1500, after fifty years of printing, there were more than 9,000,000 books.

    britannica.com

  8. 11Trithemius, 1492

    Johannes Trithemius, De laude scriptorum (In Praise of Scribes). Argued printed books would never equal handwritten codices, and that abandoning copying was laziness.

    historyofinformation.com

  9. 12Gessner, 1545

    Conrad Gessner, Bibliotheca Universalis. Later accounts quote him warning princes of a “confusing and harmful abundance of books.”

    en.wikipedia.org

  10. 13Gutenberg Bible, c. 1455

    The 42-line Bible printed at Mainz, commonly dated 1454–55. Conventional marker for European movable-type printing at scale.

    britannica.com

  11. 14Scribe vs press output

    A skilled scribe is commonly estimated at on the order of one Bible per year. A press could pull thousands of sheets in a day. Elizabeth L. Eisenstein, The Printing Press as an Agent of Change (1979).

    en.wikipedia.org

  12. 15Plato, Phaedrus, c. 370 BCE

    Socrates, via the myth of Theuth and Thamus: writing “will produce forgetfulness in the minds of those who learn to use it, because they will not practice their memory.”

    historyofinformation.com

  13. 16Telephone scare, late 19th c.

    Period newspapers called the telephone a danger because it “enters into every dwelling,” and “never a necessity.” Collected in Wondermark, “True Stuff: Alexander Graham Bell vs. Western Union.”

    wondermark.com

  14. 17Wiener, 1950

    Norbert Wiener, The Human Use of Human Beings. The automatic machine is “the precise equivalent of slave labor,” and competing with it would make the Depression “a pleasant joke.”

    en.wikipedia.org

  15. 18TIME, 2 April 1965

    TIME on computers: some hoped they would return man to “the Hellenic concept of leisure… The slaves, in modern Hellenism, would be the computers.”

    time.com

  16. 19Pew Research Center, June 2026

    52% of Americans say they are more concerned than excited about the increased use of AI in daily life, up from 37% in 2021.

    pewresearch.org