The Moment I Stopped Thinking About Files
For a long time, I assumed that digital work naturally meant working with files. Writing meant opening Word. Data meant opening Excel. A presentation meant opening PowerPoint. Once the work was finished, it was saved, named, attached, uploaded, revised, and eventually archived.
This arrangement felt ordinary because it was shared by everyone around me. In corporate life, a task often begins with a request for a file and ends with the delivery of another file. We speak of preparing a document, updating a spreadsheet, revising a deck, or converting something into PDF. Even when the real purpose is to develop an idea, examine evidence, or communicate a judgment, the conversation quickly turns toward the container.
My experience with artificial intelligence gradually disturbed this assumption. When I use a conversational AI to write, translate, summarize, compare ideas, or develop an argument, the process feels remarkably direct. The conversation stays close to language and thought. I can introduce an unfinished observation, test several interpretations, return to something discussed earlier, and slowly discover the shape of what I am trying to say. The work proceeds without much concern for pages, margins, fonts, slide layouts, or file compatibility.
The experience changes when I ask the same AI to create or revise a Word document, Excel workbook, PowerPoint presentation, or PDF. The task remains possible, but it often becomes slower and more complicated. The AI must consider styles, page breaks, chart dimensions, formulas, themes, embedded objects, compatibility rules, and countless other details. Much of that labor does not deepen the content. It protects the structure of the container.
This contrast made me wonder whether the difficulty was merely technical. Perhaps AI tools still need time to improve their handling of Microsoft Office and other complex formats. Yet the more I considered it, the more the difficulty seemed to reveal a larger issue. We may have spent decades confusing knowledge with the objects used to carry it.
A Word document can contain an argument, but it is not the argument itself. A spreadsheet can contain data, but the data does not naturally belong to Excel. A presentation can communicate a strategy, but the strategy is not made of slides. These formats serve practical purposes, yet we have become so accustomed to them that they now appear to define the work.
Artificial intelligence exposes the difference between the content and its packaging because it operates most naturally before the packaging decision has been made. It can help us understand, connect, reorganize, and transform knowledge without first deciding where that knowledge must live. Perhaps the significant change introduced by AI is not that documents will become easier to produce. It may be that documents will cease to be the place where thinking begins.
How the Container Became the Knowledge
The dominance of the document did not arise from a mistake. It emerged from the practical needs of earlier technologies. For centuries, books and manuscripts were among the most reliable ways to preserve and transmit thought. Knowledge had to take physical form if it was to travel beyond the memory and voice of its author. Paper did not merely display ideas. It gave them durability, sequence, and reach.
The typewriter strengthened the connection between writing and the page. The writer produced language through a machine that was already preparing a printable surface. Margins, spacing, line length, and page order became part of the act of composition because there was no easy separation between the text and its physical appearance.
Word processors inherited this world. They offered enormous improvements, including revision, duplication, search, and digital storage, but they preserved the basic image of the page. A document remained something that looked as though it might eventually be printed.
Spreadsheets developed from a different tradition, but they produced a similar effect. Data became associated with cells, rows, columns, formulas, and worksheets. Presentation software gave ideas a visual sequence of slides. PDF later offered a stable way to preserve the final appearance of a page across different devices and systems.
Each of these technologies solved a genuine problem. They allowed people to create, distribute, review, and preserve information at a scale that earlier generations could not have imagined. Their success gradually shaped the way organizations understood knowledge itself.
Writing came to belong in Word. Numbers belonged in Excel. Presentations belonged in PowerPoint. Communication belonged in Outlook. Collaboration belonged in shared workspaces. Each application became a territory with its own conventions, permissions, and file formats. Corporate work then became a process of moving content among these territories.
An analyst might export data from an online system into Excel, clean it, create charts, copy those charts into PowerPoint, summarize the findings in Word, send the files through email, and upload the final versions into a shared repository. The same underlying knowledge passed through several containers, each requiring adjustment and maintenance. The work could be valuable, but much of the effort concerned the movement and preservation of artifacts.
This structure also encouraged fragmentation. A decision discussed in a meeting might appear in the meeting notes, a follow-up email, a revised slide, a project tracker, and a final report. Each version might contain part of the truth. Months later, someone attempting to understand the decision would have to reconstruct its history from several disconnected files.
We learned to call this knowledge management, although much of it was actually document management. The file became the apparent unit of knowledge because the systems around us were designed to recognize and govern files. Files could be named, stored, attached, approved, locked, versioned, and archived. Ideas were more difficult to control because they could cross boundaries, change shape, and appear in several places at once.
The container therefore became administratively convenient. Over time, convenience hardened into assumption, and we began asking which application should be opened before asking what we were trying to understand.
When AI Changed the Center of Gravity
Generative AI introduces a different starting point. A person can begin with a question rather than an application. The initial object may be an uncertainty, a set of notes, a contradiction, a dataset, a draft, or an unfinished impression. The form does not have to be settled in advance.
This creates a new order of work. First comes understanding. Then comes development. After that comes transformation. Only later does rendering become necessary.
Consider the ordinary task of analyzing website performance. Under the traditional model, data is exported from an analytics platform into a spreadsheet. Someone cleans the rows, constructs formulas, creates charts, and prepares a monthly workbook. Selected charts may then be copied into a slide deck, accompanied by written commentary. The spreadsheet becomes central even though the data did not originate there and does not need to remain there.
An AI-native process can begin differently. The AI connects directly to the analytics source through an API. It retrieves the required data, applies defined business rules, compares periods, identifies unusual patterns, and produces the appropriate explanation. It can generate an interactive dashboard for analysts, a short summary for executives, a regional comparison for local teams, or a presentation for a meeting.
Excel may still be useful for inspection or special analysis, but it is no longer the necessary passage through which the data must travel. The durable assets are now the source connection, the definitions of the metrics, the transformation logic, the reporting rules, the historical context, and the questions the organization wants to answer. The workbook becomes one possible output rather than the home of the process.
Writing follows the same pattern. An essay does not begin as a Word document. It begins as a perception that has not yet found its structure. The writer may notice a recurring tension, compare several experiences, question an assumption, and slowly discover a central idea.
AI allows that development to occur without forcing the thought immediately into the conventions of a formal document. The writer can ask questions, examine alternatives, reorganize sections, explore examples, and revisit previous discussions. Once the thought has matured, it can be rendered as a web article, a report, a PDF, a speech, a presentation, or a shorter statement for social media.
The content becomes primary, while the format becomes conditional. This inversion helps explain why general AI environments often feel more liberating than AI features placed inside established applications. A tool embedded in Word usually assumes that the Word document is already the center of the activity. It helps summarize, rewrite, format, or expand what exists inside that environment.
A general conversational AI permits the user to postpone the decision about the destination. The work can remain open long enough for the right form to emerge from the content. Applications do not become useless, but their position changes. Instead of asking AI to serve the application, we can allow applications to serve the developing knowledge.
From Documents to Living Context
A more significant change appears when the unit of knowledge is no longer the individual file. For decades, a report was treated as a finished object. It contained a selected set of facts, arguments, charts, and conclusions arranged in a fixed order. Every reader received the same structure, even if their responsibilities and questions were different.
An executive might need three strategic implications. A researcher might need the methodology. A regional team might want local data. A translator might need terminology and source context. A new employee might require definitions that the original audience already understood. The static document attempts to serve everyone by choosing one path in advance.
AI makes another possibility imaginable. Instead of sending a fixed report, an organization could share a bounded body of knowledge containing verified claims, source material, evidence, definitions, relationships, assumptions, uncertainties, permissions, and version history. Each recipient could then interact with that shared knowledge through an AI interface.
One person could request a two-minute summary. Another could ask for a translation. A third could examine the evidence behind a particular conclusion. Someone else could compare the findings with the previous quarter or generate a presentation for a customer meeting. The underlying knowledge would remain shared even though its presentation changed.
I have come to imagine such a package as an AI feed cassette. The word cassette suggests something portable and bounded, yet capable of being played differently depending on the receiving device. The package would not be defined by pages, slides, or spreadsheet tabs. It would carry the content and context necessary for meaningful interaction.
In practice, many people are already moving toward a limited version of this model. Daily news consumption offers a simple example. Instead of visiting a fixed list of websites and processing every article independently, a person can ask AI to assemble a set of developments related to current interests, recent questions, ongoing projects, and longer-term concerns.
The news sites remain important as sources, but they are no longer the only interface. AI places new information within a continuing context. A story about workplace technology may connect with an earlier discussion about document systems. A market development may relate to a professional decision under consideration. A cultural or political event may reopen a question that has appeared in previous reading or conversation. The value comes not merely from selecting relevant articles, but from relating new information to an evolving history of thought.
This form of curation is different from simple personalization. Personalization can easily become a system that shows people more of what they already like. It may reinforce familiar beliefs, reward predictable reactions, and gradually narrow attention.
Contextualization can operate differently. It remembers the questions a person has been developing while also presenting evidence that complicates or challenges those questions. It supports continuity without eliminating surprise. A responsible AI feed should therefore preserve two movements at once. It should deepen the user’s existing lines of inquiry, and it should introduce material that has a credible chance of expanding them.
The safety of such a system would not come only from technical guardrails. It would also depend on the design of intellectual exposure. A useful AI should not merely protect the user from false information. It should protect the user from becoming too comfortable within a perfectly tailored world. The ideal feed is not a mirror, but a conversation partner with memory.
Why the Essay Still Deserves a Shape
If knowledge becomes increasingly fluid, contextual, and interactive, the traditional essay may appear less necessary. Why construct a fixed sequence of paragraphs when readers could simply question the underlying knowledge directly?
The answer lies in the difference between access and articulation. A body of knowledge may contain facts, evidence, relationships, and possible interpretations, but an essay offers a path through them. It records how one person has chosen to see the material at a particular moment.
The sequence matters. The pacing matters. The movement from observation to reflection matters. A carefully shaped essay does more than transfer information. It allows the reader to inhabit an order of attention.
Many essays now emerge from extended interaction with AI. A topic may not begin as a formal assignment. It can develop through conversations, news, professional experience, reading, memory, and questions that return over time. At some point, a pattern becomes visible, and the essay begins when the writer decides that one path through this larger context deserves a stable form.
This process does not weaken authorship. It relocates it. Authorship is no longer defined only by the manual production of every sentence. It also resides in the selection of questions, the recognition of connections, the judgment of relevance, the rejection of weak formulations, and the decision about what deserves to be preserved.
AI can propose several directions. It can help organize sections, clarify language, identify repetition, or produce an initial draft. The final work remains uniquely connected to the life, concerns, preferences, and long-term perspective of the person guiding the process.
Two people could use the same AI system and read the same sources, yet produce entirely different essays because their contexts would not be the same. Each person has a different history of attention. This helps explain why relatively simple, text-based structures may continue to remain useful.
Clear sections allow the perspective to shift without losing continuity. Paragraphs allow one movement of thought to develop before another begins. Such writing can appear as a web article, a printed document, a digital publication, or another format required by a particular audience.
Simple text formats are useful in this process because they keep structure visible without allowing presentation to dominate. A heading remains a heading. A paragraph remains a paragraph. The content can travel among systems while preserving its basic organization.
Their value lies less in any specific technical format than in the discipline they represent. The writing does not depend on elaborate design to hold together. Its coherence must come from the relationships among sentences, paragraphs, and sections. The structure belongs to the thought rather than to the software.
A finished essay can therefore be understood as a stable rendering of a living context. It does not contain everything the writer knows or everything that influenced the reflection. It preserves one stage of understanding. Later experiences may deepen, revise, or even contradict it. The essay remains valuable because it shows what could be seen clearly at that particular point.
In a world of constantly updating feeds and endlessly adaptive interfaces, this kind of stable articulation may become more valuable rather than less. Fluid knowledge allows exploration, while composed writing creates memory.
Becoming Curators of Understanding
Discussions about AI and creativity often begin with the fear that machines will produce the work previously performed by human beings. The concern is understandable, especially when AI can generate competent prose, images, code, summaries, and presentations within seconds.
Yet creativity has never consisted only of producing artifacts. A large part of intellectual creation happens before the artifact appears. It begins in attention. Someone notices a pattern that others have ignored. A familiar event is connected with a distant idea. A contradiction is allowed to remain unresolved long enough to become productive. A question is pursued even when it offers no immediate reward.
AI can assist with these movements, but it does not eliminate the person who decides which movement is worth following. The human role may gradually become more curatorial, although curation should not be understood as a secondary or passive activity. A curator establishes relationships. A curator decides what belongs together, what should remain separate, what deserves emphasis, and what should be questioned.
In an AI-supported intellectual life, curation includes choosing sources, defining questions, testing interpretations, preserving continuity, and maintaining openness to surprise. It also includes knowing when a developing idea has reached the point where it deserves a fixed expression.
The essay is one such expression. A report is another. A presentation, a dashboard, a conversation, or a public statement may serve different purposes. None of these has to become the permanent residence of the knowledge from which it emerged. They are moments of articulation.
This perspective also changes the meaning of originality. Originality does not require producing thought from an isolated self untouched by tools, traditions, or other voices. Human creativity has always been relational. Writers have drawn from books, conversations, memories, teachers, events, and inherited forms. AI adds another participant to that environment.
The uniqueness of the work comes from the context through which these materials are received and transformed. Personal experience, professional responsibilities, intellectual commitments, reading history, cultural background, and repeated questions all shape what a person recognizes as significant. AI may help preserve and connect that context, but it does not live the person’s life.
A future built around AI feeds and interactive knowledge packages may therefore increase the importance of cultivated perspective. When everyone has access to powerful tools, the difference between one person’s work and another’s will depend less on technical production and more on the quality of attention brought to the material.
The central question will not be who can generate the most content. It will be who can recognize which content deserves to become knowledge, and which knowledge deserves to become understanding.
For centuries, people preserved thought by placing it inside manuscripts, books, reports, and files. Those forms allowed knowledge to survive beyond the moment of its creation. They remain valuable, and many will continue to serve necessary legal, artistic, historical, and institutional purposes.
AI introduces another possibility. Knowledge may increasingly live in continuing contexts, connected sources, remembered conversations, and evolving relationships. Reports and essays will still be written, but they may no longer be mistaken for the entire body of understanding. They will become deliberate moments in a longer intellectual life.
Writing within this transition remains traditional in one sense. It still consists of sentences, paragraphs, sections, rhythm, and voice. The patient work of shaping language and discovering how one thought can lead naturally into another remains essential.
At the same time, the essay no longer has to begin from an empty page. It can grow from a cultivated field of context, much of it developed through continuing dialogue with AI. The final text is neither a mechanical product of the system nor an isolated expression of the self. It is the result of curation, judgment, revision, memory, and lived perspective.
The document is no longer the place where the knowledge lives. It is the place where, for a moment, the knowledge becomes visible.
Photo by Daniel Schludi on Unsplash
Tom’s Blog 