A Card Stops, and the Tools Become Visible

My credit card was temporarily restricted after the issuer identified a transaction that required verification. It was a routine security precaution, the kind of preventive action designed to protect cardholders when something appears unusual. The matter was resolved quickly, and the card soon returned to normal. No serious financial problem had occurred. Yet the brief interruption produced an unexpected benefit.

While the restriction was in place, several automatic subscription payments failed. What first appeared to be an inconvenience became an unplanned audit of my digital life. I had to look at the affected services one by one and decide whether each still deserved a place in my working environment.

The surprising discovery was not that some subscriptions were expensive. It was that several of them had become unnecessary without my noticing. I had once used them regularly and regarded them as essential tools. Yet when the payments stopped, my work did not. I could still organize ideas, translate documents, research unfamiliar subjects, edit files, and create finished materials. Those activities had moved elsewhere, mostly into my conversations with AI.

The change had not arrived as a formal decision. I had not compared every application with an AI service and announced that the old tools had lost. I had followed convenience. When I needed to reorganize a rough set of notes, I asked AI. When I needed a PDF translated or reformatted, I asked AI. When I wanted to compare sources, draft an essay, improve an image, or extract a structure from scattered thoughts, I increasingly began in the same place. Separate applications remained installed, and their subscriptions continued, but they had ceased to be the natural entrance to the work.

The blocked card revealed the gap between payment and attention. A subscription can survive long after the habit that justified it has disappeared. Automatic renewal makes this easy because it converts an old decision into a continuing expense. We keep paying for the possibility of using a capability, even after another tool has absorbed it.

This small incident gave the familiar claim that “AI will replace software” a more concrete meaning. The shift was no longer a distant prediction about the technology industry. It had already occurred in the ordinary sequence of my own work. The more useful question was not whether every application would vanish. It was what an application had truly been providing, and whether that value still required a separate interface, subscription, and container.

The Age of Rented Capabilities

The rise of software as a service changed the relationship between users and tools. Instead of buying a program and keeping it for years, people began renting access to a growing collection of capabilities. One subscription organized notes. Another handled documents. Another stored books and articles. Others edited images, converted PDFs, managed references, scheduled work, or improved writing.

These services offered more than technical functions. They offered a designed way of thinking and working. WorkFlowy turned thought into expandable lists. Notion joined pages, databases, calendars, and boards within a flexible workspace. Craft made digital writing feel polished and visually coherent. Obsidian transformed folders of Markdown files into a network of linked ideas. Each application proposed its own answer to a human difficulty: how to give usable form to information that would otherwise remain scattered.

The interface was part of the value. Users learned where commands were located, how items related to one another, and how information should be shaped to fit the system. A good interface reduced friction, but it also trained the user to think according to the application’s structure. The outline encouraged hierarchical thought. The database encouraged properties and views. The graph encouraged links. Over time, the distinction between organizing knowledge and operating the software became hard to see.

Cloud storage strengthened this relationship. Once years of notes, templates, links, comments, and routines accumulated inside an application, leaving became costly. The difficulty was not limited to exporting text. The meaning of the information often depended on relationships created by the service. A list could be downloaded, yet its filters, views, permissions, and history might not survive. Convenience at the beginning became dependence later.

This model worked because specialized software often performed its task better than a general tool. A dedicated PDF application knew more about PDFs. A presentation program offered precise control over slides. A research service combined search with citation management. Paying for several subscriptions made sense when each provided a capability that could not be obtained elsewhere with comparable quality.

AI changes that calculation because it does not present itself as one more specialized tool. It offers a broad field of capabilities through a common language. The same conversational space can help with translation, research, structure, formatting, image creation, spreadsheet analysis, and document production. The user no longer has to rent a different interface for every class of task.

The economic attraction is easy to understand. Traditional SaaS sells permanent access to a narrow competence, whether the customer uses it every day or twice a year. General AI offers competence on demand. A person who occasionally needs to combine PDFs, create a diagram, translate a report, or reorganize notes may no longer see a reason to maintain four separate subscriptions. The AI service does not have to surpass every specialist application in every respect. It only has to be good enough for the actual work the user performs.

That final qualification matters. Many subscriptions are not displaced because AI becomes objectively superior to the application. They are displaced because users discover that they never needed the application’s full precision in the first place.

Intelligence Moves to the Front

Conventional applications require users to translate intentions into operations. A person who wants to prepare a report must decide which program to open, locate the relevant files, understand the available functions, and perform a sequence of commands. The application does not need to understand why the report is being made. It waits for the user to operate it correctly.

AI reverses this relationship. The user begins with an intention expressed in ordinary language: compare these documents, identify the main differences, prepare a table, and turn the result into a presentation for a management audience. The system interprets the request, selects methods, uses tools, and assembles an output. Human attention moves away from commands and toward objectives.

In that arrangement, AI becomes both interface and worker. It is the interface because the user communicates through language rather than menus. It is the worker because it performs a chain of operations that previously required direct human action. The distinction between asking and doing begins to narrow.

Adobe offers a useful example. Creative Cloud contains powerful applications developed for professional control, yet many subscribers use only a small portion of that power. If someone needs to extract pages, edit a photograph, remove a background, translate a PDF, or produce a formatted document, conversational AI can now perform much of the work without requiring mastery of Acrobat, Photoshop, or InDesign. The finished PDF or image still exists, but the path to it no longer begins with a specialized application.

This does not make professional design knowledge worthless. Exact typography, color management, accessibility, non-destructive editing, video timing, print production, and brand consistency still require disciplined judgment. What loses value first is procedural knowledge: which menu contains a command, which sequence of filters creates an effect, or which export setting converts one format into another. The skill shifts from operating the tool to directing and evaluating the result.

Microsoft faces a related change. For decades, Word, Excel, and PowerPoint have been the visible rooms in which office work takes place. Adding an AI assistant to each room is helpful, but it may represent only an intermediate stage. If an agent can analyze a workbook, draft a report, and build a presentation from the same request, users may no longer need to move consciously among the applications. The Office programs continue to process and render the artifacts, but they become part of the background machinery.

Search-oriented AI services face pressure from the same movement. Perplexity gained attention when the ability to search the web, synthesize results, and provide sources distinguished it from models that relied mainly on training data. Once leading AI services gained search, research, citation, file analysis, and agentic action, that distinction weakened. A feature that once defined a product became an expected part of general intelligence.

Notion, Craft, and Obsidian will not face identical outcomes, but each must answer a new question. If AI can organize fragments, create structures, connect related ideas, and retrieve them through conversation, how much value remains in the manual arrangement of a knowledge workspace? Their databases, files, and collaboration systems may remain useful. Their status as the place where thinking begins becomes less secure.

The application does not have to disappear from the computer to vanish from experience. It can remain technically present while losing its position as the place where intention enters the system.

What Must Remain Behind the Interface

Once AI becomes the main entrance to digital work, the location and form of data become more important, not less. An intelligent agent can work across many tasks only when it can reach the relevant material. Information locked in a proprietary format or inaccessible service limits both the user and the AI.

This creates renewed interest in portable formats. Markdown is valuable because it is readable as text, editable by many tools, and easy for AI to interpret. A folder of Markdown files does not depend on one company’s interface for its continued existence. Obsidian benefits from this arrangement because its notes remain ordinary local files even if the application is removed. The interface may become optional while the knowledge survives.

Portability, however, cannot be reduced to the ability to export visible text. A modern workspace contains relationships as well as content. Comments reveal how a decision was reached. Permissions define who may act. Version histories show what changed. Database properties connect records. Approval states distinguish drafts from official documents. A CSV or folder of text files may preserve the words while losing the structure that gave them organizational meaning.

The stronger principle is therefore not “everything should be plain text.” It is that users and organizations should be able to retrieve their content, structure, metadata, and history in forms that other systems can understand. Open formats matter, but so do documented APIs, reliable exports, stable identifiers, and the ability to move information without destroying its relationships.

Managed cloud systems also retain a necessary role. Companies need secure storage, identity controls, audit records, retention policies, backup, and formal approval. AI can create a document, but creation alone does not establish which copy is authoritative. It does not determine who has the right to view it, who approved it, or how long it must be preserved.

Seen from this angle, Microsoft’s durable strength may lie beyond Word, Excel, and PowerPoint. OneDrive and SharePoint store organizational documents. Entra ID manages identity and access. Microsoft Graph connects information across services. Purview supports governance and compliance. Azure provides infrastructure. Even if employees interact mainly with an AI agent, these systems can remain the controlled environment in which the agent works.

Google has a parallel advantage through Gmail, Drive, Docs, Sheets, identity, search, and its wider account ecosystem. The competition between Microsoft and Google may not be decided by which company has the more attractive word processor. It may depend on which environment gives AI the richest authorized context while preserving enterprise control.

A new architecture begins to appear. AI occupies the foreground as the conversational layer and active worker. Portable files protect personal and institutional ownership. Cloud platforms maintain identity, permissions, history, and continuity. The visible application becomes thinner, while the infrastructure behind it becomes more consequential.

From Task Processor to Outcome Owner

The same change that moves applications into the background also changes the organization of human work. AI is strongest when a task has a recognizable input, a defined output, available reference material, and criteria by which the result can be checked. Many office jobs contain a large number of such tasks.

Consider a traditional translation team. A manager receives requests, assigns documents to translators, monitors deadlines, and collects completed files. Each translator may work within a narrow boundary. The source text arrives, the translation is produced, and the finished document is submitted. Understanding the wider publication strategy may be helpful, but it is not always required for completing the assignment.

This is the kind of arrangement that AI can compress. A model can receive the source text together with terminology databases, style guides, previous translations, product documentation, and audience instructions. It can generate a first version rapidly and revise it in response to feedback. It can also compare the translation against the source, flag inconsistent terminology, and adapt the output to a required format.

The immediate temptation is to imagine that ten translators will be replaced by one manager who performs quality assurance. That may occur in some settings, especially where the work is repetitive and the consequences of error are limited. Yet reduced cost can also increase demand. Articles that were never translated because of budget constraints may now be localized. Existing teams may handle far more material rather than shrink in direct proportion to the productivity gain.

Either way, the task-only role becomes harder to defend. A professional who receives one item, performs one operation, and submits it without responsibility for what happens before or after is competing with the form of work AI handles best.

The surviving human role is broader. Someone must decide which material deserves translation, identify the intended Japanese audience, resolve technical ambiguity, judge cultural and legal implications, coordinate with global and local stakeholders, and determine whether the final publication serves its purpose. That person is no longer defined by the act of translating sentences. The role is closer to a language-market publication owner.

This helps explain the claim that everyone will become a manager. It does not mean every employee will supervise other people. It means more workers will have to manage objectives, context, agents, standards, exceptions, and consequences. They will be responsible for the movement of work rather than one isolated operation inside it.

Traditional managers are not protected by their titles. A manager who distributes tasks, requests status updates, and compiles routine reports may also be performing highly automatable work. The durable managerial function is not task allocation. It is the capacity to define an outcome, reconcile competing interests, recognize exceptions, set a defensible standard, and accept responsibility for the result.

AI can read more documents than any individual employee and maintain awareness across many parallel activities. Yet access to context is not the same as living within a situation. Organizational life includes informal commitments, unresolved tensions, personal trust, shifting priorities, and decisions that have never been written down. AI may identify patterns across the record, but it does not automatically know which relationship must be protected or which exception is justified.

A large context window does not create accountability. The human role moves upward from execution, but it does not disappear.

Lock-In Finds a New Home

The movement away from specialized SaaS can feel like liberation. Fewer subscriptions mean fewer interfaces to learn, fewer databases to maintain, and fewer recurring payments for services that are rarely used. Portable files appear to restore control to the user.

Yet AI can create a new form of dependence that is harder to see. The most valuable asset may no longer be a document stored in a proprietary application. It may be the accumulated context held by an AI service: years of conversations, preferred writing patterns, recurring instructions, connected accounts, agent configurations, and knowledge of ongoing projects.

A user may export every Markdown, Word, and PDF file and still lose much of the working relationship developed with the system. The files contain outputs, but not always the reasoning history, corrections, preferences, or routines that made future work easier. Lock-in moves from the file format to the context surrounding the file.

Organizations face the same danger at a larger scale. An enterprise agent may learn how reports are prepared, which sources are trusted, how approvals are routed, and how exceptions are handled. If those arrangements exist only inside one provider’s memory, embeddings, or workflow configuration, changing providers becomes difficult even when the underlying documents are portable.

Concentration also increases operational risk. Ten narrow applications can fail independently. One AI service that performs ten categories of work creates a common point of failure. An outage, policy restriction, pricing change, model update, or security incident can interrupt many activities at once. The simplicity visible to the user may rest on a highly concentrated dependency.

Reliability presents another challenge. Conventional software is designed to perform defined operations consistently. AI interprets requests probabilistically and may produce different results as models, prompts, or context change. Variability supports creative exploration, but it becomes dangerous when the work involves financial calculations, legal obligations, technical instructions, or regulated records.

Well-formed output can make the risk harder to detect. A translated document may read fluently while altering the force of a requirement. A table may look complete while placing a number in the wrong category. A PDF may preserve its appearance while losing an annotation, accessibility tag, or hidden field. As execution becomes easier, verification becomes more demanding.

Security risks also expand when AI moves from advising to acting. A conversational assistant that drafts text has limited power. An agent with access to email, storage, calendars, databases, and publishing systems can make consequential mistakes. It may expose confidential material, modify the wrong record, send an unfinished message, or follow an instruction that conflicts with a higher authority.

These risks do not justify returning to a separate application for every task. They require a more disciplined division of roles. AI should perform and coordinate work. Portable formats should preserve ownership where possible. Managed systems should protect official data and record authorized actions. Human beings should remain responsible for goals, exceptions, and consequences.

The Application Recedes

The subscriptions exposed by the blocked card did not all represent the same kind of value. Some were little more than interfaces for occasional tasks. Others provided access to content, preserved records, or supported collaboration. Once that distinction became visible, the future of software looked less like a universal extinction and more like a sorting process.

Applications whose main value lies in formatting, organizing, converting, summarizing, or presenting information are the most exposed. Their capabilities can be absorbed into general AI, especially for personal work and low-risk business tasks. A separate subscription becomes difficult to justify when the same result can be requested from the AI service already used throughout the day.

Systems that preserve authoritative data, execute transactions, manage access, coordinate teams, or satisfy regulatory obligations occupy a stronger position. They may lose their visible interfaces, but their underlying functions remain necessary. The user may never open the application, while an agent continues to rely on its database, permissions, and audit history.

Software companies will therefore have to decide what they truly own. A polished interface is no longer a secure advantage. Durable value will come from trusted data, execution rights, collaboration networks, professional precision, governance, or infrastructure that AI cannot reproduce through language alone. Companies that possess none of these may discover that they were selling a temporary arrangement of buttons.

Users face a related decision. The sensible goal is not to eliminate every application or place every document inside an AI conversation. It is to remove tools that duplicate capabilities now available elsewhere, keep data in portable forms when possible, and retain governed systems where continuity and accountability require them.

The human position changes with the software. When applications demanded direct operation, competence was often measured by procedural fluency. People proved their value by knowing how to use the tool. As AI assumes more of that work, competence moves toward framing problems, selecting evidence, setting standards, judging results, and understanding consequences.

The blocked card did not cause this transformation. It interrupted an automatic process long enough for the transformation to become visible. The subscriptions were still renewing, but the center of work had already shifted. The applications remained on the screen, yet they were no longer where the work began.

The application is vanishing from the foreground, not because digital systems have become unnecessary, but because intelligence is becoming the interface to them. Behind that interface, files must still endure, records must still be governed, and actions must still be authorized. In front of it, the human task is no longer to press every button. It is to decide what should be done and to remain answerable for what follows.

Photo by Joan Gamell on Unsplash