
An Unexpected Return to Familiar Software
When I read a Gizmodo report in November 2026 announcing that Anthropic was launching Claude Sheets, Claude Whiteboards, and Claude Meetings, my first reaction was a clear sense that something felt off. Just days before its November 19 stock market debut, Anthropic was stepping straight into territory long controlled by Microsoft Office and Google Workspace. The company was introducing a spreadsheet to challenge Microsoft Excel and Google Sheets, a shared brainstorming board, and a video call assistant built to compete with tools like Otter.ai. On the surface, the move looked backward. For the past couple of years, much of the talk around artificial intelligence has pointed toward a future where autonomous agents handle routine work in the background, making traditional office software unnecessary. Seeing a leading AI company invest so much energy in building another spreadsheet raised an obvious question: are we moving forward, or are we rebuilding the old office with slightly smarter furniture?
When I looked closer at how people actually work today, the reasoning behind Anthropic’s strategy made much more sense. Chat windows are remarkably useful for drafting ideas or asking quick questions, yet they fall short when the task involves large tables of numbers, detailed project plans, or group reviews. Scrolling through hundreds of lines of chat text is a poor way for a person to check a financial forecast or compare regional figures across quarters. Until artificial intelligence can handle every decision without supervision, human review remains part of daily business. People still need a shared screen where they can inspect what the model has produced, spot a strange number in a column, and adjust an assumption before sharing the results with colleagues.
There is also a practical business reality behind the product design. Corporate workflows have spent decades taking shape around spreadsheets, shared documents, and slide decks. Asking a company to drop those familiar tools overnight in favor of a blank chat box creates too much disruption, even when the underlying AI is strong. By offering a familiar grid in Claude Sheets and a recognizable board in Claude Whiteboards, Anthropic makes adoption easier while aiming for something larger underneath. If workers spend their whole day inside Microsoft or Google applications, Anthropic risks becoming an invisible engine running behind someone else’s screen. Owning the workspace lets the company hold the user’s ongoing business context directly, instead of watching it stay locked inside rival platforms.
From Independent Programs to Presentation Layers
Once you look past the familiar surface of rows, columns, and whiteboards, a bigger structural shift comes into view. Traditional office software was built around separate programs that each held both the data and the way that data appeared on screen. Information lived inside isolated containers marked by file extensions such as .docx, .xlsx, and .pptx. Because those containers did not share a live context with one another, office workers spent a huge part of their week acting as human bridges between programs. Someone would take notes in a meeting, copy the action items into a Word document, rebuild the numbers in Excel, and paste screenshots of those charts into PowerPoint for a management review.
An AI-centered setup flips that relationship around. Instead of treating each program as a separate vault holding its own copy of the facts, the core data, decisions, and records live in a shared backend designed for machine reasoning. Once the knowledge sits in that common layer, familiar applications step back and act as presentation layers. A spreadsheet stops being the place where numbers are permanently stored and becomes a table view generated when someone needs to check calculations or compare figures. A slide deck becomes a visual summary created for a group discussion, and a written document becomes a readable report suited for careful review. The differences between applications narrow into differences of output format, chosen based on who needs the information and what they plan to do with it.
Seen from this angle, Anthropic’s new office suite looks less like a throwback to old software and more like a bridge to the next stage. When Anthropic highlights that Claude carries the same context from a whiteboard to a spreadsheet to a dashboard without forcing users to repeat themselves, it is pointing toward a world where the walls between tools start to fade. Keeping familiar visual frames intact gives teams a comfortable place to stand while the real center of work shifts underneath. Over time, the grid and the board stop acting as separate worlds and become interchangeable windows looking into the same living body of work.
Living in the Transition Through Markdown and Delegation
This shift is not a distant theory about the future. In everyday personal work, many of us are already living inside its early stages. In my own routine, whenever I ask an AI model to translate materials, combine scattered notes, or help build a report, I use plain text formats as the main source. Markdown has become my standard format for written knowledge, paired with CSV or JSON when the job involves structured numbers or records. Plain text uses far fewer tokens than heavy proprietary document formats, and its clean structure fits naturally with how language models read headings, lists, and relationships.
To keep projects tidy during this transition, I still follow a simple folder layout. A main working folder holds the core Markdown files, paired with an assets folder for images and supporting files, and an output folder set aside for finished deliverables in standard office formats. When I need to translate a PDF, for example, I no longer wrestle with the PDF directly or retype sections into a word processor. I ask the AI to turn the source into clean Markdown first, do the translation and editing inside that text layer, and convert the finished piece back into a PDF or a Word file at the very end as a presentation format.
As a result, direct manual work inside office software has dropped sharply in my daily routine. I rarely open Microsoft Word or Google Docs to type and format a document by hand, and I no longer click through cells, write formulas, or build macros in Excel or Google Sheets. Even when my job involves Google Workspace, where work logs and project records are tracked in Google Sheets and turned into reports in Google Docs, my role has shifted to giving clear instructions and reviewing the results. I explain what I need through chat or voice, let the AI draft the report or update the table, and ask for changes until the result is right. Compared with a few years ago, hands-on operation of office programs has faded into the background, showing in real daily practice that these tools are already turning into output formats rather than places where we build every piece by hand.
The Dilemma Between Local Repositories and Cloud Freedom
Treating Markdown as the main source solves one set of problems, yet it quickly brings up another: where should that knowledge actually live? Right now, anyone building an AI-assisted workflow runs into an awkward split between local file storage and the cloud. Because I do not trust a single computer to hold years of writing and work records, keeping files only on my MacBook is never an option. A sudden hardware failure or a lost laptop would erase everything stored locally. To protect against that risk, I keep my local Markdown files on my MacBook while backing them up to both GitHub and iCloud. The setup works reliably, yet it still feels like a patchwork arrangement built for an in-between period.
Ideally, I would prefer the clean, device-free freedom of cloud storage. Google Drive has long stood for that ideal in my mind, letting a set of documents stay within reach across a laptop, a tablet, and a phone without worrying about local syncing. Pointing Gemini to a specific folder in Google Drive and asking it to work across the files inside feels natural and very convenient. For a long time, however, Google Drive treated Markdown like an afterthought, offering weak preview tools and placing plain text well behind Google Docs. Even though Google recently added a native Markdown viewer, the service still feels like a cloud filing cabinet built for human browsing rather than a fast, lightweight backend for AI agents. Microsoft OneDrive brings its own hurdles, especially on company accounts where strict security rules and complicated permission settings slow down everyday use.
At the same time, the leading AI labs have pulled advanced workflows back onto the local hard drive. Tools like Claude Code and Codex gained their early momentum by reading, editing, and organizing files inside local project folders. With Antigravity joining Gemini’s desktop app, all three major platforms now offer similar ways to work directly on local directories. Seeing Gemini, Claude, and ChatGPT reach parity is welcome, yet the underlying approach still feels surprisingly old-fashioned. Relying on local folders, nested subfolders, and file paths ties modern AI work back to a specific machine and to a filing system invented decades ago for personal desktop computers.
Why the Developer Workflow Is Not the Final Destination
Much of the current enthusiasm for local Markdown workflows comes from software engineers, especially dedicated users of Claude Code who pair their agents with Obsidian and GitHub. Inside the world of software development, that combination makes total sense. Programmers spend their careers working inside local repositories, version-controlled text files, and code editors, so treating personal knowledge like a software project feels familiar and effective. Watching an AI agent move through a local vault of linked Markdown notes and push updates to GitHub matches the exact habits engineers already trust.
From outside that engineering circle, however, the praise for Obsidian often feels like a step backward dressed up as a breakthrough. Users open Obsidian’s network graph view, admire a glowing web of connected dots, and announce that they have built an AI-ready second brain. In truth, that web of lines is mostly visual comfort for human eyes. Beneath the clean interface and community plugins, Obsidian remains a polished cousin of text and code editors like VS Code. It still assumes that knowledge has to be broken into separate text files, sorted into human-readable folders, and maintained through habits borrowed from software engineering.
That same assumption currently holds back what people call vibe coding. Today, vibe coding lets a person guide software creation through plain conversation, yet the whole process still rests on traditional developer tools. Directories, configuration files, Markdown instructions, and Git repositories remain visible at every step. A stubborn bias still shapes much of the AI world: the belief that artificial intelligence works best only when users adopt the tools and rituals of programmers. True vibe coding, and the next real stage of AI-assisted knowledge work, will begin when that bias fades. When Markdown editors, manual folder trees, and GitHub repositories disappear from everyday view for non-programmers, people will stop bending their habits to fit developer tools and start working directly with their ideas.
Dynamic Outputs and the Disappearance of Manual Input
Picturing what comes after the Markdown era means letting go of the idea that a backend knowledge store needs to be readable by humans at all. Today, my workflow still starts with a deliberate step where I ask the AI to convert an incoming document, note, or table into Markdown or CSV. In the next stage, that step should drop away. Instead of naming a file format, a user will hand over raw materials (text, audio clips, photos, screenshots, and numbers) and let the AI store them in whatever app-free backend structure works best for its own memory and reasoning. People do not need to read the raw database any more than a reader needs to inspect the raw code inside an ebook. We only need the system to produce accurate, clear outputs whenever we ask for them.
Taking that idea one step further, even the deliberate act of saving and filing information is likely to blend into normal conversation. I can already see this change in how I communicate every day. Whether I am answering an email or sending a message in a work chat, I barely type on a keyboard anymore. Using AI-powered voice dictation on my MacBook and smartphone, I speak my thoughts out loud in natural sentences and let the software turn them into clean text for colleagues and friends. As speaking becomes the main way we think aloud, ask questions, and give feedback, the AI will continuously pick up context from those exchanges and update its cloud-based knowledge layer on its own. Filing a document into a folder will soon feel as dated as punching holes in paper for a binder.
On the output side, leaving rigid file containers behind opens the door to far more flexible results. Fixed pages and locked grids were compromises from the era of paper printouts and desktop monitors. Tools like NotebookLM already point toward a more fluid approach, taking a group of backend sources and turning them into interactive summaries, audio briefings, or custom web pages shaped around a user’s question. Because modern language models write clean web code with ease, HTML displayed in a browser stands out as the most natural all-purpose presentation format. Unlike a frozen PDF or a static PowerPoint deck, an HTML output can adjust to any screen, let the reader filter numbers or open supporting details on the spot, and live behind a shareable URL that fulfills the device-free promise of the cloud.
Context Boundaries and the Intelligence of Restraint
Even if files, folders, and traditional applications fade into the background, one tough practical problem still stands in the way of a seamless future: how context moves across platforms and stays separated across different parts of life. At the broad level, users today split their time across competing AI ecosystems built by Google, Anthropic, and OpenAI. There is no universal layer like a web browser that lets a thoughtful conversation held in ChatGPT yesterday carry straight into Gemini today. Ironically, the one place where cross-model sharing already works is inside the very local project folders I hope to outgrow. A local skill file or project directory can be read by Codex, Claude Code, and Antigravity alike, giving old file-based engineering habits a practical edge until cloud platforms learn to work together.
Inside a single platform, a quieter tension shows up around memory and personalization. Modern AI systems have become remarkably good at recalling past conversations and matching a user’s habits, yet that strength can turn intrusive fast. Bringing helpful background into a new chat adds depth and saves time, while pulling personal thoughts into a strictly professional request creates awkward overlap. Opening a fresh chat clears the slate and prevents unwanted mixing, yet it also wipes away the shared history that makes an AI partner thoughtful, forcing the conversation back to a shallow start.
For now, I manage these borders by hand through a deliberate split of tools. I use Gemini for personal intellectual work, long-form essays, and everyday reflection, while keeping ChatGPT mainly for business tasks. For structured work that needs repeatable steps (such as translating PDFs or preparing pages for a CMS), I rely on local project folders and skill files so each operational task stays cleanly separated from everything else. It is a practical system for 2026, yet it still depends on human discipline to keep the walls standing. The next real leap in artificial intelligence will not come from storing ever larger oceans of personal data, nor from copying familiar office menus onto new screens. It will arrive when an AI develops an intelligence of restraint, sensing when to draw on shared history, when to shift between work and personal life, and when to leave the past untouched without asking the human being to draw the border by hand.
Leave a Reply