The next form of software
From App to Stack
The app organized millions of people around one finished product. AI makes the opposite arrangement possible: software continuously reorganized around one person.
00 · The inherited form
The app feels natural only because it arrived first.
For the last twenty years, we have been trained to think in apps.
Need to edit a photograph? Find an app.
Need to manage customers? Find an app.
Need to create an invoice, design a presentation, track expenses, plan a trip, prepare your taxes, organize a project, or manage your health records? There is an app for that.
The app has become so familiar that it feels like the natural form of software.
It is not.
An app is the form software took when programming was scarce, interfaces had to be designed in advance, and one finished product had to serve a large population.
Software was expensive to create. A team of programmers had to anticipate what millions of people might want to do, write the code, design the screens, test every button, publish the product, and maintain it for years.
That cost required scale.
The same accounting software had to serve millions of businesses. The same design software had to serve millions of creators. The same customer-management software had to serve millions of salespeople.
The result was the app: one fixed piece of software designed for everyone.
Artificial intelligence changes the economics that created it.
01 · System note
Software Is Language
People tend to treat computer code as something fundamentally different from ordinary writing.
It is not.
A programming language is exactly what its name suggests: a language.
It has vocabulary, grammar, syntax, structure, and meaning. Instead of using English to write an article, a programmer uses Python, JavaScript, SQL, or another language to tell a computer what to do.
A large language model can write code for the same basic reason it can write this article.
It has learned the patterns of the language.
It can write a sentence in American English. It can write a paragraph. It can write a chapter. It can revise the chapter when you tell it what is wrong.
It can do the same thing with software.
It can write a function, build a database query, create a web page, design an interface, connect one process to another, and revise the code when the result does not work as intended.
To the AI, code is not a mysterious substance. It is another language.
This means programming is no longer as scarce as it once was.
Software does not always have to be designed months in advance by a company hoping to sell the same product to millions of people. It can increasingly be written at the moment it is needed, for the person who needs it.
That is the beginning of the shift from app to stack.
02 · System note
The App Was Built Before You Arrived
Every app contains assumptions about you.
Its designers decided which information matters, which fields you should complete, which sequence you should follow, and which choices you should be allowed to make.
They decided what a customer is.
They decided what an expense is.
They decided what a project is.
They decided what a report should look like.
They decided what belongs on the screen and what does not.
You arrive after all those decisions have been made.
You can usually adjust a few settings, choose a template, add a custom field, or pay for a more expensive version. But you are still operating inside software designed for a large population.
You must learn the app’s language.
You must understand its menus, categories, screens, and workflow. You must adapt your work to the software because the software cannot fully adapt itself to you.
We have lived with this arrangement for so long that we rarely question it.
But once AI can write and rewrite software through ordinary conversation, the limitations of the mass-market app become much harder to accept.
Why should you spend hours learning how an app works when you can simply describe what you want?
Why should your business operate according to categories created by a software company?
Why should you wait for a feature request to become popular enough for a vendor to build it?
Why should ten million people receive the same interface when no two people work exactly alike?
The app made sense when software had to be built once and distributed to everyone.
It makes less sense when software can be created for you.
03 · System note
Claude Cowork and ChatGPT Work Are Teaching a New Habit
Claude Cowork and ChatGPT Work are important not merely because they can perform useful tasks.
They are teaching people to think differently.
Instead of opening a spreadsheet, arranging columns, writing formulas, and formatting a report, you say:
“Analyze these expenses and show me where spending increased.”
Instead of opening presentation software, selecting a template, and building slides, you say:
“Turn this material into a presentation for tomorrow’s meeting.”
Instead of opening several applications to research a customer, update a record, write an email, and prepare a proposal, you say:
“Prepare everything I need for my meeting with this customer.”
Today, these systems often accomplish the work by operating existing apps.
That is an intermediate stage.
They use the tools available because our information is still scattered across accounting apps, design apps, email apps, project apps, and customer-management apps.
But the deeper lesson is already taking hold.
People are discovering that they do not want to operate software.
They want software to operate on their behalf.
Once that expectation becomes normal, the app begins to look like unnecessary machinery standing between the person and the result.
04 · System note
Connecting Apps Is Not the Destination
The current solution is to connect AI to apps.
You grant the AI access to your email, accounting system, calendar, documents, customer records, and other services. The AI then attempts to work across them.
This sounds sensible, but anyone who has tried it knows how fragile it can be.
Access is difficult to grant.
Permissions are confusing.
Connections break.
Authentication expires.
One system allows reading but not writing.
Another exposes only part of its functionality.
A feature available through the app is unavailable through its API.
A connector works one day and fails the next.
The app may contain your data, but it decides how much of that data your AI may access and what your AI may do with it.
The entire arrangement is backward.
Your AI should not need permission from a dozen software companies to work with your information.
Your information should belong to you.
The AI should work directly with your information.
The app is currently the owner of the environment, and your AI is treated as a visitor. It receives limited access under rules created by the app company.
The personal stack reverses that relationship.
Your environment belongs to you.
Outside services become temporary tools that your AI uses when necessary.
Move the center of the system.
Choose a mode. The tools may look similar; the ownership relationship is not.
You adapt to software built before you arrived.
The app defines the categories, workflow, interface, and access. Your information lives inside its environment.
The AI works across apps as a visitor.
It can delegate and coordinate, but permissions, APIs, authentication, and product boundaries still define what it can do.
Software is created inside an environment you own.
Your data endures. Models can change. Interfaces can appear for a task and disappear when the task is complete.
05 · System note
One Person, One Database
The foundation of the personal stack is not an app.
It is your own data environment.
Imagine that every individual has a private database account—something conceptually similar to having your own Supabase environment.
The important point is not the particular technology. The important point is ownership and organization.
Your information lives in your database.
Your files live in your storage.
Your AI works with both.
At the highest level of the database is not a company, an application, or a software vendor. Beneath you may be your businesses, projects, properties, accounts, relationships, purchases, documents, schedules, health records, travel, education, and anything else that belongs to your life.
Today, the same person is divided across hundreds of systems.
Your financial life is inside one company’s application.
Your customer relationships are inside another.
Your photographs are somewhere else.
Your medical information is scattered among providers.
Your documents are divided across drives, inboxes, portals, and devices.
Every app creates its own partial version of you.
The personal stack brings those fragments back together.
It creates one enduring environment organized around the individual.
This does not mean one giant table containing everything. A proper database can contain many tables, relationships, permissions, and boundaries.
The difference is that those structures belong to you.
You are not a record inside someone else’s application.
The applications and services become records inside your environment.
06 · System note
The AI Does Not Store Your Information
The AI itself is not the database.
This distinction matters.
A language model does not need to store your tax records, customer history, medical files, or property information inside itself.
The model provides intelligence.
The database provides memory.
The AI reads from the database, reasons about the information, performs work, and writes the results back.
That is a far cleaner arrangement than expecting the AI model itself to remember everything.
The model may change.
You may use a different model next year.
You may use several models for different purposes.
Your data should survive all of them.
The enduring asset is your personal stack: your information, structure, permissions, history, and procedures.
The AI is the intelligence that works with it.
The database is where your actual records live.
07 · System note
“Do My Taxes”
Consider a familiar task.
Today, doing your taxes usually means finding tax software, creating an account, uploading documents, answering a long sequence of questions, entering information the government or financial institutions may already possess, reviewing unfamiliar forms, paying a fee, and authorizing submission.
The app controls the process.
In the personal-stack model, you simply say:
“Do my taxes.”
Your income records are already in your database.
Your business expenses are already categorized.
Your charitable contributions, property information, investments, payroll records, prior returns, and supporting documents are already part of your stack.
Your AI gathers the relevant information, applies the current tax rules, identifies uncertainties, asks you the few questions that require human judgment, prepares the return, and presents the result for approval.
There is no tax app.
There is no permanent tax interface.
There is no new account.
There is no need to upload another copy of information you already own.
A temporary screen may appear so you can review the result. It may show what you owe, explain the important decisions, and ask whether you approve filing.
When the work is complete, the screen can disappear.
The records remain in your stack.
This is a very different idea of software.
The software exists for the duration of the task.
08 · System note
Software Becomes Temporary
Suppose you are considering buying a rental property.
You ask your AI to evaluate the property.
It creates the necessary database tables, gathers the available information, builds a financial model, creates a comparison screen, and lets you adjust the assumptions in ordinary language.
“Assume the roof must be replaced in three years.”
“Use a lower occupancy rate.”
“Compare this property with the other two.”
“Show me the price at which the investment becomes attractive.”
For a few hours, you have what looks like a specialized real-estate investment application.
But there is no app company.
There is no subscription.
There is no onboarding.
There are no features you do not need.
The AI writes the necessary code, creates the temporary interface, and connects it to your data.
When the decision is finished, the interface can disappear.
The useful information remains.
The software was created for one purpose and one person.
That is the personal stack.
09 · System note
From Configuration to Conversation
An app allows configuration.
A stack allows conversation.
Configuration means choosing from options the developer anticipated.
Conversation means changing the system in ways no developer had to predict.
You can say:
“Stop organizing my customers by industry. Organize them by the problem they hired us to solve.”
“Add a separate approval process for expenses above $500.”
“Do not create a weekly report. Contact me only when something unusual happens.”
“Combine these two categories. I no longer make that distinction.”
“Create a simpler screen for this process.”
“Remove that screen. I do not need it anymore.”
The AI changes the code, database, rules, or interface accordingly.
You are not requesting a feature from a software vendor.
You are changing your own system.
This could happen every day.
A personal stack is not a finished product. It evolves as quickly as your thinking does.
10 · System note
The Personal Operating System
Over time, your database, files, rules, history, agents, and temporary software form something larger than a collection of tools.
They become your Personal Operating System.
Not an operating system for the computer.
An operating system for your work and life.
It knows how your information is organized.
It knows which decisions require your approval.
It knows what can be completed automatically.
It knows how you prefer reports, messages, plans, and analyses to be prepared.
It knows your businesses, projects, property, responsibilities, and relationships because those things are represented in your own data environment.
You speak to it in ordinary language.
It writes the code it needs.
It creates interfaces when they are useful.
It removes them when they are not.
It changes as you change.
The personal stack is not a collection of apps connected to AI.
It is the replacement of apps by an environment that can continuously create the software you need.
11 · System note
From App to Stack
The app will not disappear immediately.
For some time, Claude Cowork, ChatGPT Work, and other AI systems will continue operating the applications we already use. That is how the transition begins.
- First, the AI helps us use apps.
- Then it operates the apps for us.
- Then we begin questioning why the apps are there.
- Then our data moves into environments we own.
- Then the AI writes more of the software directly.
- Then interfaces become temporary.
- Then the app stops being the center of the digital world.
- The individual becomes the center.
The app was software built once for millions of people.
The stack is software continuously created for one.
That is the shift now beginning.
From app to stack.