The Token Dividend
OpenAI, Google, Anthropic, and SpaceXAI will not pay Calendly for us. They will recreate scheduling—and many other narrow software functions—with their own models and code, then compete to capture the subscriptions we no longer pay as token spend.
The app bill is being rewritten.
The prediction, precisely stated
The labs will absorb the function, not the vendor invoice.
The future is not Claude quietly buying me a Calendly subscription. The future is Claude no longer needing Calendly to perform ordinary scheduling.
That distinction changes the economics. An aggregator keeps the underlying subscriptions alive and tries to become the interface above them. An absorber recreates enough of the underlying capability that the separate line item disappears from the user’s wallet.
Today I may pay one company to schedule meetings, another to host a small site, another to collect a form, another to move data between services, and another to turn the resulting information into a dashboard. Each subscription exists because software had to be designed in advance, generalized for a market, hosted as a product, and operated through its own interface.
Generative code changes that premise. A frontier AI company already owns the intelligence, the coding agent, the identity layer, the execution environment, and the meter. If it can create a small scheduling surface, a lightweight site, a recurring workflow, or a one-purpose dashboard inside the AI product, it does not need to reimburse an app company. It can make the feature part of the intelligence service.
The token dividend is the monthly software spend released when an AI platform recreates a narrow app function—and then captures some or all of that released budget as model usage.
This is the wallet consequence of the move from apps to stacks. The First Stack Generation will not necessarily assemble a larger pile of paid software. It will increasingly ask one intelligence layer to generate, operate, and revise the small capabilities that used to require separate products.
The apps do not all vanish. But many of their invoices can.
Interactive forecast
Run the wallet conversion.
The amounts below are deliberately illustrative, not price forecasts. The mechanism is the prediction: recurring app rent becomes available for variable AI consumption.
Six fixed subscriptions
One intelligence budget
Where absorption begins
Thin apps are exposed first.
The easiest subscription to absorb is the one whose value is mostly a small interface wrapped around rules the AI can understand and code it can regenerate.
Scheduling
Availability rules, booking pages, reminders, and rescheduling are legible workflows. Once the AI controls calendar context and can generate the surface, the separate personal scheduler becomes optional.
Light hosting
A landing page, calculator, internal portal, or small public tool can be written, deployed, and revised inside the AI environment without a separate site-builder relationship.
Forms
Intake questions, conditional paths, validation, and structured output can be generated for one purpose instead of rented as a generalized form product.
Workflow glue
Many automations are simple translations: when this happens, read that, transform it, and place the result here. Agents can increasingly own the translation directly.
Dashboards
A recurring question does not always need a permanent analytics product. The stack can create a living view around the question and rebuild it when the question changes.
Document utilities
Formatting, conversion, comparison, extraction, and assembly are becoming native acts of the model rather than reasons to maintain a shelf of single-purpose subscriptions.
They are learning to own the capability.
Anthropic has made recurring work native.
Claude Cowork now supports scheduled tasks that run remotely on a cadence and produce finished outputs. That is not yet a full replacement for every scheduling product, but it moves scheduling logic and recurring execution into Claude itself. Anthropic’s documentation describes the native scheduling layer.
OpenAI can build and deploy the small app.
ChatGPT Sites lets users create, publish, and share lightweight websites and interactive apps from Codex. A category of work that once required a site builder, hosting account, and development handoff can now remain inside the AI workspace. OpenAI describes Sites as a home for focused websites and lightweight apps.
Google owns both intelligence and adjacent rails.
Gemini can create and manage calendar events, while Google AI Studio can generate production apps and deploy initial projects directly to Google Cloud. Google does not have to cross as many company boundaries to turn intelligence into an operating capability. Gemini already acts on Google Calendar, and AI Studio now builds and deploys apps inside Google’s own stack.
SpaceXAI can generate and publish the product surface.
Grok Build creates working apps and websites, publishes them on its own URLs, supports custom domains, and can activate SpaceXAI models without requiring users to manage separate API keys. Its product announcement already combines code generation, hosting, distribution, and model consumption.
What the prediction does not claim
Software remains. The retail app layer thins.
Absorption is not magic. The AI can internalize code and interface faster than it can replace regulated rails, proprietary networks, durable records, or institutional accountability.
Code-shaped value
- simple interfaces and booking pages;
- small databases with replaceable schemas;
- single-purpose transformations;
- lightweight workflow routing;
- generic publishing and hosting surfaces.
Rail-shaped value
- systems of record with deep history;
- payments, identity, telecom, and regulated transactions;
- marketplaces with proprietary supply or demand;
- high-stakes governance and contractual accountability;
- specialized infrastructure whose reliability is the product.
The contest for wallet share
The subscription savings will not be left unattended.
The AI company does not need every absorbed dollar. It only needs the user to feel that more tokens cost less than rebuilding the old app stack.
That is why usage credits matter beyond the immediate question of limits. In the preceding essay, I argued that people buy continuity when an agent is already inside valuable work. This prediction extends the same logic across the household or business software budget.
If the AI platform removes four or five monthly subscriptions, the user’s reference point changes. A $50 refill no longer competes only with waiting for the weekly limit to reset. It competes with the $75 or $150 of app rent the AI layer has made unnecessary. The model company can raise its share of the wallet while the user still experiences consolidation, convenience, and possibly a lower total bill.
This is why I expect the frontier companies to build more native execution surfaces even while they promote connector ecosystems. Connectors are the bridge into the current world. Generated native capabilities are the route to owning the next one.
The strategic prize is not being the app that sits on top of all the other apps. It is becoming the environment in which fewer separate apps need to exist for the user at all.
My prediction
The AI winner will not merely earn a new subscription. It will inherit the old ones.
OpenAI, Google, Anthropic, and SpaceXAI are competing for more than the monthly AI line item. They are competing to turn scheduling, hosting, forms, automation, dashboards, and document utilities into generated capabilities inside their own systems. The apps they absorb will release a token dividend. The company that converts that dividend into useful completed work will take the largest share of the human AI wallet.