For more than twenty years, the software industry ran on a simple equation: building a system was expensive, slow and required large teams; keeping it online cost almost nothing. The monthly subscription was born to amortize that upfront investment. Every month you were, in practice, paying for the work someone did to design the screens, model the database and write the business rules.
That equation has changed. And like any shift that touches the revenue model of an entire industry, it deserves careful discussion, not slogans.
What AI did to the cost of building
A huge share of the software we use every day is what developers call CRUD: create, read, update and delete records. Customer databases, inventory control, appointment calendars, to-do lists, an expense spreadsheet with an interface on top. These systems are useful, often indispensable, but technically predictable.
That is exactly the kind of software AI has learned to build best. Coding agents now write, in minutes, the form, the validation, the filtered list and the API that used to take a team weeks. Tools that generate apps from a plain-language description already put this within reach of people who have never programmed. In César itself, a request like "I want to track tool loans in my building" becomes a working app in the same conversation.
This does not mean software engineering stopped mattering. Security, performance, reliable integrations and well-designed products still require experienced people. But the layer that weighed most on cost, turning a process into screens and tables, became far cheaper. And what becomes cheap to produce tends to become cheap to buy.
Subscriptions for CRUD lost their justification
When any company can have its own database, and an assistant can assemble a custom one in minutes, charging $20 a month per user to access forms and reports is hard to defend. Not because the software is bad, but because the value it delivers no longer lives in the existence of screens.
There is also a fairness problem subscriptions always carried, and it is now more visible:
- Light users subsidize heavy users. The customer who opens the system twice a month pays the same as the one who depends on it all day.
- Price does not track value. A busy month and an idle month cost the same.
- Per-seat pricing penalizes growth. A company hires someone and the bill goes up, even if the vendor's effort is unchanged.
- Customers pay for inertia. Many subscriptions stay active because cancelling is a hassle, not because they deliver something every month.
None of this is new. The difference is that while building was expensive, there was no realistic alternative. Now there is.
What is still expensive
If building got cheap, what still costs real money? The answer lies in what happens after the software exists, every time it works for you.
AI at runtime. Every time a model reads an email, understands a bill, summarizes a meeting or decides which action to take, specialized hardware is being consumed. A simple request costs a fraction of a cent. A task that involves searching the web, reading long documents and executing ten steps costs much more. That cost is real, variable and proportional to use.
Physical-world and third-party resources. A phone call, a WhatsApp message sent to a contact, issuing an invoice, generating an image, editing a video, querying a paid database. Each has a provider on the other side charging per unit.
Continuous autonomous work. An assistant that watches your inbox, tracks your flight status and warns you about a due date before you remember is working even when you are not looking.
These are the costs that grow with use, and that is precisely where people perceive value. Nobody cares how many tables a system has. People care about the bill that was not late, the appointment rescheduled without a phone call, the invoice issued on its own after a sale.
Charge for what software does, not for what it is
The natural consequence is a model where price follows the work performed: credits consumed as AI and expensive resources are used. This is not an invention. It is how electricity, telephony and cloud computing have always worked. You pay for what you consume, and each unit has a clear cost.
Done well, this model brings concrete advantages to the people paying:
- Proportionality. Light users pay little. Those who extract a lot of value pay more, and that makes sense.
- Transparency. You can see what credits were spent on: a search, a call, a generated document.
- Aligned incentives. The vendor earns when the customer uses the product and gets results, not when the customer forgets to cancel.
- Freedom to grow. Adding people, creating apps and storing data costs nothing extra. What costs is the heavy lifting.
Where care is needed
It would be dishonest to present usage-based pricing as risk-free. It solves old problems and raises legitimate concerns that need answers:
Predictability. Companies and families plan budgets. A bill that swings from month to month creates anxiety. Fixed-price credit packs, monthly allowances included in the plan and usage alerts solve most of this.
Fear of using. If every request feels like a taxi meter running, people ration usage and lose exactly the value of the product. Cost per action must be low enough that everyday use feels relaxed, leaving meaningful consumption for what is genuinely heavy.
Clarity. "Credits" cannot become an opaque currency. Customers need to understand, before asking, whether something is cheap or expensive, and be able to check their balance at any time.
No surprises. A warning when the balance is running low, never an unexpected automatic charge, and expensive tasks confirmed before they start.
That is why the most sensible model is rarely an extreme. An affordable monthly base that covers access and a usage allowance, combined with credits for anything beyond it, usually balances predictability and fairness.
How César charges
César was designed around this logic. Structure, such as organizing contacts, calendar, bills, trips, packages and even the apps you create through conversation, is not what drives the price. What consumes credits is real work: AI thinking and acting, research, images and videos, calls, messages sent on your behalf.
There is a free plan with credits every month, for anyone who wants to use it day to day without paying. Paid plans include a larger allowance, and you can check your balance by simply asking César. When the balance gets close to empty, it warns you ahead of time, with no surprise on the bill.
The same idea applies to businesses on César Empresas: a company does not pay per seat or per screen. It pays for the customer service the assistant handled, for automatically issued invoices, for work that no longer has to be done by hand.
The end of a model, not of software
None of this means software as a product will disappear, or that every subscription is unfair. Systems with heavy infrastructure, regulatory responsibility or critical data still have real fixed costs that justify recurring fees.
What is ending is the idea that screens and records, on their own, are worth a monthly fee. When AI drives the cost of building close to zero, value moves elsewhere: to what software can do in the real world on behalf of the people who use it. It is fair that price moves with it.
For software buyers, the question worth asking at every renewal is simple: am I paying for the work this system does for me, or just for the right to keep opening the same screens?
If you want to try an assistant that charges for what it does, César has a free plan and takes five minutes to start. See also how AI is changing the way we use software.