There was a time when the promise of software was access.
You logged in. You clicked through menus. You paid, quietly and predictably, for a seat.
The arrangement was simple, and for a while, it seemed almost natural: the more people who needed the software, the more seats you sold, and the more valuable the company became.
Now, quietly and more unsettlingly, that logic is beginning to fray.
We find ourselves in a moment where the interface is no longer the center of gravity. Where the user, once the unit of value, is no longer the only one doing the work. And where a new and slightly unnerving question has begun to circulate:
What, exactly, are we paying for?
The quiet shift
The most significant change is not that AI can write code or generate images.
It is that AI can do things.
Not assist. Not accelerate. But execute.
An AI agent does not wait for a user to log in. It runs workflows, completes tasks, and delivers results independently.
And once this happens, something subtle but profound breaks.
The old assumption was that value scaled with the number of people using the software. More users meant more work, more output, more value.
But if software itself performs the work, that assumption becomes difficult to maintain.
As one analysis puts it, the value is no longer tied to how many users log in, but to what the software actually does.
The unit of value begins to shift:
From seats → to actions From access → to outcomes
And with it, the business model begins to move.
The erosion of the seat
Faced with this reality, many have reached for a dramatic conclusion:
AI will kill SaaS.
But this is not quite right.
What AI threatens is not the existence of SaaS, but one of its most cherished assumptions: that charging per user makes sense.
For decades, per-seat pricing was elegant because it was easy to understand. It aligned with how software was used.
Now, it does not.
AI agents behave less like users and more like workers. They perform tasks continuously, unevenly, and often invisibly.
A single workflow may involve hundreds of steps, multiple systems, and no human interaction at all.
To charge for this as if it were a person logging in begins to feel, if not incorrect, then at least incomplete.
And so the industry is beginning, slowly, to adjust.
Usage-based pricing. Outcome-based pricing. Hybrid models that attempt to reconcile the old world with the new.
Even the most traditional players are experimenting with combinations of seats and usage, trying to align price with value in a world where value is harder to see.
This is not a collapse. It is a recalibration.
The strange fear of replacement
Alongside this economic shift, a more theatrical fear has taken hold.
It is often expressed in a particular way:
That AI, combined with a little ingenuity, might allow anyone to recreate complex enterprise software in an afternoon.
The phrase varies. The sentiment does not.
“Why pay for Salesforce,” it asks, “when you could just build it?”
There is, in this idea, a certain appeal. It flatters us.
It suggests that what once required years of effort, teams of engineers, and accumulated knowledge might now be conjured with a few prompts and a free afternoon.
But this, too, rests on a misunderstanding.
The muted weight of software
Enterprise software is not merely code.
It is memory.
It is the accumulation of edge cases, exceptions, regulatory constraints, integrations, and decisions made long ago by people who are no longer present.
It contains:
- —workflows shaped by real organizations
- —data models refined through use
- —assumptions tested, broken, and rebuilt
These things are not visible in the interface.
They are not easily reproduced.
They are, in a sense, the software’s history.
AI can generate forms. It can scaffold interfaces. It can approximate functionality.
But it does not, in an afternoon, recreate the slow sediment of complexity that defines a true system of record.
To believe otherwise is to confuse the surface with the substance.
The unbundling
What is happening instead is more subtle.
AI is not replacing software wholesale. It is slipping between the user and the application.
Where once a person navigated a UI, now they might ask a question.
Where once they clicked through workflows, now those workflows may be orchestrated on their behalf.
The application does not disappear.
It recedes.
It becomes, increasingly, something like infrastructure: present, necessary, but no longer the place where the user spends their time.
This gives rise to a peculiar inversion.
The more powerful the software becomes, the less visible it may be.
The redistribution of value
And so the question returns, slightly altered:
If users no longer log in, where does value live?
Not in the interface. Not in the seat.
But in three more latent places:
- —the data the system holds
- —the workflows it governs
- —the outcomes it produces
This is why some software will prove more resilient than others.
Systems of record, with deep data gravity and embedded logic, are difficult to replace.
Thin layers of convenience, whose value lies primarily in their interface, are more exposed.
The distinction is not always obvious from the outside.
But it becomes clear when the interface is removed.
The compromise
And yet, as with many technological shifts, the reality is neither total disruption nor total continuity.
SaaS companies are not abandoning their models overnight.
They are adapting them.
Behind the scenes:
- —AI performs tasks
- —usage is tracked
- —costs are calculated
On the surface:
- —pricing remains legible
- —interfaces remain familiar
- —seats, in some cases, persist
Hybrid models emerge not because they are elegant, but because they are tolerable.
They allow both provider and customer to adjust, gradually, to a world in which the old assumptions no longer quite hold.
The deeper unease
What gives this moment its particular tension is not only economic.
It is philosophical.
For years, we equated software with interaction.
To use something was to engage with it directly.
Now, we are entering a world in which software acts without being seen.
Where work is done without a user present. Where value is delivered without a login.
And this raises a quiet, unsettling possibility:
That what we once thought of as “using software” may no longer be the primary way in which software creates value.
