Every technology cycle starts with a simple prediction.
Streaming would kill television.
The cloud would kill on-premise software.
AI agents will now kill software seats.
The logic appears straightforward.
If an AI agent can handle the work of five customer-service employees, a company should need fewer Salesforce licenses.
If recruiting and payroll become more automated, Workday should lose users.
If anyone can create an advertisement with a prompt, Adobe should lose subscribers.
Software companies traditionally charge according to the number of humans using their products. Fewer humans should therefore mean less software revenue.
Except that is not what the latest numbers show.
The seat is not disappearing.
It is getting a machine coworker.
And the software company wants to charge for both.
SaaS is building a second meter
Microsoft now has more than 20 million paid Microsoft 365 Copilot seats.
At the same time, traditional Microsoft 365 commercial seats still grew 6% year over year.
Customers are not simply replacing normal software subscriptions with AI.
They are paying for:
The Microsoft 365 subscription
The Copilot subscription
Autonomous-agent usage through Copilot Credits
The Azure infrastructure underneath it
Microsoft described this transition directly as a move toward “seats plus consumption.”
Copilot Credit consumption nearly doubled sequentially in its latest quarter, while almost 60% of Dynamics service customers were already purchasing usage credits.
The old meter is still running.
Microsoft has simply installed another one.
Salesforce is following the same playbook.
Agentforce annual recurring revenue reached $1.2 billion, up 205% year over year. Customers had consumed 3.8 billion agentic work units.
But Salesforce’s traditional subscription and support revenue also grew 14%.
Again, the core business has not vanished.
Salesforce is adding a new billing layer around the work performed by agents.
The traditional SaaS model was:
Employees × price per seat.
The emerging model is:
Human access
AI assistants
autonomous actions
data usage
completed outcomes.
That changes the investment question.
The important issue is no longer simply how many seats AI will eliminate.
It is how much of the value created by digital labour the software company can capture.
Fewer workers can still mean more software revenue
Imagine a customer-service department with ten employees, each requiring a software license.
An AI agent allows the company to operate with five employees instead.
The software vendor loses five seats.
But the agent now resolves thousands of cases, updates records, checks contracts and routes approvals. If the vendor charges for those actions, its consumption revenue can exceed the lost seat revenue.
The customer saves money because it needs fewer employees.
The software company can still earn more because it is now charging for the work rather than only for the worker.
That is the real SaaS-AI thesis.
AI can reduce human employment without reducing software spending.
But only if the vendor controls the workflow, the data and the billing system.
The SaaS cannibalization game
Software incumbents face three choices.
Protect the existing seat
A company can limit automation and preserve its current pricing model.
That protects revenue in the short term.
It also gives competitors an opportunity to offer customers a more automated product.
Cannibalize itself
A company can build agents that reduce the number of humans required to use its own software.
That threatens seat revenue.
But it gives the vendor a chance to replace seats with consumption, transaction or outcome-based pricing.
Allow another platform to cannibalize it
This is the worst outcome.
A Microsoft, OpenAI, Anthropic or internally developed agent becomes the customer’s main interface.
The underlying SaaS product may remain necessary, but it is reduced to a background database.
The agent platform owns the customer relationship.
The SaaS vendor keeps the cost of maintaining the system while another company captures the pricing power.
That makes self-cannibalization the rational move.
Not because software companies want to destroy their seat businesses.
Because they cannot afford to let somebody else do it first.
