AI agents could change who interacts with business software, how often a screen gets opened and whether a seat remains a good proxy for value. They could also create more usage and new types of software business.
That is a forecast, not a proven outcome.
It still feels uncomfortably familiar.
Collab365 spent years helping Microsoft 365 professionals through events and long-form content. Then generative AI changed how people found answers and how quickly product guidance aged. Helen and I have been rebuilding the business around that change.
So when somebody says agents will hollow out software interfaces, I do not hear a distant theory. I hear a smaller version of a problem we have already lived.
The screen may stop being the main event
Most software pricing grew around human users. A person received a licence, signed in and clicked through the product.
An agent changes the interaction. It may retrieve a record, update a field, prepare a summary or trigger a workflow while the person stays in chat.
The underlying software is still doing valuable work. The difference is that fewer people may need to visit its interface for every step.
The tempting leap is to claim that a 200-person team will become 40 people and a vendor will lose 80 per cent of its seats. We cannot know that. Agents may augment people, create new usage or require new controls rather than simply removing users.
The safer conclusion is that vendors need more than one way to measure value.
Microsoft already uses more than one charging model
Microsoft's current Copilot licensing documentation shows the mixed model clearly. Some experiences are covered by a Microsoft 365 Copilot add-on licence. Other agent scenarios can use metered, consumption-based billing through Copilot Credits, particularly when a user without that add-on accesses shared tenant data through eligible Microsoft 365 or Copilot Chat access.
That does not prove per-seat SaaS is ending. It proves that agent activity does not always fit neatly into a user-seat box.
The commercial questions become:
- Is value created by a person having access, by an action completing or by compute consumed?
- Who approves an agent's actions?
- How are identity, permissions and audit records preserved?
- What stops an automated loop creating an unexpected bill?
- Which screens remain essential for judgement and exception handling?
Those are not cosmetic pricing questions. They change product design, procurement and support.
What our own experience taught me
Our previous model was not destroyed by one Microsoft product on one morning. Search behaviour, customer expectations, content freshness and AI-assisted answers all moved together.
Calling that "Copilot gutted our business" made a punchy headline. It gave one product too much credit for a much wider change.
The useful scar is this: do not confuse the current delivery mechanism with the value underneath it.
If customers value a resolved problem, protect that outcome. Expect the interface, pricing unit and route to the outcome to move.