SYSTEM DESIGN
FIELD NOTE 18 / 18What does “SaaS is dead” actually mean?
AI can change how we use applications, and could make some unnecessary. The useful question is which responsibilities move, which remain and who takes them on.
An application can disappear from view before its responsibilities disappear from the business.
SHARE THIS FIELD NOTE
Pass the useful signal on.
What is actually changing?
Imagine asking your AI to prepare a customer follow-up, update the opportunity and let a colleague know what needs to happen next.
Today, you might do that by opening your CRM, writing an email and sending a message in Slack. In an agent-assisted setup, you could explain the task once, review the proposed actions and let the system carry them out.
You might never open those applications yourself.
That is a meaningful change. But has the software disappeared? Or has the way you use it changed?
I think that distinction gets lost in the phrase “SaaS is dead”.
There is a plausible argument underneath the headline. Understanding it means looking at which parts of an application AI could take over, which parts would remain useful and where the responsibility for running them would go.
AI can be a convenience layer
When we work together in a room, we can talk, point at something and move a sticky note. When we are apart, software carries much of that interaction.
If I explain something to my agent and it gets the message to you, how much does it matter whether I opened Slack myself? I care that you receive the right message, understand it and can respond.
The application can still be there. I simply spend less time navigating it.
That is what I mean by AI as a convenience layer. It can reduce the effort between deciding what needs to happen and getting it done across different tools.
Microsoft describes a similar direction in its account of moving from apps to agents: people express what they want, while applications provide the underlying capabilities, records and permissions. That is a vendor’s architectural view, rather than proof that every business can operate this way today. But it illustrates the distinction well.
In that arrangement, SaaS remains part of the system. The user’s attention moves elsewhere.
Replacing the application is a bigger claim
The stronger version of the argument goes further.
Suppose an AI provider could offer the functions I need directly, coordinate them through agents and give me one place to oversee the work. Some applications in between could become unnecessary.
That is a future I can understand.
But an application usually contains more than the screens we recognise. Behind a customer record are decisions about who can see it, which changes are allowed, how activity is recorded and what happens when two people update it at once.
An agent that sends a message to another agent does not automatically replace those functions.
Direct agent communication already has technical foundations. The A2A protocol, for example, defines how agents from different systems can communicate and collaborate. It establishes a way for agents to interact; it does not establish that the business applications supporting them are redundant.
For an existing application to become unnecessary, its useful responsibilities would have to be removed, recreated or supplied somewhere else.
For a narrow tool, that might be manageable. For a system holding years of customer history and business rules, it is a much larger undertaking.
This is also where the interests of established software companies matter. They can make their existing capabilities available to agents. Salesforce’s AIforce documentation describes that approach: external interfaces can use the platform’s data, automation and rules while Salesforce remains underneath.
An application becoming easier to access through AI could strengthen its role in a business, even if fewer people visit its screens.
Humans will still need somewhere to see the work
My interest in convenience comes with a condition: I want to see what my AI did, check the result and make changes.
Consider that customer follow-up again.
Did the system update the correct opportunity? Did it record a possibility as a commitment? Did it send the message, or only prepare it? What information did it use?
A short confirmation saying “Done” would leave too much unanswered.
The interface might show the proposed email, the exact record changes and any unresolved questions together. I could approve some actions, edit others and stop the rest.
That interface does not have to belong to each underlying application. An AI provider could supply it. It also does not have to be a chat window. Comparing alternatives, reviewing a campaign or correcting several records may be easier in a visual workspace.
The role of the UI can change as more execution happens in the background. It becomes a place for understanding, judgement and control, alongside whatever work people still want to do directly.
Giving an agent access makes that relationship more important. The convenience has to justify the access, and the person granting it needs a clear understanding of its limits.
A subscription can disappear while the responsibility remains
There is another reason to be careful with the headline.
SaaS describes a way of delivering software as a service. If an AI company hosts the capabilities, maintains them and charges for access or use, software is still being delivered as a service. The supplier and commercial arrangement may have changed.
Individual products could lose customers. Some categories could become harder to justify. Businesses could buy fewer separate tools and rely more heavily on one provider.
Those would be substantial changes without amounting to the end of SaaS.
For the buyer, the useful assessment is specific. What work would the replacement perform? What would it cost to run, check and maintain? Who would handle failures? Could the business retrieve its data and change providers later?
Removing a subscription is valuable when the resulting arrangement works better. Otherwise, a visible software expense can turn into less visible integration, review and maintenance work.
Start with the work you want to change
I can see how we might reach a point where agents coordinate more work directly and some familiar applications are no longer needed.
I am cautious about treating that as an everyday operating reality across a business. Different tasks have different requirements, and a successful demonstration leaves plenty of questions about ongoing operation.
For a revenue team, there is already a useful place to start: identify a recurring task where people spend too much time carrying information between applications.
Then examine what the system would need to understand, what it could be allowed to do and how someone would check the result.
That exercise might lead you to connect existing software, replace a narrow tool or build a new capability. Each is a reasonable outcome if it improves the work.
“SaaS is dead” compresses all of those possibilities into one dramatic conclusion.
The question I would ask first is more precise: which part of our software setup could become unnecessary, and what would we need to trust in its place?
Evidence and further reading.
Microsoft’s architectural view of agents coordinating work while applications retain records, capabilities and permissions.
02A2A protocol documentationA technical foundation for communication between agents, not proof that the applications supporting them are redundant.
03Salesforce: Introduction to AIforce for AdminsSalesforce describes external interfaces using its existing data, automation and business rules.
