Why SaaS, the dominant model of the last twenty years of enterprise software, is being quietly replaced by something with a different shape. We are calling it Knowledge as a Service.
The most valuable moment in a conversation with Claude or ChatGPT is not the answer. It is the suggestion at the end. The brilliant prompt for what to ask next. The line that makes you realize you were thinking too small.
If you have used any modern AI assistant for more than a week, you know the pattern. You ask a question. You get an answer. And underneath the answer, the system surfaces three more questions you had not thought to ask, often better than the one you started with. You follow one of them, get another answer, and three more open up. Twenty minutes later you have understood something you did not know you were trying to understand.
That dynamic is not a feature. It is a category. And it is, in our view, the shape that the next twenty years of enterprise software will take. We have started calling it Knowledge as a Service. KaaS. This post is about what changes when software stops being a tool and starts being a thinking partner.
SaaS Was Right For Its Era
Software as a Service was the dominant business model of the last twenty years of enterprise software, and it was right. Selling perpetual licenses on installed software was a worse fit for how businesses actually used software than a subscription with continuous updates. SaaS unlocked a generation of companies, and it produced enormous value.
But the shape of a SaaS product is unmistakable once you start looking for it. You license a tool. You use the tool. The tool sits there, a box of features, until you decide to use it again. The value of the contract is, in practice, capped by the imagination of the user. If your team only thinks of three ways to use Salesforce, you get the value of three workflows. If your analyst only knows ten queries to run in Tableau, you get the value of ten dashboards. The software does not push you to ask the eleventh question. It does not even know there is an eleventh question. It waits.
This was fine in the era it was built for. Humans were the only intelligent component in the system; software was the storage and the workflow. The intelligence layer was you. The tool was a place to put your intelligence.
What Changed
Two things have changed at the same time, and they reinforce each other.
First, the cost of the intelligence layer has collapsed. A capable language model can now read a document, draft an analysis, propose three counterarguments, and suggest the next investigation worth running, all in seconds and for fractions of a cent. The intelligence is no longer scarce. It is sitting next to the storage. It is free to participate.
Second, and more important, the shape of how people interact with capable AI has trained an entire generation of users to expect software that does more than wait. The expectation is no longer “give me what I asked for.” It is “give me what I asked for, and tell me what I should be asking next.” Once you have used software that behaves this way, going back to software that just sits there feels like reading a book that refuses to turn its own pages.
This second shift is the one that matters more for the category argument, because it changes what users are willing to pay for. The contract value of a SaaS tool was capped by the user’s imagination. The contract value of a system that expands the user’s imagination has no obvious cap. Every answer creates the next question. Every interaction compounds into a deeper one. The platform’s value grows with the user’s curiosity, rather than depreciating with the user’s familiarity.
“The contract value of a SaaS tool is capped by the user’s imagination. The contract value of a system that expands the user’s imagination has no obvious cap.”
The Three Properties of KaaS
If you accept that something has changed, the next question is what to call the thing that has replaced SaaS, and how to recognize it when you see it. We think Knowledge as a Service is the right name, and we think it has three identifying properties.
It pulls the user forward rather than waiting. The platform does not just answer the question; it surfaces the next question worth asking. The user moves through the system in a sequence the system is helping to construct, not a sequence the user has to imagine on their own. This is the shift that AI assistants have already made obvious in the consumer space and that enterprise software is now starting to absorb.
Its value compounds with engagement rather than depreciating with familiarity. SaaS contracts mature into renewal conversations, and renewal conversations turn on whether the customer used enough of the feature set to justify the seat count. KaaS contracts mature into expansion conversations, because the customer is using the platform to ask questions they could not have formulated when they signed the contract. The renewal is not a defense; it is a foregone conclusion, and the conversation is about how much further the customer wants to go.
It treats data and context as the product, not the substrate. SaaS products built features on top of data. KaaS products treat the data, and the connections between data, as the actual deliverable. The interface is just how the customer touches the data. This is why so many of the AI-native companies that look like SaaS at first glance are not, structurally, SaaS at all. They are selling access to a continuously growing knowledge graph that happens to have a UI on top of it. The graph is the product. Everything else is delivery.
The Hardest Domain Is the Physical World
The easy version of this argument is that consumer AI assistants are KaaS, and therefore the category generalizes. We think that undersells what is happening, and it understates how difficult the category actually is to build outside a chat interface.
KaaS-style behavior is starting to show up across enterprise software wherever the underlying data is rich enough to support it. Legal research platforms surface the next case to read. Financial analytics systems flag the correlation the analyst did not know to query. Industrial monitoring platforms catch the failure mode the operator did not know to look for. In each case the platform stops being a passive tool and becomes an active participant in the user’s thinking. That is the dynamic. But the difficulty of delivering it varies enormously by domain, because the data underneath is structured differently.
Text is the easy case. Most of the consumer AI experience that has trained users to expect KaaS behavior is happening on top of language data, which has a single uniform structure: tokens in sequence. A model that has read enough text can plausibly suggest the next thing to read. The substrate is forgiving.
The physical world is the hard case. Sensors, satellites, infrastructure logs, building telemetry, climate observations, IoT streams. The data is fragmented across thousands of incompatible sources, indexed in incompatible coordinate systems, captured at incompatible time scales, owned by parties with incompatible governance models. To deliver KaaS behavior on top of physical-world data, the platform has to do all of the following before it can even attempt to surface the next question: federate across the sources without forcing them into a single schema, index everything natively in space and time, encode the relationships between objects in a queryable knowledge graph, and reason across the whole thing at the speed of a conversation. None of this is trivial. Most of it has not, until now, existed.
This is the work we have been doing at SeerAI for the last five years. Geodesic, our platform, is the AI infrastructure layer that turns physical-world data into something a model, an agent, or an enterprise can reason on the way you reason with Claude. A facilities team asks where indoor air quality is degrading fastest across their portfolio. Geodesic answers, and surfaces the question they had not thought to ask: which buildings show the same degradation pattern correlated with seasonal outdoor air quality, and what intervention sequence has worked at peer buildings. An analyst asks for a fusion of two data feeds. Geodesic answers, and suggests a third feed that materially changes the picture. Every answer becomes the next question. Every customer interaction becomes a deeper interaction.
We mention this not to claim that Geodesic is the only example of KaaS, but to point out that it is, as far as we are aware, the only one in the physical-world domain that is shipping in production today. The category is real, and someone has to build it for the hard cases. We have been quietly doing that work.
What It Takes to Build One
If you accept that KaaS is the next category, the practical question for any team building software now is what it actually takes to deliver. Our experience suggests three architectural commitments that are difficult to retrofit, which is part of why most existing SaaS platforms will struggle to make the transition rather than evolve into it gracefully.
The data has to be first-class, not the substrate. KaaS systems treat the data and the relationships between data as the actual product. Most SaaS platforms treat data as the thing the features operate on. Inverting that hierarchy is not a feature change; it is a rebuild. We made this commitment at the first line of code, which is why Geodesic’s knowledge graph is the platform rather than a component of it.
Space and time have to be native, not metadata. For any KaaS platform that touches physical reality, where and when are the primary keys. They are not optional fields. Snowflake and Databricks were built around tables and rows; geography and time are columns you bolt on. Palantir’s ontology can be configured to handle them but is not architecturally native. Esri is spatially native but treats time as an afterthought. Geodesic was designed from the first line of code with where and when as equal citizens with what. This is the architectural decision that makes everything else possible at scale.
The platform has to surface the next question, not just answer the current one. This is the user-facing property of KaaS, and it is harder to deliver than it looks. It requires not just a model that can suggest follow-ups, but a knowledge graph rich enough that the suggestions are non-obvious and a reasoning layer fast enough that the suggestions arrive while the user is still in the flow. We spent five years getting all three of these aligned. We do not think they can be assembled in months.
The Question Worth Sitting With
If your software is supposed to make you ask better questions, what does that mean for how you measure it, how you price it, how you sell it, and how you build it?
It probably means usage metrics matter less than question-depth metrics. It probably means seat-based pricing is the wrong shape, because the value scales with what each user does, not how many users you have. It probably means the sales motion stops being feature demos and starts being demonstrations of how much further the customer can see. It probably means the engineering work shifts from building features to building the data layer that makes new questions answerable.
The dominant model of enterprise software is changing, and the change is being driven by something more fundamental than “add AI to your product.” SaaS was a category. KaaS is, we think, the next one. For domains organized around text, plenty of teams will reach this destination. For domains organized around the physical world, the work is harder, and the platforms that have already done the architectural work will have a head start that is difficult to close. Geodesic is one of those platforms. We think it is the right one for any team that needs KaaS-grade reasoning over the world outside the chat window. If that describes the problem you are trying to solve, we should talk.