The dominant software category of the past twenty years was workflow automation: tools that took a process a human was doing manually — managing a sales pipeline, tracking support tickets, scheduling marketing campaigns — and made it faster, more consistent, and measurable. The value proposition was operational efficiency. Enterprise CRM incumbents, enterprise workflow platforms, developer collaboration platforms — these are workflow automation companies. They succeeded by replacing human memory and coordination overhead with structured data and process automation. The analogy that explains why the next wave looks different is this: workflow automation ate the process layer. AI is eating the knowledge layer. And the knowledge layer is a different beast entirely.
A workflow can be described as a decision tree. It has nodes, branches, conditions, and endpoints. You can formalise it, diagram it, and automate it. Knowledge is not like this. Knowledge is contextual, relational, tacit, and often contested. It lives in documents and conversations and relationships and professional judgment. The reason knowledge management software has historically failed — the enterprise intranet graveyard, the wiki that nobody updates, the knowledge base with 40% stale articles — is that the software tried to apply workflow logic to a fundamentally different kind of information object. Knowledge doesn't flow through a process; it gets accumulated, challenged, refined, and transmitted through human cognition in ways that don't reduce to nodes and branches.
What changes with AI is not the nature of knowledge, but our ability to build tools that interact with it at the right level of abstraction. Retrieval-augmented generation allows a knowledge platform to surface relevant content without requiring the user to know how the content is organised. Semantic similarity models allow knowledge to be connected across conceptual proximity rather than taxonomic hierarchy. These are not incremental improvements to search; they're a different interaction model with the knowledge layer. For education, this matters specifically because the knowledge that needs to be managed isn't just institutional documentation — it's student learning, curriculum logic, assessment evidence, teacher professional knowledge, and the accumulated pedagogical experience of a school community. Building platforms that can hold all of that and make it actionable is the infrastructure challenge of the next decade in education technology.
We're not saying SaaS is over. Workflow automation still has decades of growth left in verticals that remain deeply manual. What we are saying is that the companies with the highest durable value over the next decade will be the ones that figure out the knowledge layer — how to store it, transmit it, verify it, and build trust around it. In education, that means curriculum knowledge graphs, verifiable student learning records, portable assessment evidence, and the institutional intelligence that currently lives in a school's staff room and retires with its senior teachers. The dominant knowledge infrastructure platform for learning will be a company that makes this as obvious a purchase as CRM was for a growing sales team in 2005. That company is not built yet. We think it will be built from Australasia.