From AI Tools to ‘Intelition’: How Shared World Models Will Reshape Software

Enterprise AI is moving beyond the notion of “smart tools” that people invoke on demand. A new paradigm is emerging in which human and machine intelligence work within a shared understanding of the business, continuously shaping decisions and actions together. The article this analysis draws on proposes a term for that shift: intelition.

Intelition describes what happens when people and AI agents co-perceive, co-decide, co-create and co-act in the same living model of an enterprise and its environment. Instead of a user calling an API or opening a chat window, AI becomes an ever-present collaborator embedded in the organization’s ontology, its world model and each individual’s personal interface.

For enterprise technology leaders, this has architectural consequences. Intelition is not a single feature or product category; it’s an organizing principle that ties together unified ontologies, durable world models and human-centric interfaces. The convergence of these three forces is already visible in the market through the strategies and research agendas of players like Palantir, Google, Meta, Apple and Tim Berners-Lee’s Solid ecosystem.

This article unpacks those forces and what they imply for how you’ll design systems, govern data and think about “applications” in the next software era.

From ‘Using AI’ to Continuous Co‑Production

Most current enterprise AI patterns still treat models as external services: you send a prompt, they send back an answer; you add a “human in the loop” step to an agent workflow. This framing implicitly positions the human as a user standing outside the system, occasionally reaching in to “use” AI.

Intelition inverts that relationship. Instead of discrete invocations, humans and agents continuously co-produce the state of the system: the data, the logic, the plans and the actions. The boundary between “the app,” “the AI” and “the user” starts to dissolve into a single shared model of work.

Practically, that means:

• AI agents operate on the same objects, policies and events that people do, rather than on ad hoc text prompts alone.
• Decisions and workflows are shaped in real time by both human judgment and machine reasoning.
• The system learns not just from historical data but from this ongoing joint activity.

To support this, enterprises need three things working together: a unified ontology that gives humans and agents a common language of the business, world models and learning mechanisms that give AI durable understanding over time and a personal interface that puts individuals in control of how their intent and data participate in this shared system.

Why a Unified Enterprise Ontology Matters

Palantir CEO Alex Karp recently argued that “all the value in the market is going to go to chips and what we call ontology,” framing this shift as “only the beginning of something much larger and more significant.” In this context, ontology is not an academic abstraction; it is the explicit model of an enterprise’s objects and their relationships—customers, policies, assets, events—and the permissions and actions that connect them.

Palantir extends this into what it calls a “kinetic layer,” which encodes not only what exists but what can be done, by whom, under which constraints. In an intelition-centric architecture, that kinetic layer is where AI agents and humans share the same verbs: approve, dispatch, escalate, reconcile, notify and so on.

However, most enterprises are far from this state. The SaaS era left organizations with fragmented object and process models. Every application ships with its own schema and workflows; legacy systems add more divergence and gaps. Despite years of investment in data warehouses and lakes, few organizations have achieved a truly consolidated enterprise ontology.

From an AI architecture standpoint, this fragmentation is more than an integration headache. Agentic systems that must reason and act across functions—linking suppliers to operations, regulators to compliance teams, customers to case management—need a shared, machine-operable view of the enterprise. Without it, AI remains confined to narrow use cases inside single applications.

As organizations link and federate ontologies across internal and external domains, a new software paradigm emerges. AI agents can be authorized to traverse and act across that federated graph in ways that mirror how people think about the business, not how individual systems were historically designed.

Ontologies as the Foundation for Agentic AI in the Enterprise

Agentic AI tools—systems that can plan, call functions, orchestrate workflows and act with some autonomy—are already appearing in many enterprises. But their scope is usually bounded by a single vendor’s tools or a single application’s API surface.

By contrast, when built on top of a unified or federated ontology, agentic systems can:

• Resolve references across systems—recognizing that a “customer account” in one application maps to a “policyholder” in another and an “asset owner” in a third.
• Apply consistent business logic—such as eligibility rules, risk thresholds or compliance policies—wherever relevant objects appear in the graph.
• Respect security and permissions at the ontology level, not just at individual application endpoints.

Karp describes the goal as “tether[ing] the power of artificial intelligence to objects and relationships in the real world.” That tether is the ontology: it anchors large language models and other AI components to structured representations of the business and its environment.

For technology leaders, prioritizing ontology work—often seen as tedious modeling or integration plumbing—becomes strategically important. It is a prerequisite for moving from isolated AI pilots to intelition: a state where AI and humans operate in a shared understanding that spans suppliers, regulators, customers and internal operations.

World Models, Durable Memory and the Limits of Today’s LLMs

Even with a robust ontology, today’s AI systems have a key limitation: they are excellent at handling context within a session, but they do not inherently accumulate understanding over time. Each major retraining is effectively a reset, not a continuation of a single evolving worldview.

Continual learning requires durable memory: mechanisms by which models can update and refine their internal representations without starting over. Research efforts are now probing how to add this capability while remaining grounded in existing large language model (LLM) architectures.

Google, for example, has announced work on “Nested Learning” as a potential approach. While the researchers do not claim to have solved the broader challenge of building world models, they position Nested Learning as a way to embed continual learning and more durable memory directly into the training and operation of LLMs. The stated ambition is that this kind of architecture could eventually make periodic wholesale retraining obsolete.

In parallel, Meta’s chief AI scientist Yann LeCun has been advancing a different but related thread. In 2022 he published a blueprint for “autonomous machine intelligence,” proposing a hierarchical joint-embedding architecture (H-JEPA) that uses world models to make predictions. His critique is blunt: “LLMs are good at manipulating language, but not at thinking.”

Since then, Meta has moved parts of this blueprint into practice with open source models V-JEPA and I-JEPA, designed to learn representations from images and video. These systems aim to learn how the world behaves over time from raw sensory data, not just how words tend to co-occur in text corpora.

For enterprises, the immediate takeaway is not that one research direction has “won,” but that world models and continual learning are becoming central to the AI roadmap. Intelition ultimately depends on agents that can retain and refine a model of the enterprise and its environment, informed by ongoing interaction, not just static training data.

How Emerging World‑Model Research Connects to Intelition

World-model research and ontology work reinforce each other in an intelition architecture.

Ontologies provide explicit structure: named entities, relationships, events, permissions and actions. World models—whether built via approaches like Nested Learning or JEPA-style architectures—provide implicit structure: statistical expectations about how those entities and events evolve over time, what typically follows what and which interventions change outcomes.

When you combine the two, several capabilities become possible:

• Richer forecasting and simulation: Agents can use world models to estimate how changes to objects and relationships in the ontology (such as policy changes or supply disruptions) are likely to play out.
• Continuous adaptation: Durable memory allows agents to refine their expectations as new data arrives, without discarding prior learning. That is key for volatile domains like operations, risk and customer behavior.
• Shared situational awareness: Humans and agents can reason over the same evolving picture of the enterprise, expressed both in the ontology’s graph and in the patterns learned by world models.

However, the underlying research is still in motion. The Google work explicitly notes that Nested Learning does not by itself solve world modeling. LeCun’s critique of LLMs underscores that current systems, despite their capabilities, fall short of robust “thinking” about the physical and social world.

Enterprise leaders should therefore plan for architectures that can incorporate these advances as they mature, rather than assuming today’s prompt-driven LLM usage patterns will remain the dominant model. Intelition implies a long-term trajectory toward systems that accumulate and operationalize a history of joint human–machine activity.

The Personal Intelition Interface: Beyond ‘User’ and ‘App’

The third force shaping intelition is the personal interface—the way individuals inhabit this shared world of humans and agents. In this paradigm, people are not peripheral “users” of isolated apps and AI endpoints; their primary way of participating in work and life is through an always-on, context-aware interface that can act on their behalf.

This is not just another productivity application. It is envisioned as the main surface through which a person expresses intent, sets goals and mediates actions across a federated digital economy.

Recent moves in the market signal how seriously the industry is taking this idea. In May, Jony Ive sold his AI device company, io, to OpenAI with the explicit goal of accelerating a new AI device category. Ive framed his motivation not only in terms of innovation but also responsibility, acknowledging that past innovations have had both “wonderful” and “harmful” unintended consequences and emphasizing a determination to “try and be useful.” The implication is that getting personal intelligence devices right is about more than first-mover advantage; it is about shaping how individuals experience AI in their daily lives.

In parallel, Apple is exploring architectures specifically aimed at understanding “user intent” on-device, with lower compute requirements and latency than cloud-centric LLM approaches. Its work on UI-JEPA is one example: an approach that focuses on analyzing the user’s on-device activity to infer what they are trying to do, rather than sending everything to centralized servers for modeling.

In an intelition context, a personal interface backed by such on-device intelligence would serve as the individual’s agent in the broader enterprise and data ecosystem—negotiating permissions, orchestrating tasks and filtering what information and actions are appropriate in a given context.

On‑Device Intent and the Rebalancing of Data Power

Moving user-intent understanding onto the device has broader implications than performance and battery life. It challenges the prevailing model of the digital economy, where centralized entities profile users and monetize their intent and behavior at scale.

Tim Berners-Lee, the inventor of the World Wide Web, has criticized this dynamic directly: “The user has been reduced to a consumable product for the advertiser … there’s still time to build machines that work for humans, and not the other way around.”

His work on the Solid standard, begun in 2022, aims to do exactly that. Solid is a specification for secure personal data management that gives individuals control over where their data is stored and how it is shared. It is explicitly well-suited to pair with new personal AI devices that must respect user autonomy while still participating in rich agentic workflows.

One concrete example is Inrupt—a company founded by Berners-Lee—combining Solid with Anthropic’s Model Context Protocol (MCP) to create “Agentic Wallets.” In this pattern, personal control over data is not a cosmetic feature; it is the architectural safeguard as systems gain the ability to learn and act continuously on behalf of the user.

For enterprises, this points toward a future in which interacting with customers, employees and partners increasingly involves negotiating with their personal agents, which operate under user-controlled data and intent policies. Designing systems that can respect and leverage this shift will be an important part of any intelition-aware strategy.

Strategic Takeaways for Enterprise Technology Leaders

The underlying message of the intelition framing is that the next software era is not hypothetical—it is already taking shape. Three forces are simultaneously advancing and beginning to converge:

• Enterprise ontologies supply the nouns and verbs: the objects, relationships, actions and permissions that define how your business operates.
• World-model and continual-learning research aims to give AI durable memory and a capacity to update its understanding over time, beyond one-off retrainings.
• Personal interfaces and on-device intent understanding put individuals at the center, with agents that work for them and mediate how they participate in larger systems.

Together, these form the substrate for intelition—continuous co-production between humans and AI agents in a shared model of reality.

In practical terms, enterprise technology leaders and data decision-makers can start preparing by:

• Elevating ontology and data modeling from a back-office concern to a first-class strategic capability, especially where cross-functional agentic workflows are envisioned.
• Tracking and experimenting with emerging approaches to durable memory and continual learning, with an eye toward how they might integrate with your enterprise graph and event streams.
• Exploring how personal agents and secure personal data standards, such as Solid, will change customer and employee interaction patterns, and what governance frameworks you will need.

Most importantly, treat AI not as a set of tools you bolt onto existing systems, but as a co-evolving participant in the architecture of your enterprise. That is the essence of intelition—and the direction in which the software ecosystem is already moving.