Present day enterprise systems have been built and designed under the assumption that humans sit at the centre of every decision. That is about to change.
For years, enterprise architecture has revolved around two layers: Systems of Record and Systems of Engagement. These layers defined the modern enterprise architecture. They were successful because they matched how work used to happen: humans making decisions, systems supporting them.
This way of working is reaching its limit. As AI agents begin to execute work across systems, a gap is emerging that neither of these layers were designed to solve.
From identity infrastructure to understanding enterprise behavior
My own background has been in building infrastructure for identity and trust at scale.
At my former venture, ForgeRock, which raised 275 million dollars and reached a 2.8 billion dollar valuation at IPO on the NYSE (FORG), we focused on securing and managing identity across complex digital ecosystems. That experience made one thing very clear: identity is never isolated. It requires context to be meaningful. Who is acting, what they are trying to do, and how that relates to everything else happening around them is what ultimately determines whether they can be trusted.
My focus has since shifted to how enterprise systems behave when AI agents, not humans, are doing the work. That work has led to IndyKite.
Where the current stack falls short
Enterprises already have most of the foundation required for AI-driven execution: large volumes of data, capable models, and a growing set of orchestration and automation systems.
Problems emerge when work spans multiple systems, for example when an identity event triggers a workflow in a CRM, which then triggers billing, access control, and downstream actions. At each step, instead of context being preserved, it is reinterpreted or reduced to only system specific representations. Identity is re-evaluated in each system, relationships are implicit, and the original intent is no longer visible in downstream execution. Trust is reconstructed later by piecing together system logs and audit trails, not understood at the moment the action is taken.
This is manageable when humans are in the loop, because people can compensate for what systems fail to capture, interpreting ambiguity, reconstructing intent, and filling in missing pieces. But as systems increasingly act independently, that limitation becomes a bottleneck.
The shift to the agentic enterprise
We are rapidly moving toward the era of the agentic enterprise, where AI agents will not be confined to a single platform or application. They will operate across systems, trigger workflows, combine signals, and take actions that previously required coordination between multiple tools and teams. This changes what enterprise architecture needs to provide.
Many enterprise systems involved in workflows, automation, and decisioning are increasingly part of execution flows, triggering actions and passing control across boundaries. This requires enterprise architecture that can ensure execution remains understandable and governable across systems.
Without a layer that maintains that continuity, execution becomes fragmented, difficult to interpret, and hard to control at scale.
The new context layer
Jaya Gupta from Foundation Capital is leading much of the thought leadership around this missing layer for AI agents. In a recent interview, Jaya noted the significance of ‘context graphs’ in capturing decision traceability and the role it can play in decisioning and governing agentic systems.
“Agents can read data and take action, but they still don’t know why decisions get made. That reasoning is what we call a decision trace. Those decision traces are scattered across tools, buried in Slack, and sometimes they don’t get recorded at all.
The winners of the future will be the companies that can capture those decision traces and turn them into context graphs.”
What makes this especially powerful is that these decision traces extend their value beyond the moment of execution. When captured and connected over time, they create a feedback loop that strengthens the system’s understanding of how work unfolds. Decisions become part of a growing body of operational knowledge, revealing why actions were taken, how outcomes developed, and how similar situations can be addressed in the future.
Over time, context graphs turn this accumulated decision history into a foundation for continuous learning across agent workflows. Traceability evolves into an operational asset that informs future actions, enabling AI systems to draw on precedent, context, and experience as they operate.
IndyKite and the System of Intelligence
This is the space we are focused on at IndyKite. We call it a System of Intelligence.
At its core is a live context graph that continuously captures and connects identity, relationships, data signals, decisions and actions across systems. This real time, contextual understanding becomes the basis for control.
Decisions, actions, and outcomes are continuously captured and connected, creating a persistent understanding of how the enterprise operates. This enables patterns to emerge across workflows, decisions to be evaluated against outcomes, and agents l to make better choices based both current context, and accumulated experience.
It allows real-time decisions to be made with full awareness of context, enables policies to be applied consistently, and gives AI systems the ability to operate across systems without losing track of previous actions, identity or intent.
It is a live intelligence layer that reflects how the enterprise is actually operating at runtime, forming the foundation for reliable and continuously improving agent-driven execution across enterprise systems
Three architectural foundations
To make this work, three architectural ideas come together. A context graph models how entities and relationships evolve over time. A data mesh ensures that data remains distributed and owned at the source while still being usable across the organization. Identity acts as the common layer that ties actors, whether human or machine, to actions and intent.
Individually, these concepts are well understood. Together, they form a foundation for reasoning about enterprise activity in real time.
Getting AI agents enterprise production ready
The era of the agentic enterprise is already here. AI agents are increasingly shifting from supporting work to executing it, creating a control, trust and governance crisis. I strongly believe the only way to solve this is to leverage unified context for real-time execution.
Without context and real-time control, automation becomes fragmented and hard to govern. With it, enterprises gain a clear, shared understanding of identity, intent, data, and can effectively enforce action as work happens.
This is the role of the System of Intelligence, a runtime layer that connects existing systems and preserves meaning as execution flows across them. It enables agents to act with awareness and enterprises to maintain control, while continuously learning from how decisions are made and executed across the organization.
Learn more about The IndyKite Platform here.
