Your data says what happened. Your catalogue says why. The process says how.
Institutions already hold a great deal of context. The part AI and automation need most sits between them, in code and in people.
Banks and insurers are not short of context. Entity and transaction data say what happened and when. Catalogues, policies and documented lineage say why: what a term means, who owns it and which obligation it serves.
What sits between them is the how: the process that turns a rule into a reported figure, step by step. It is carried in application code, in spreadsheets and hand-offs, and in the knowledge of the people who run it. It is the one kind of context most institutions do not hold in any system.
Why the how matters for agents
A person working a regulatory process fills the gap from experience. They know which figure is net of provisions, which source takes priority and which colleague to call. An agent has none of that. Ask it for a figure and it will find a field with the right name and answer fluently, with whichever number it found first.
A bigger context window does not help. It gives the agent more candidate numbers to choose from. What the agent needs is the question the figure serves, the approved rule that produces it, and the steps in between, in order.
One governed version of the rule
Process context holds the how alongside the what and the why. xflow is built to hold each rule once, with the same version used for design, validation, evidence and any agent that references it. The rule an agent applies is then the rule the regulator was shown, and the move to agentic processes can happen at the institution's own pace.