Nothing about the discipline — everything about the tempo. An AI agent screens, structures and rebalances thousands of times a day. Each of those moments must honour a mandate written by humans and ruled on by a board.
Screening The universe is decided before the trade
An agent admitting an instrument to the permissible universe is applying the board’s screen, not its own judgement. The screen must be machine-readable and the agent must be unable to widen it by inference — “this looks close enough to what the board approved” is exactly the failure the discipline exists to prevent.
Structuring Murabaha, ijara, sukuk — step by governed step
Structure workflows have ordered, ruled-on steps. An agent may prepare and sequence them; it may not silently substitute a step, reorder what the ruling assumed, or improvise an economically-equivalent shortcut. Equivalence is a ruling, not an optimisation.
Rebalancing Held inside the screen, continuously
Markets move; holdings drift toward the edge of the screen. The agent that rebalances at 3 a.m. must be checking the live mandate at 3 a.m. — not the version cached at onboarding. Drift into non-compliance is a governance event, not a rounding error.
Finality Reserved acts wait for the board
Approving a structure, certifying a product, ruling on a contested screen — reserved acts stay advisory in the agent’s hands until a named member of the board makes them final. Speed applies to everything before that moment. Never to the moment itself.
Record The board can look back — and reproduce the answer
Every consequential act keeps its chain: which mandate applied, which screen ran, who made it final. Months later, the board and the auditor can replay the question “was this within the mandate?” and get the same answer — from the record, not from recollection.