4. The Agent Ecosystem: A Network of Specialised, Coordinated Agents
A viable and practical agentic architecture for post-trade operations is built on a network of agents with defined, generic capabilities that operate across each functional domain of the trade processing lifecycle. Rather than building bespoke automation for each process in isolation, this approach creates reusable agent types that are configured and deployed across Statics and Account Management, Trade Allocation, Trade Control, Trade Confirmation, Settlements, and Corporate Actions
The Mailbox Agent serves as the universal intake layer, continuously monitoring communication channels across all process domains. It identifies incoming instructions, exceptions, counterparty responses, and custodian notices, and routes them to the appropriate specialist agents for processing
The Document Parsing Agent provides the cognitive translation layer. It ingests communications in any format — email body text, spreadsheet attachments, SWIFT messages, third-party platform exports, PDF notices — and converts them into structured, standardized data that downstream agents and processing systems can act upon. In the context of Trade Allocation, this means converting allocation details from whatever format a client or fund manager sends them into the standard format required for system update. In Corporate Actions, it means extracting event details from custodian bank notices with precision and without manual rekeying.
The Task Tracker AgentThe Task Tracker Agent manages proactive outreach across the lifecycle. Rather than waiting for a deadline to be missed, this agent identifies pending items — unsigned confirmations, unmatched allocations, open trade control breaks, pending settlement fails — and proactively initiates communication with internal teams and external counterparties. In the Settlements domain, it goes further, using counterparty behavioral patterns to predict fails before they occur and issuing pre-emptive reminders to mitigate them.
The System Update Agent executes decisions within defined authority parameters. It raises setup requests in internal tools and third-party platforms, updates allocations on relevant systems, resolves trade control breaks based on standard operating procedures, performs T+1 reconciliation of confirmation documents against booking values, initiates stock borrow transactions to cover settlement fails, and processes corporate action elections and cashflow setups. This is the agent that directly reduces the human processing workload.
The Orchestration Agent provides the
coordination layer across the entire ecosystem. It manages the sequencing of information
gathering, reasoning, and action across the specialist agents, determines when human
intervention is required and routes exceptions to the appropriate human authority, and
maintains the audit trail of decisions and actions across the workflow.
The agents must operate under human supervision, with checks and balances in the form of
approvals and sign-offs, with a control switch to hold back action when required. The
convenience of having all inputs to make informed decisions is a powerful enabler for
efficient processing.
