Agents are coming to investment management.

Agents can already build personalized portfolios, rebalance accounts, generate proposals, check restrictions against client IPSs, compare client notes to cash redemption requests, propose client-specific trade ideas, find ways to improve trade execution quality, monitor portfolios against a client’s financial plan, resolve reconciliation failures, investigate operational breaks, and generate personalized reports with commentary commingling the firm’s perspective with client-specific needs.

They can do this proactively, continuously, and at the scale of every account, every position, and every trade for the world’s largest financial institutions.

But agents cannot do any of this on today’s investment management software.

Investment management software was not designed. It accumulated.

Over the last 40 years, each layer of investment management tools patched a gap left exposed by the layer before it. Every workflow became a category of software vendors, each with a handcrafted tool built to patch that one specific workflow.

Fixed income platforms patched the inability of trading systems to handle bonds. Rebalancers patched the inability of advisor workstations to optimize portfolios. UMA platforms patched the inability of rebalancers to coordinate accounts, sleeves, and managers. Direct indexing platforms patched the inability of UMA platforms to tax optimize. Spreadsheets patched everything else.

Eventually it became easy to confuse this patchwork for a system. But it is not a system. A system is something whose parts are designed to work together as a unified whole. The current stack is a patchwork of tools, each designed for one specific workflow, stitched together after the fact by the person using them.

The real system was the people who learned which tool could be trusted for a given piece of data, which information needed to be copied over from one tool to another, which report needed to be adjusted to capture holdings the tool didn’t support, which approval happened outside the tool, which exception mattered, and which spreadsheets in which shared folders filled the inevitable gaps left by all of the tools.

To do real work safely, agents need an operating system.

They need a single source of truth for the data they query, the context required to interpret it, the controls that govern what they can see and do, and the audit trail that explains what happened – with the ability for humans to safely monitor, review, and intervene as necessary.

Perhaps surprisingly, this operating system reduces to just seven primitives:

Data. The complete state of the world, normalized and tracked in real-time. Accounts, households, positions, tax lots, security master, market data, trading venues, custodians, books and records, restrictions, transactions, and post-trade booking — all integrated in one data model.

Context. The nuance required to interpret that data correctly. Account strategies, sleeves, household relationships, advisor preferences, eligibility rules, tax assumptions, and the history that connects them. Plus the memory of every previous interaction with that client, portfolio, and account.

Engines. The compute that changes the state of the system. Order management, execution management, portfolio accounting, portfolio optimization, pre-trade compliance, post-trade booking, allocation, scenario analysis, reconciliation, reporting, and best execution.

Controls. The permissions that govern what each user and each agent can see and do. Who can see what, who can do what, what requires approval, what must be blocked, and what must be escalated.

Queries. The ability to ask any question across the data model and get an answer in seconds.

Actions. The operations that change the state of the data model. Buy, sell, rebalance, allocate, approve, reject, route, cancel, book, report, explain.

Audit trail. A complete, queryable record of every action taken by every human and every agent, with the context, data, and reasoning behind it.

Every workflow in investment management — for every user, every asset class, every currency — reduces to a composition of these primitives. A UMA platform is a set of workflows. A fixed income platform is a set of workflows. A rebalancer, a reporting tool, a proposal tool, a reconciliation tool, a direct indexing platform are all workflows built on top of the same core primitives.

For 40 years, investment management software has meant one tool per workflow.

One tool for buying a bond. One tool for buying an individual stock. One tool for rebalancing a portfolio. One tool for generating proposals. One tool for conducting tax-transition analyses. One tool for explaining what happened. And so on.

These tools were the only way to deliver this work in the era they were built. That era assumed software was hard, integrations were brittle, and humans would sit between the tools and make them work together.

That era is now over, and the largest financial institutions in the world are building the future with us.