We previously introduced the idea that AI is the new operating system for investment management firms. Not a chatbot. Not a point solution. The underlying architecture that orchestrates how the entire firm operates. The concept resonated. Now comes the harder part: What does the blueprint actually look like?
This article proposes a modern, layered architecture designed for how firms should operate going forward. It reflects lessons learned from real implementations — including some hard-won lessons about what does not work. The result is a practical framework that balances curated vendor capabilities, custom-built solutions, and self-service empowerment.
The AI Operating System Stack
The AI Operating System is not a single product. It is a stack of five interdependent layers, each with a distinct role. Think of it like a traditional OS. You need a file system, bundled applications, system services, a kernel, and a user interface. Remove any layer and the system breaks.
A key enabler of this architecture is the Model Context Protocol (MCP) — an open-source standard for connecting AI applications to external systems. MCP provides a universal way for AI agents to discover and use tools, access data, and call functions across the firm’s technology landscape. It is the connective tissue that allows the layers of this stack to communicate with each other.
Here is the stack, from foundation to user:
Layer 1: Enterprise Data Platform (The Foundation)
This is the file system of your AI Operating System. A modern Enterprise Data Platform — whether a Lakehouse architecture or a Snowflake Data Cloud — provides the governed, trusted, accessible data that everything above depends on. Capabilities like Snowflake Cortex and Snowpark add native AI and machine learning directly within the platform. Without this layer, nothing above it works reliably. As we wrote previously, your Data Strategy is your AI Strategy.
Layer 2: Curated Vendor Capabilities (Connect)
Do not recreate the wheel. Your firm likely already runs on Microsoft 365, Salesforce, Snowflake, Seismic, and FactSet. These platforms are now AI-accessible through MCP connectivity. Leverage what you already own. The goal is to curate these capabilities into your stack, not replace them. Every system you connect expands what your AI can see and do.
Layer 3: Enterprise AI Capabilities (Build)
This is where your IT team — or your consulting partner — creates competitive advantage. Custom MCP servers encode your firm’s proprietary workflows as composable building blocks that any AI agent can call. Skills and Plugins package firm-specific knowledge into reusable AI workflows — a Skill tells the AI how to accomplish a task; a Plugin bundles related Skills with the server connections they depend on. Firms often have a bias toward buying solutions. Sometimes building is unavoidable. And when it is done well, it creates capabilities competitors cannot replicate.
Layer 4: The Master Agent (AI Orchestration)
This is the heart of the operating system. Every firm should have a flagship AI experience — a single, intelligent entry point that becomes the firm’s primary AI interface. The Master Agent orchestrates everything below it: dispatching tasks to the right specialists, sequencing multi-step workflows, and synthesizing results into something more valuable than any individual piece. We will unpack this layer in detail below.
Layer 5: Self-Service & Empowerment (The Goal)
The ultimate objective. Staff across the firm create personal AI agents, explore data conversationally, generate reports on demand, and schedule AI tasks for themselves or their teams. The Skills and Plugins from Layer 3 are what make this possible — without them, users are limited to generic AI interactions. With firm-specific Skills installed, the AI operates with the firm’s institutional knowledge, not just the model’s general knowledge. AI becomes second nature. This is where the firm realizes the full promise of the AI Operating System.
MCP: The Nervous System
If the Master Agent is the brain, MCP is the nervous system. Every system in your ecosystem — Salesforce, Snowflake, SharePoint, Outlook, custom Python scripts — can be exposed as an MCP server.
Establishing MCP connectivity to your vendor solutions is like granting AI superpowers. Suddenly, data is accessible. Functions like sending an email, querying a database, or updating a CRM record are exposed as callable tools. This is a fundamental shift. AI is no longer limited to what it knows. It can act across the entire firm’s technology landscape.
The architecture we recommend treats MCP servers as satellite brains — small, specialized servers for specific domains. For vendor solutions, you connect what the vendor provides: a SharePoint MCP server, a Salesforce MCP server, a Snowflake MCP server, for example. For proprietary systems — your firm’s valuation models, internal research databases, or custom risk engines — you build dedicated MCP servers that expose those capabilities to the AI layer. Instead of loading every tool into one monolithic agent, the Master Agent uses MCP tool discovery to find the right capability on demand. You avoid the monolith. You build lightweight, reusable, composable components. Your technical team manages them independently of the agents that use them.
The Master Agent: Compound Intelligence
The Master Agent can delegate, orchestrate, and choreograph. It passes the right tasks to the right specialists, sequences multi-step workflows, and makes judgment calls about which capabilities to engage and in what order. The result is compound intelligence — insights that no individual tool or data source could produce on its own.
Consider an example. An analyst asks the Master Agent to prepare a pre-meeting brief for a client. This is not a single tool call. The Master Agent retrieves the client’s recent activity from Salesforce, pulls portfolio holdings and performance data from Snowflake, mines SharePoint for previous meeting decks and notes, and checks the compliance system for any outstanding items. Then it synthesizes all of this into a single, coherent briefing document — identifying themes, flagging risks, and suggesting talking points. No individual MCP server could produce that output. The value comes from the orchestration — the ability to weave together disparate data sources and specialist capabilities into something exponentially more valuable than the sum of its parts.
This is the key insight. The Master Agent creates compounding value. Each MCP server is a specialist. It does one thing well. But the Master Agent understands how those specialties relate to each other. It knows that a change in portfolio holdings might affect the client conversation. It knows that a recent compliance flag should inform the relationship manager’s approach. The intelligence is in the connections. The whole is not just greater than the sum of its parts — it is exponentially greater.
The Master Agent pattern is powerful for several additional reasons:
- It simplifies training by focusing user interactions around a single interface.
- It scales naturally as new MCP skills are added over time.
- It provides a consistent experience that reflects the firm’s brand and persona.
- And it offers a sense of corporate identity and pride.
It can embody the firm’s unique expertise, its workflows, and its “secret sauce.”
AI Agents represent a digital labor opportunity. It is now cost-effective to give every employee an intelligent assistant. The Master Agent makes that real.
Conclusion
The key insight of this architecture is decomposition. Break every process into reusable building blocks. Use AI to orchestrate them together. Expose them via MCP so any agent can call them. Then put it in the hands of every employee.
This is what a modern AI Operating System looks like:
- The Enterprise Data Platform provides the foundation.
- Vendor solutions are connected.
- Enterprise AI Capabilities, built with software engineering discipline, create competitive advantage.
- The Master Agent orchestrates it all into compound intelligence.
- And Self-Service puts AI in the hands of everyone across the firm.
The firms that build this stack will not just use AI. They will be powered by it.
Olmstead helps investment managers design and build their AI Operating System — from data foundation through operational transformation to client experience — guided by our AI Strategy frameworks and SPARTAH responsible AI guardrails. Contact us today to learn more.