Brookstreet Intelligence 2.0

Investment Architecture

Access to frontier AI models is rapidly commoditised. The durable advantage is no longer which firm can access the most advanced model, but which firm has the data discipline, information architecture, and workflow design to turn those models into reliable, production‑grade systems in institutional investment settings. Brookstreet Intelligence 2.0 is built explicitly around this principle, treating models as interchangeable components inside a broader operating architecture rather than as standalone tools.

The platform combines: autonomous multi‑agent workflows, a proprietary knowledge base, structured internal data infrastructure, MCP‑enabled interoperability with internal and external systems, parallel reasoning and cross‑validation across models, autonomous orchestration of agents, and human‑led investment judgement at the decision layer. In practice, that means AI systems that do not simply answer questions, but execute sourcing, diligence, monitoring, and reporting workflows end‑to‑end—under human oversight.

• We began building this architecture early, in 2023, with some of the first AI concepts covered by Fortune’s profile of Brookstreet’s AI fund management approach.

• Brookstreet Intelligence 2.0 was formally released to GPs and LPs in April 2026 as the intelligence backbone of our Brookstreet 2.0 platform.

• Further we profiled in “Merging Agentic AI, MCP and Human‑Led Insight in Private Equity,” co‑developed with Third Bridge.

Technical Appendix

Brookstreet Intelligence 2.0 reflects a deeper shift in how artificial intelligence is engineered and deployed in production. The frontier is no longer defined by individual models simply becoming larger or more powerful, but by how multiple models and tools are composed into coordinated systems. In practical terms, this means moving away from a single AI generating outputs toward an architecture in which different agents operate in parallel—researching, reasoning, critiquing, synthesising, and executing—with conclusions refined through interaction rather than produced in a single pass.

This approach draws directly from ideas advanced by Andrej Karpathy, one of the most influential figures in modern AI. With a background spanning Stanford research, early OpenAI, and leading AI at Tesla for production‑grade autonomous driving, he sits across the three worlds that matter: top‑tier research, frontier model development, and high‑stakes real‑world execution. His framing has helped move the field away from thinking about models as standalone tools and toward thinking about them as components within larger, orchestrated systems.

Technically, intelligence is increasingly being built in layers. At the foundation sit large language models and specialised models for reasoning, retrieval, vision, and code. Above that sits an orchestration layer that decomposes tasks, assigns roles, routes outputs, and manages feedback loops between agents. One agent may be used for research, another for structured reasoning, another for critique, and another for execution. Their outputs are checked, challenged, ranked, or combined through validation steps before any final answer or action is produced. This creates a system of parallel reasoning and cross‑verification that is materially stronger than relying on a single model response in isolation.

Brookstreet extends that logic through a multi‑agent, multi‑LLM council with autonomous orchestrator and sub-agents. Different frontier models are deployed side by side, each bringing distinct strengths, training biases, and reasoning styles. The result is a distributed intelligence architecture closer to an investment committee: one agent proposes, another questions, another tests assumptions, another structures execution. Conviction is built through convergence, structured disagreement, and refinement. This is a more sophisticated and resilient way to operate than linear prompt‑response workflows.

Using emerging Model Context Protocol (MCP) standards, these agents can securely call into internal systems (knowledge base, data rooms, KPI repositories, portfolio monitoring, governance archives) and external grounded‑truth providers (such as expert‑interview and specialised data networks), under strict permissions and audit trails. Governance is built in: human decision rights and approval thresholds are explicit; every workflow is observable and auditable; role‑based access controls protect sensitive information and respect regulatory and LP requirements.

At a higher level, this aligns with Garry Kasparov’s centaur principle: human and machine working together outperform either alone. Within Brookstreet Intelligence 2.0, AI provides speed, scale, memory, pattern recognition, and continuous analytical support, while human judgment remains central to interpretation, strategic framing, and final capital‑allocation decisions. The objective is not automation for its own sake, but amplified judgment.

The implication is clear. Competitive advantage in AI is no longer defined by access to a model; that layer is rapidly commoditising. The real advantage now lies in architecture; how intelligence is orchestrated, governed, validated, and embedded into live workflows. Brookstreet Intelligence 2.0 is designed precisely at that level: not as a standalone tool, but as an integrated, multi‑agent intelligence system embedded across the full investment management lifecycle, for both GPs, LPs and PortCos.