Brex’s Multi-Agent Network Replaces Dashboards with Executive Assistants

“The half-life of code has declined so significantly with agentic coding, it’s actually quite easy for anyone else to kind of try on for size a variety of different pieces of tech.” This observation by James Reggio, CTO of Brex, encapsulates the fundamental challenge and opportunity facing mature fintech companies today: how to integrate rapidly evolving AI technologies without sacrificing the stability and compliance required of a regulated financial institution. Reggio recently sat down with Swyx and Alessio Fanelli of Latent Space to detail Brex's disciplined, three-pronged strategy for navigating this transformation, focusing heavily on their internal agent platform and their counter-intuitive cultural approach to talent acquisition.

Brex’s strategy rests on three pillars: Corporate AI, Operational AI, and Product AI. Corporate AI focuses on internal workflows, aiming to 10x employee productivity across every function by leveraging off-the-shelf and custom tools. Operational AI targets the high-cost, high-risk areas inherent to finance—including KYC, underwriting, fraud detection, and dispute resolution—with the explicit goal of lowering the cost of operations while maintaining stringent regulatory standards. Finally, Product AI ensures that Brex remains essential to its clients by introducing new features that help customers justify Brex as an integral part of their own corporate AI strategy to their boards.

The most profound shift appears in the operational domain, where Brex found that simplicity trumps complexity. Rather than pursuing overly engineered reinforcement learning models, the company prioritized building agents driven by standard operating procedures (SOPs). “We realized that SOP-driven agents beat overengineered reinforcement learning in finance operations,” Reggio noted, emphasizing that breaking down complex workflows like KYC and underwriting into auditable, repeatable steps unlocked rapid automation and guaranteed the necessary compliance. This approach ensures that even as automation increases, the outputs remain fully explainable and accountable, a non-negotiable requirement in fintech. These operational agents, such as the KYC Agent and Underwriting Agent, run atop a centralized internal structure—the Brex Agent Platform—which serves as the "secret fourth pillar" enabling scalability and consistency across the entire organization.

This foundational platform includes an LLM Gateway for routing requests, a Prompt Manager for managing SOPs, a centralized Knowledge Base for understanding the business deeply, and an Evaluation Framework for rigorous testing. This internal tooling acts as the force multiplier, decoupling the core application logic from the rapidly changing landscape of large language models and frameworks. By providing this abstraction layer, Brex ensures that its engineers are not bogged down maintaining bespoke integrations but can instead focus on developing specialized agents that deliver tangible business value, whether through internal cost savings or enhanced customer experiences.

The clearest manifestation of the Product AI pillar is the Brex Assistant, a feature designed to replace traditional, cumbersome dashboards with an executive assistant (EA) model. Reggio explained that for employees using Brex, the ultimate goal is for the product to “completely disappear,” leaving only the corporate card itself and the automated assistance handling expense documentation, travel booking, and procurement policies. This assistant is not a monolithic AI but rather an orchestrator in a multi-agent network, coordinating specialist sub-agents—like the Audit Agent, Procurement Agent, and Reimbursement Agent—to complete multi-turn conversations and tasks. This distributed, hierarchical structure allows Brex to maintain specialization and accuracy across diverse financial workflows while presenting a unified, intuitive user interface.

A crucial component of this strategy is Brex’s unique approach to talent, encapsulated in their "Quitters Welcome" philosophy. Recognizing that the best builders often possess a founder's mindset, Brex intentionally seeks out individuals who have either started or plan to start companies. This approach appeals to talented builders because Brex can offer them "problems to solve that are interesting... but with instant distribution," deploying new financial AI applications to 40,000+ companies immediately. Reggio explained that they prefer to hire those with high agency and product taste, providing them with challenging problems and the resources to build solutions that have immediate, large-scale impact. This cultural decision ensures that the AI team—which Reggio described as a small, tight-knit group of roughly ten young, AI-native engineers paired with experienced staff—remains lean, ambitious, and focused on pushing production-grade agents.

Furthermore, this focus on agentic tools has subtly shifted the role of the engineer. As AI tools handle more boilerplate and repetitive coding tasks, the value metric moves away from vanity metrics like the percentage of code generated by AI. Instead, the focus is increasingly on the second-order effects: maintaining a healthy codebase, managing "slop" and "drift" introduced by generative models, and preserving deep code ownership. Reggio highlighted the importance of moving beyond traditional metrics, noting that the engineer's role now resembles that of a supervisor or mentor, guiding and structuring the work done by the AI agents. This paradigm elevates the human role to one of architectural design and strategic oversight, ensuring that the human attention—the ultimate scarce resource—is directed toward the most critical business and technical challenges.

The discussion underscored that for Brex, the AI transformation is not merely about adopting new tools but fundamentally restructuring their operations and product philosophy around agentic systems, ensuring that even in a high-stakes, regulated environment, speed and innovation remain paramount.