Brex’s AI Hail Mary — With CTO James Reggio

2026-02-06 · Source: Latent Space: The AI Engineer Podcast · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, FinTech AI · Depth: Advanced, extended

Summary

Brex CTO James Reggio details the financial institution's three-pillar AI strategy: corporate AI for 10x employee workflows, operational AI for cost and compliance, and product AI to position Brex as a customer's AI solution. Brex has built an internal AI platform featuring an LLM gateway, prompt/version management, evals, and cost observability. The company emphasizes SOP-driven agents over complex reinforcement learning for operational tasks like KYC and fraud, achieving faster automation. Brex also fosters an "employees build their own AI stack" philosophy, allowing teams to choose between tools like ChatGPT, Claude, Gemini, Cursor, and Windsurf, rather than enforcing a single solution. The firm's multi-agent "network" architecture uses an executive assistant-style orchestrator coordinating specialist agents for multi-turn conversations, moving beyond one-shot tool calls. Brex has re-interviewed its entire engineering organization using agentic coding to upskill staff and measures AI adoption through second-order effects like code quality and maintainability, not just lines of code.

Key takeaway

For AI Architects and VP of Engineering/Data considering large-scale AI integration, Brex's approach highlights the value of a flexible, platform-centric strategy. Focus on building an internal AI platform that supports diverse tools and encourages operational teams to define SOP-driven agents. Prioritize robust evaluation frameworks, including LLM-as-judge for multi-turn systems, to ensure compliance and avoid regressions, rather than solely chasing vanity metrics like "% of code written by AI."

Key insights

Brex's AI strategy leverages a multi-pillar approach, prioritizing auditable, SOP-driven agents and empowering employee choice in AI tools.

Principles

• Simplicity often outperforms over-engineered solutions in operational AI.
• Empower employees to choose their AI tools for better adoption.
• AI amplifies both good and bad architectural decisions.

Method

Brex employs a multi-agent network architecture where an orchestrator agent coordinates specialist sub-agents through multi-turn conversations to complete complex tasks, enhancing system modularity and evaluation.

In practice

• Implement an LLM gateway for centralized prompt, version, and cost management.
• Break down operational problems into granular, auditable SOPs for agent automation.
• Use LLM-as-judge for multi-turn agent evaluations to assess complex interactions.

Topics

• Brex AI Strategy
• Multi-Agent Architectures
• Operational AI Automation
• AI Adoption & Fluency
• AI in Finance

Best for: AI Architect, VP of Engineering/Data, Executive, CTO, Director of AI/ML, MLOps Engineer

Related on AIssential

• AI is becoming the new enterprise interface: shaping customer discovery, shopping, service, surveillance, and internal workflows, often through platforms companies do not fully control. — AI is transitioning from a free experimental tool to a costly, regulated, and strategically critical enterprise interface.
• Advancing AI adoption in EU public administrations: Future directions and opportunities under the Apply AI Strategy — The EU's Apply AI strategy guides public administrations to strategically adopt AI for efficient, trustworthy, and people-centric services.
• Brex bets on ‘less orchestration’ as it builds an Agent Mesh for autonomous finance — Brex's Agent Mesh uses decentralized, role-specific AI agents communicating via message streams to achieve high automation.
• The AI Interview: Kaynaz Behdin — Stellantis drives AI for measurable business performance, integrating it deeply across operations with robust governance.
• The Deployment Company, Back to the 70s, Apple and Intel — Major AI developers are deploying human engineering teams to facilitate enterprise AI adoption, mirroring 1970s mainframe implementations.
• The European Union is deploying AI across strategic sectors — The EU is strategically deploying trustworthy AI across key sectors to enhance competitiveness.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Latent Space: The AI Engineer Podcast.