Brex’s AI Hail Mary — With CTO James Reggio (acquired for $5B by Capital One!)
TL;DR
Brex CTO James Reggio details how the company is executing a three-pillar AI strategy—corporate adoption, operational automation, and product innovation—to transform into an agentic finance platform, achieving 5x growth with 99% burn reduction by deploying a hybrid architecture that pairs internal LLM infrastructure with a custom multi-agent framework.
🎯 Three-Pillar AI Strategy 3 insights
Corporate AI enables 10x workflow gains
Brex adopts and buys AI tooling across every business function to exponentially improve internal workflows, ensuring AI literacy is uniform across the organization rather than siloed.
Operational AI reduces institutional costs
The company builds and buys solutions specifically to lower the cost of operations as a financial institution, including automated underwriting pipelines that eliminated human intervention in KYC and customer onboarding.
Product AI embeds into customer strategies
New features are designed so Brex becomes part of customers' own corporate AI strategies, deployed to roughly 40,000 customers ranging from Fortune 100 companies to startups.
👥 Team Architecture & Culture 3 insights
Hybrid AI team pairs veterans with AI-natives
The dedicated 10-person AI team combines "AI-native" 20-year-olds with staff-level engineers who understand legacy systems, creating tight pods that navigate existing codebases while exploiting new agentic capabilities.
"Quitters welcome" founder hiring philosophy
Brex actively recruits ex-founders and future founders, celebrating when alumni leave to start companies, using the value proposition of solving interesting problems with instant distribution to 40,000 customers rather than starting from zero.
Mobile engineer to CTO transition
James Reggio broke the "backend-only CTO ceiling" by leveraging founder experience and business leadership skills over pure technical depth, representing a rare front-end-to-CTO path.
⚙️ Technical Infrastructure 3 insights
Evolution from internal gateway to Mastra
Starting with a hand-rolled LLM gateway in January 2023 for prompt management and cost monitoring, the team now uses Mastra (TypeScript) for the agentic layer while maintaining Kotlin/Elixir backends, treating tech choices as disposable due to agent-assisted coding.
Custom multi-agent orchestration framework
Brex built a proprietary system where an orchestrator agent "DMs" sub-agents (travel, policy, expense) via multi-turn natural language conversations rather than single RPC tool calls, enabling complex clarification workflows that off-the-shelf frameworks couldn't support.
MCP as imperative system interface
While sub-agents use MCP to connect to traditional systems, agent-to-agent coordination happens through conversational interfaces, functioning like an organizational chart where an executive assistant consults specialists before responding.
📈 Business Impact & Operations 2 insights
Dramatic efficiency metrics
The AI transformation contributed to 5x revenue growth alongside a 99% reduction in burn rate over 18 months, driven by internal automation and operational AI deployment.
Universal agentic coding adoption
Unlike companies where AI tools are siloed, Cursor and agentic coding are used uniformly across all 300 engineers, including engineering managers who maintain technical hands-on work as part of the "operate at all levels" value.
Bottom Line
Structure your AI transformation around three distinct pillars—internal tooling adoption, operational cost reduction, and customer-facing product innovation—while building centralized "startup-like" teams that pair AI-native talent with domain experts to deploy multi-agent systems capable of natural language coordination rather than simple API tool calls.
More from Latent Space
View all
🔬There Is No AlphaFold for Materials — AI for Materials Discovery with Heather Kulik
MIT professor Heather Kulik explains how AI discovered quantum phenomena to create 4x tougher polymers and why materials science lacks an 'AlphaFold' equivalent due to missing experimental datasets, emphasizing that domain expertise remains essential to validate AI predictions in chemistry.
Dreamer: the Agent OS for Everyone — David Singleton
David Singleton introduces Dreamer as an 'Agent OS' that combines a personal AI Sidekick with a marketplace of tools and agents, enabling both non-technical users and engineers to build, customize, and deploy AI applications through natural language while maintaining privacy through centralized, OS-level architecture.
Why Anthropic Thinks AI Should Have Its Own Computer — Felix Rieseberg of Claude Cowork/Code
Anthropic's Felix Rieseberg explains why AI agents need their own virtual computers to be effective, arguing that confining Claude to chat interfaces severely limits capability. He details how this philosophy shaped Claude Cowork and why product development is shifting from lengthy planning to rapidly building multiple prototypes simultaneously.
⚡️Monty: the ultrafast Python interpreter by Agents for Agents — Samuel Colvin, Pydantic
Samuel Colvin from Pydantic introduces Monty, a Rust-based Python interpreter designed specifically for AI agents that achieves sub-microsecond execution latency by running in-process, bridging the gap between rigid tool calling and heavy containerized sandboxes.