Agent + Human Ops:

How AI is Changing Roles and Workflows at Brex

Overview

Brex CTO James Reggio spends a lot of time thinking about how AI can make Brex disappear. “Nobody wakes up and says ‘I can’t wait to spend time in Brex,’” he says.
“We believe the best user interface is just the card that lets you swipe and go. Most of our investments on the product AI side are oriented around the thinking of eliminating Brex from your overall experience. The best AI is the AI that you don’t notice because it’s taken work away.”

Camilla Matias, Brex’s COO, shares this thinking, except she’s looking inward at the company’s operations — using AI to reduce the toil of the most manual, prescriptive work. “If I was starting the company today in this new AI era, I’d build operations completely different,” says Matias.

AI Pillars at Brex

Brex has three AI pillars:

• Product AI, the customer-facing AI features.
• Operational AI, the internal use of AI to scale the business.
• A corporate AI workstream, the AI tools employees use to do their day-to-day work.

“There’s actually a hidden fourth pillar, which is an AI platform that serves both our product and operational pillars,” says Reggio.

Treat Internal Operations with External Rigor

To enable more internal AI adoption, Reggio identified two early hurdles:

1. Shortening the time from idea to execution.
2. Eliminating legal, procurement, and onboarding barriers.

To solve the first challenge, he built an internal platform in Retool fashioned from the product’s external AI features, where employees could build, test and deploy agents.

This “agent platform” now has:

• A prompt management system that allows users to refine implementations.
• Multi-model testing and evaluation frameworks.
• API integrations for automated workflows.

Redesign Roles for Agent, Not Task, Management

Matias believes in moving quickly and not being constrained by traditional roles. “Before AI, if you wanted extremely efficient operations, you needed very specific teams dealing with specific client segments. But that’s constrained by how many groups you have and how you enable connection between them. With AI, that’s not a bottleneck anymore.”

Workflows Designed for AI

Instead of fitting AI into human workflows, Brex finds workflows where AI excels.

• Time consumption—targeting high-volume, time-intensive processes.
• AI’s advantage over humans—assessing how much better LLMs perform tasks.
• Implementation speed—seeking quick wins.

Conclusion

As Brex transforms its internal operations, the future seems promising for efficiency gains. “I think we can operate 5x to 10x more efficiently,” says Matias, reflecting on AI’s potential to reshape workflows.