How we're building AI-native operations at Brex

What if we could start from scratch?

A few months ago, our CEO Pedro Franceschi asked me a deceptively simple question: "If we were starting Brex today, how would we build?" Not how would we improve, but how would we build from scratch in an era where AI is real, accessible, and rapidly evolving?

The answer crystallized something we'd already been working toward: we'd build everything AI-native. AI would be a core component of every Brex product, operational decision, and workflow.

And this is what we’re doubling down on today. We're rapidly rebuilding our foundation with AI at the center — creating systems that get smarter, faster, and more precise with every interaction. And unlocking exponential improvements in speed, quality, and consistency.

Three AI pillars, one foundation

Operational AI is the focus of this article, but it’s just one pillar of our AI strategy. At Brex, AI is not a collection of tools but an integrated approach that touches every part of the business across three key pillars:

First, we build AI into every Brex product to empower customers to spend smarter and move faster. Internally, we also build and buy AI capabilities to automate, augment, and accelerate how we operate. I may be biased, but I think this is where the real leverage lives. And, lastly, we purchase AI tools to multiply every employee’s productivity. Think Glean for knowledge search, Cursor for engineering, and Claude for writing.

Today, I will dive deeper into operational AI, detailing how we're building an AI-native operations org and why this shift is essential for any business leader planning for the next decade.

Building AI-native operations

What do we mean by “AI-native operations?” Brian Klaja, Senior Director of Operations Enablement here at Brex, explains it with a great analogy: If traditional operations is like managing a fleet of taxis, AI-native operations is a network of self-driving vehicles.

With taxis, every trip needs a driver, every route requires navigation, and maintenance happens one car at a time. On the other hand, self-driving cars handle the vast majority of journeys autonomously. They learn from road conditions and self-diagnose issues.

Essentially, we’re building an operations org where AI is the driving force behind the entire design, functionality, and decision-making processes. Here’s what that looks like in practice:

• AI handles the first attempt at most work, such as customer cases, approvals, and transaction reviews.
• Humans focus on exceptions, using those edge cases to retrain and refine the system.
• Compliance and operations experts become workflow designers and policy engineers.
• We build systems that improve and compound over time.

Our journey to AI-native operations is guided by the following core principles we have developed:

1. Customer-first: AI must be neutral or positive for the customer experience, empowering self-service and enabling instant resolutions (e.g., chatbot problem solving, instant approvals/credits).
2. AI-fluent talent: Non-engineering roles must have analytical skills for this plan to work.
3. Governance and explainability: All systems must be auditable, traceable, and transparent.
4. Platform over point solutions: Build iteratively for multiple tasks at a time to enable reusability.
5. Adoption by design: Automation tools need to be accessible to and manageable by non-technical employees, requiring analytical skills but not deep coding expertise.
6. Build vs. buy discipline: Sometimes buying makes more sense than building. Align decisions to the core principles.

Building an AI-native workforce

This is about evolving roles, not eliminating them. This shift necessitates an investment in new skills, demanding that talent be more technically and analytically proficient. We’re refocusing human work on problem-solving, critical thinking, ethical judgment, managing AI systems, and collaborating with AI as a partner. As such, our AI-native operational structure will have three levels of work:

Level 1: AI-automated execution — Fully automated, no human involvement needed for standard cases.

Level 2: AI-augmented teams — Handle exceptions from Level 1, perform evaluations, and continuously retrain the system. These are prompt engineers and workflow analysts who understand both the business and the technology.

Level 3: Policy and process design — Senior domain experts who ensure our AI systems operate within the right frameworks, handle escalations, and translate compliance requirements into machine-enforceable rules.

Continuing the self-driving car analogy from above, employees will become sophisticated fleet managers and system engineers. They will optimize the fleet's overall performance, improve the AI's driving algorithms based on new data and rare incidents, and ensure the passengers — our customers — have a smooth experience.

Where it’s already working and compounding

Our AI-native transformation is already live across several key areas of our operations:

• KYC and AML: Identity verification, document checks, sanctions screening, and risk assessments are increasingly automated with AI. Our KYC agent already handles the end-to-end application review.
• Customer experience: Our service agent (powered by Brex customer Sierra AI) and other agent-assist tools are improving containment rates and resolution quality, enabling instant resolutions without transfers. Customers now get their responses 90% faster, saving them 15,000 hours per year.
• Implementation: New customers onboard faster with AI-guided configuration flows and intelligent document processing, dramatically reducing manual lift.
• Disputes: Intake, evidence gathering, outcome prediction, and resolution drafting all happen more efficiently with agent assistance.
• Fraud and compliance: Agentic AI is leveraging behavioral biometrics and graph analytics to detect and prevent complex fraud scenarios before they escalate.

Each domain is being transformed in parallel, built on a shared platform so the benefits compound across the company. Every improvement cascades through the organization. When our KYC system gets faster, sales closes deals quicker. When fraud detection improves, underwriting gets better data. When customer support becomes more efficient, we learn which product features need work.

This creates what we call the AI-native operations flywheel: tighter feedback loops, better data, stronger models, improved operations. The efficiency gains are just the beginning. The real win is building an organization that gets exponentially better over time.

Why this shift is essential now

First, the technology has caught up to the hype and expectations. AI is real, fast-moving, and capable of doing work we used to reserve for humans. Customer expectations are also rising. Speed and consistency are no longer tradeoffs, and AI enables us to deliver both.

As Brex continues to grow, AI is also critical to scaling our systems to achieve and accelerate our next chapter of growth. And last but not least, AI is a path to profitability. We're targeting 5% cost-to-serve by FY28, and operational AI is how we'll get there.

Building for the long term

If you still think AI is just about efficiency, you're probably thinking too small.

This isn't about doing the same work faster. It's about doing better work and building AI-native systems that scale intelligently, adapt rapidly, and enable our teams to spend time on what truly matters.

We're not automating for its own sake. We're redesigning operations so Brex can serve more customers, with higher precision, and unlock greater leverage without compromising on trust, control, or experience.

This transformation requires leadership commitment, cross-functional collaboration, and the willingness to rebuild processes from scratch rather than automate what already exists. It's a cultural shift as much as a technical one.

I believe it will be one of the defining transformations of our next decade. And I encourage other business leaders to rethink your operations in the age of AI — and build it right, from the ground up.