Fintech Breaks is betting that the future of enterprise AI isn’t better orchestration – it’s less than that.
As generic AI agents move from co-pilots to autonomous systems, Brax CTO James Rezio says traditional agent orchestration frameworks are becoming a hindrance rather than an enabler. Instead of relying on a central coordinator or rigid workflow, Brex has created what it calls an “agent mesh”: a network of narrow, role-specific agents that communicate in plain language and work independently – but with full visibility.
“Our goal is to use AI to effectively make breaks disappear,” Rezio told VentureBeat. “We are aiming for full automation.”
Breaux learned that for its purposes, agents needed to work in narrow, specific roles to be more modular, flexible, and auditable.
Rezio said the architectural goal is to enable every manager in an enterprise to have “one point of contact within Brex that is handling the totality of their responsibilities, whether it’s expense management, requesting travel, or approving expense limit requests.”
Travel with brakes assistant
The financial services industry has long adopted AI and machine learning to handle the massive amounts of data it processes. But when it comes to bringing AI models and agents, Industry adopted a more cautious path In the beginning. Now, There Are More Financial Services Companies Including Brex AI-powered platform launched And multiple agent workflows,
Brex’s first foray into generic AI was with its Brex Assistant, released in 2023, which helped customers automate certain finance and expense tasks. It provides suggestions for completing expenses, automatically fills in information, and monitors expenses that violate policies.
Reggio admits that the brakes assistant works, but it’s not enough. “I think to some extent, it remains a technology where we don’t fully know its limitations," He said. "There need to be a lot of patterns around this that are being developed by the industry as the technology matures and more companies are building with it."
Brex Assistant uses multiple models, including Anthropic’s cloud and custom Brex-models, as well as OpenAI’s API. The Assistant automates some tasks but is still limited in how low-touch it can be.
Rezio said Brax Assistant still plays a big role in the company’s autonomy journey, primarily because its agent flows into the Mesh product application.
AGent Mesh to change the orchestration
The consensus in the industry is that multi-agent ecosystems, in which agents communicate to accomplish tasks, require an orchestration framework to guide them.
Reggio, on the other hand, has a different approach. "The deterministic orchestration infrastructure … was a solution to the problems we saw two years ago, which was that agents, like models, hallucinate a lot,” Reggio said. “They’re not very good with a lot of tools, so you need to give them these degrees of freedom, but in a more structured, rigid system. “But as the models get better, I think it’s starting to cap the range of possibilities that are expanding.”
More traditional agent orchestration architectures focus on either a single agent that does everything or, more commonly, a coordinator/orchestrator plus tool agents that explicitly define the workflow. Reggio said both frameworks are more rigorous and solve problems more commonly seen in traditional software than AI.
Rezio argues that the difference is structural:
- traditional orchestra: predefined workflow, central coordinator, deterministic path
-
agent trap: event-driven, role-specialized agents, message-based coordination
Agent mesh relies on linking together a network of many smaller agents, each specializing in a single task. Agents, once again using a hybrid mix of models like the Brex Assistant, communicate “in plain English” with other agents over a shared message stream. A routing model quickly determines which tools to implement, he said.
A single reimbursement request triggers multiple actions: a compliance check to align with expense policies, budget verification, receipt matching and then payment initiation. While an agent can certainly be coded to do all of this, this method is “brittle and error-prone”, and it responds to new information shared through the message stream anyway.
Rezio said the idea is to clarify all those different tasks and delegate them to smaller agents instead. He compared the architecture to a Wi-Fi mesh, where no single node controls the system – reliability emerges from many small, overlapping contributors.
“We found a really good fit with the idea of basically incorporating specific roles as agents on top of the best platform to manage specific responsibilities, like how you might delegate accounts payable to one team versus expense management to another team,” Rezio said.
Brex defines three main ideas in the agent mesh architecture:
- config, where the definitions of agents, models, tools, and subscriptions reside
-
MessageStream, a log of each message, tool call, and state change
-
Clock, which ensures deterministic order
Brex also built evaluations into the system, in which the LLM acts as a judge, and an audit agent reviews each agent’s decisions to ensure they adhere to accuracy and behavior policies.
success so far
Brax says it has achieved substantial adoption among customers in its AI ecosystem. Brex did not provide third-party benchmarks or customer-specific data to validate those benefits.
But Rezio said enterprise customers using Brex Assistant and the company’s machine learning systems “are able to achieve 99% automation, especially for those customers who are really leaning into AI.”
This is a notable improvement from the 60 to 70% of Brex customers who were able to automate their expense processes prior to the launch of Brex Assistant.
The company is still in the early stages of its autonomy journey, Rezio said. But if the agent mesh approach works, the most successful outcome may be invisible: Employees are no longer thinking about expenses at all.
<a href