Jan 17, 2026 · 1h 13m · latent-space
Brex’s AI Hail Mary — With CTO James Reggio (acquired for $5B by Capital One!)
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Brex CTO James Reggio details the fintech company's three-pillar AI strategy—Corporate, Operational, and Product AI—highlighting their multi-agent orchestration architecture, internal developer tooling, and founder-driven engineering culture.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 16.2% of the talking time here. How this is scored →
The hosts as informed peer 5.9 Guest teaching 6.2 Guest disagreement 1.5 The hosts pushing back 2.4
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 14:58 Defending against LangChain superiority
James pushes back on Swyx's defense of LangChain, maintaining that early framework shortcomings forced Brex to build proprietary tooling and switch to Mastra.
Hardest push from the hosts ▶ 32:53 Challenging AI code reviewer claimsSwyx firmly rejects the notion that companies can solve code slop by simply layering more automated AI reviewers onto AI-generated code.
Biggest teaching moment ▶ 40:26 RL failure versus simple SOP agentic workflowsJames educates the hosts on how their expensive bet on reinforcement learning failed compared to simple web research agents executing granular SOP prompts.
The host holds their own ▶ 56:00 Suggesting saturation tracking through failing evalsAlessio leverages his experience with Baris AI to introduce a novel paradigm for tracking model capability progression, which the guest enthusiastically adopts.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Hiring Ex-Founders and Brex's 'Quitters Welcome' Philosophy | 6 | 5 | 2 | 4 | Swyx and Alessio challenge the trendy practice of hiring ex-founders, questioning whether founder tenure is too short or an anti-signal. James acknowledges their concern but counters with Brex's deliberate 'Quitters Welcome' philosophy and how instant distribution attracts top talent. | |
| Engineering Organization and the Dedicated AI Center of Excellence | 5 | 5 | 1 | 2 | Alessio asks about engineering organization structure and AI adoption disparities across teams. James explains their product domain divisions and the deliberate creation of a centralized 10-person AI center of excellence designed like an internal disruptor startup. | |
| Pod Composition and Managing Internal AI Perceptions | 5 | 6 | 2 | 3 | Alessio raises the common cultural problem of non-AI engineers feeling alienated or unvalued. James reframes this, explaining that because Brex engineering heavily rewards revenue impact, core product engineers remain proud of driving direct business metrics over speculative AI projects. | |
| Brex Agent Platform Architecture, TypeScript, and Mastra | 6 | 6 | 3 | 3 | Swyx presses on Brex's choice of Mastra over established tooling like LangChain. James defends the decision by noting ergonomics and historical limitations of early LangChain before explaining their greenfield TypeScript stack. | |
| Multi-Agent Architecture and the Brex Assistant | 6 | 7 | 1 | 2 | James gives an in-depth breakdown of Brex's multi-agent architecture for the Brex Assistant, explaining how single-agent tool overloading failed and why encapsulated, inter-agent direct messaging modeled after human org charts worked better. | |
| Model Context Protocol vs. Multi-Turn Agent Conversations | 7 | 6 | 2 | 3 | Alessio and James discuss whether Anthropic's Model Context Protocol (MCP) should govern agent-to-agent communication. James clarifies that MCP suits single-turn imperative tool calls, whereas multi-agent collaboration requires conversational, multi-turn dialogue. | |
| Deep Dive into Brex's Three AI Strategic Pillars | 5 | 7 | 1 | 1 | James details Brex's three AI strategic pillars (corporate, operational, product). He explains how operational AI directly slashed costs in compliance and underwriting while transforming support staff into prompt and eval authors. | |
| ConductorOne Self-Service Tooling and Model Flexibility | 6 | 6 | 1 | 1 | Swyx expresses enthusiasm for Brex's ConductorOne Slack-based model provisioning. James explains their philosophy of avoiding vendor lock-in by letting employees choose models dynamically and using usage metrics in enterprise renewals. | |
| Managing Code Slop, Review Rigor, and Shared Understanding | 7 | 5 | 2 | 4 | Swyx challenges the idea that AI review tools alone can fix code slop, arguing human ownership must scale. James agrees with the downside of codebase drift and unrigorous reviews while Swyx pushes back on tool vendors claiming automated cures. | |
| The Evolution of Engineering Craft and Junior Workflows | 6 | 6 | 1 | 2 | James shares findings from a college dinner where new grads used LLMs for design docs and architecture rather than blind code generation. Alessio and James discuss how junior vs senior workflows differ in design and schema specification. | |
| Code Consistency Rules, Greptile, and Semantic Review in CI | 7 | 5 | 1 | 2 | Alessio brings up historical parallels like Danger Systems for semantic linting in CI. James validates this by detailing their adoption of Greptile and automated Claude Code review checks in GitHub Actions. | |
| Operational Lessons: From Reinforcement Learning to SOP Agents | 5 | 8 | 1 | 1 | James shares a key technical failure and pivot: Brex invested heavily in reinforcement learning for credit underwriting, only to find simple web research agents and prompt-based SOPs vastly outperformed complex RL models. | |
| Ideal Customer Profiles and Retool-Driven Prompt Iteration | 6 | 6 | 2 | 3 | Swyx presses James on whether Brex's expanded ICP represents a retreat back to the SMB segment. James clarifies the specific revenue and transaction thresholds defining their commercial segment, and highlights their Retool-based prompt management tooling. | |
| Knowledge Base Grounding and the Decision to Partner with Sierra | 6 | 7 | 2 | 3 | Swyx questions why Brex bought Sierra instead of building customer support agents in-house. James explains that building low-code management UIs and domain telemetry for CX leaders is undifferentiated work best outsourced. | |
| Multi-Turn Evals, User Simulations, and Hallucination Mitigations | 7 | 6 | 1 | 2 | Alessio suggests using forward-looking, intentionally failing evals to track model progression over time. James enthusiastically adopts the idea and explains Brex's multi-turn eval techniques and prompt guardrails against phantom agent delegation. | |
| AI Fluency Framework, Upskilling, and Agentic Coding Interviews | 5 | 7 | 1 | 1 | James explains Brex's AI fluency framework and reveals that Brex instituted mandatory agentic coding re-interviews for all existing engineers and managers to force hands-on exposure and accelerate upskilling. | |
| Headcount Planning, Engineering Leverage, and Industry Headwinds | 6 | 7 | 2 | 3 | Swyx probes on whether AI productivity is causing layoffs. James takes a nuanced stance, pointing out that AI amplifies bad architecture alongside good output, and details a deep dive into Brex's graph of audit and review agents collaborating on expense fraud. |
Statements from this episode (39)
Reggio: Brex structures its AI strategy into corporate, operational, and product pillars
“We have like three pillars to our AI strategy. We have our corporate AI strategy, which is how are we going to adopt
and like buy AI tooling across the business and basically every single function to be able to 10 X our workflows. And we have our operational …”
Brex launches 'Quitters Welcome' recruiting pitch encouraging future founders
“We actually launched sort of like a new recruiting and employee value proposition for Brex a couple months ago called Quitters Welcome, where we actually intentionally are leaning into this idea that we have a disproportionate number of folks who go on to beco…”
Reggio: Brex has roughly 40,000 customers from Fortune 100 to startups
“You can come into this business And build like financial AI applications and instantly have that deployed to roughly 40,000 customers across you know, the fortune 100 down to, you know, tens of thousands of startups.”
Brex formed a 10-person internal AI team to disrupt its product
“And AI is one of those areas where we have another team of
Just roughly about 10 people who are focused primarily on LLM applications, and we wanted to create a bit of a separation there because the way that we were thinking about this, and this is actually so…”
Reggio: Cursor adoption is uniform across Brex engineering
“It's actually fairly, fairly uniform across the entire engineering department. It's actually kind of funny, like one of our largest cursor users is actually an engineering manager.”
Reggio: Deep Domain Experience Can Impede AI-First Problem Solving
“I think part of it is like sometimes the too much experience or too much knowledge of how to solve a problem and actually be an impediment to thinking differently about it and thinking about it from like an AI first lens.”
Reggio: Brex Card Product Drives 60% of Direct Revenue
“The folks who say we're kind of a card product, they drive 60% of our direct revenue.”
Brex replaced human KYC underwriting with automated web research agents
“We've we set up a completely automated Pipeline for evaluating customer applications to get them onboarded instantly to Brex, which is something that used to require human intervention either for underwriting or KYC. But now we basically have a series of agent…”
Reggio: Agentic coding has significantly shortened the half-life of code
“Because the half-life of code has declined so significantly with agentic coding. It's actually quite, ah, easy for us and for anyone else to kind of try on for size, a variety of different pieces of tech to figure out what is going to be most ergonomic for sol…”
Reggio: Half of Brex agent applications run on Mastra, half internally
“About half of the applications that we're building right now on the agent layer are running on Mastra and then the other half are actually still running on like yet another internally developed framework. Which is a framework that's focused more on networks of…”
Reggio: The Ideal Brex UX Is Just Swiping the Card
“Like the best UI UX for Brex is just the card. Like every single thing that you have to do in the software beyond just swiping the card is like an opportunity for AI to eliminate some work for you.”
Brex built a custom framework where sub-agents direct message each other
“We kind of just hit the eject button and built our own framework, which is one in which we have agents that are able to basically DM with other agents and have multi-term conversations amongst themselves to coordinate to complete a task to or like to complete …”
Reggio: Multi-turn agent interaction beats treating sub-agents as single tool calls
“There's actually a lot of value in having like multi-turn conversations from like the orchestrator or the assistant to like the sub agent, whereas like, you know, a tool call is basically just like one RPC.”
Reggio: MCP belongs on conventional systems, not inter-agent communication
“I think of like the MCP and tool usage as being like the interface to all of our conventional imperative systems, not at the AI space.”
Reggio: Multi-agent systems work best when structured like human org charts
“It's actually helpful for us to basically conceive of it as an org chart. And like, it's the agent org chart with you know, my EA is DMing other specialists and having brief conversations to support me as their client.”
Brex transitions operations staff from executing SOPs to building AI prompts
“She's been leaning really aggressively to help every member of the operations organization start rethinking their role as being not people who kind of execute against an SOP, but are people who are going to like build prompts, build evals, and like be, become …”
Reggio: Brex targets 80% automated application decisions within 60 seconds
“Like we, as soon as we put our mind to it, we said like, look, no, we want to hit 80% automated acceptance rate for all all startup and commercial businesses that apply for Brexit. Like we want a decision within 60 seconds that's fully touchless, no humans inv…”
Reggio: Brex does not pick single AI tool winners internally
“We're not going to try to pick winners in the horse race between the foundational model providers or the agentic coding tools, or like basically anywhere where there's an active horse race. What we do and instead of like trying to pick a single solution is we …”
Reggio: Rapid AI code generation increases incident response risks from knowledge drift
“As well as, like, maybe one of the other facets of being able to generate a lot more code more quickly is, like, the drift between team members as far as, like, understanding of the code that's in their services increases is like everybody's moving faster and …”
Swix: AI productivity expectations will force engineers to own far more code
“Every engineer is just gonna own more code. Period. And be parachuted in and be expected to ramp up and be productive and also fix bugs. And if you're on, you know, pager duty or whatever to just because I mean, everyone's going to try to be more efficient and…”
Reggio: Took one-month leave to code full-time with Brex AI team
“Basically went on leave for a month and join the join the team that the AI team that we were building just to go and build alongside them.”
Reggio: CS students use AI for design docs rather than raw code generation
“And I was surprised to hear the consensus was that most people there were using agents to collaborate on like building a design document and like,
Collaborating on the architecture of the solution that they want to build, and then they'd be asking it to like e…”
Reggio: Brex relies on Greptile for agentic code review
“We're big fans of creptile and we use them for basically all of sort of the Smarter than linting, ah, like, agentic code review, ah, that's been the one solution that we have aligned around that has served us extremely well.”
Reggio: Brex chose full TypeScript over Kotlin/Elixir for agent codebase
“When we started building this new agent code base, like, as we were saying, like, we were answering the question, what would you do if you built a, you know, a Brex disruptor today? And it's like, it wouldn't be to pick Kotlin and Elixir as the backend. And so…”
Reggio: Simple web research agents outperformed RL for credit underwriting at Brex
“We made this big investment. We were working with some outside, like the, like a company that specializes in this and the performance we ended up getting was inferior to just building a, like a web research agent.”
Reggio: The real challenge in operational AI is codifying unwritten SOPs into prompts
“The challenge is really articulating and refining prompts to reflect. Reflect the execution of the SOP and like reflect all the sort of institutional knowledge that isn't written down so that agents can properly replace like the humans or the contractors that …”
Reggio: Tens of thousands of ROI-negative SMB customers almost threatened Brex
“It ended up being a huge burden for the business almost existential for us to have those tens of thousands of customers that all were ROI negative.”
Reggio: Brex sets minimum ICP at $1M revenue or $10K monthly card spend
“Right now our minimum threshold is, is like a million dollars a year in recurring in, in annual revenue or like 10,000 dollars or more per month in, in card transactions as kind of being like the low end of our ICP”
Reggio: Brex built its internal operational AI management platform in Retool
“Most of the UI UX for this platform is built in retool. And so like you can basically go into retool and there's like a prompt manager, a tool manager, an eval manager.”
Reggio: Brex uses Sierra because CX agent administration is not differentiated
“That's like solving problems that are not differentiated enough for us. I think what, what's interesting about the Sierra that has been really helpful is that again, it's really easy for like the UI and UX of basically administering a Sierra agent is something…”
Brex evaluates multi-agent networks using simulated user agents and LLM judges
“And so what we do there is we try to adopt some of the state of the art for multi-turn evals where we will we'll basically have a, an agent embody the user and like, you know, have basically the end user agent is given an objective and then we basically have i…”
Reggio: Brex built LLM circuit breakers but does not use them
“Like I, we kind of built a couple of those circuit breakers or like the ability to put those circuit breakers in. And I don't believe we're using them for anything.”
Reggio: Brex gives spot bonuses and biweekly spotlights for AI workflows
“We'll do like spot bonuses for people who have like particularly novel uses of AI on, in their day to day. In our company, all hands, every two weeks, we'll do an AI spotlight. And it's very rarely somebody in EPD for the most part is folks in ETMs, ops financ…”
Reggio: Brex engineering interview revamp makes agentic coding mandatory
“We adapted our interview loop to be more AI sort of agentic coding native. So instead of we had like a coding and a system design question that we basically have revamped into a project where we'll give you like a brief, Before you come on site and then like a…”
Brex forced all existing engineers to pass its new AI interview
“As soon as we had the interview ready to ship, we started, we said everybody in engineering, including all the managers are going to have to go through this interview. And so we re-interviewed everybody internally.”
Reggio: Agentic development amplifies bad architecture, yielding less net capacity increase
“I view agentic development as being something that amplifies all the good, just as much as it amplifies all the bad. And it amplifies sloppiness, poor architectural thinking misunderstanding of the requirements. Like there are, for all of the acceleration of g…”
Brex will freeze engineering headcount at 300 while targeting doubled efficiency
“I like having 300 engineers. Like I would love, love to just, you know, a year from now have 300 engineers, but we're still, you know, a 30, 50, a hundred percent more efficient.”
Reggio: Constraining LLMs to deterministic DAGs undersells their planning power
“My intuition has been that trying to craft LLMs into deterministic workflows and DAGs is, is kind of underselling like the power that they have to actually plan and execute more in a more sophisticated, like fluid way.”
Brex built two-tier audit agents to flag and filter expense policy violations
“We built this audit agent that can, like, ingest your SOP and look, also ingest your.
And what it does then is it's basically always looking for potential violations. And what it does is it is extremely zealous. Like it wants to have a minimum number of false …”