Staff Product Manager, AI

Why join us

Brex is the intelligent finance platform that enables companies to spend smarter and move faster in more than 200 markets. We combine corporate cards, banking, spend management, bill pay, and travel into one AI-native platform — trusted by companies like DoorDash, Coinbase, Anthropic, OpenAI, Robinhood, Zoom, Plaid, Reddit, and SeatGeek.

Working at Brex lets you push your limits, challenge the status quo, and collaborate with some of the brightest minds in the industry. We're committed to building a diverse team and an inclusive culture, and we believe your potential should only be limited by how big you can dream.

Following Capital One's acquisition of Brex, AI is one of our biggest areas of investment — with real backing and stability behind the bet.

Product at Brex

The Product team is at the forefront of Brex's mission to empower employees anywhere to make better financial decisions. With a deep understanding of the business, we identify and scope out the most impactful opportunities for Brex to tackle. We are responsible for aligning cross-functional teams — such as Engineering, Legal, Compliance, and Design — on key decisions. We set strategy and drive products from inception to launch, enabling Brex to grow rapidly and help our customers reach their full potential.

Unlocking AI Agents for the Finance Admin

Companies buy Brex to run their spend, and the person on the hook for that is the finance admin — a controller, an accounting manager, a finance ops lead. Their job comes down to two things they can never quite get ahead of: making sure money is being spent compliantly and documented in time to close the books, and knowing where the company's money is actually going.

Today they do both by hand, at a scale that doesn't work. Roughly 4% of travel and entertainment spend is out of policy — for a company spending $10M a year, that's $400K in violations, duplicates, missing receipts, and overspend. Most companies never find it, because finding it means a human reading transactions one at a time. The ones who do find it either staff a review team or bolt on a separate audit product with its own login. On the other side, the questions that actually matter — where did spend grow this quarter, which team is drifting, what's driving the number — sit behind a queue for an analyst, or a spreadsheet the admin builds themselves at month-end. Nobody is continuously watching.

The AI team is working to give every finance admin two people they can't hire today: an auditor and a business analyst.

• The auditor. An agent that reads a company's own expense policy and audit rules and reviews every transaction against them — not a 10% sample. It investigates what looks wrong, gathers the missing context by going back to the employee who spent the money, groups what a human should look at, recommends the action, closes what it can safely close, and surfaces the patterns underneath: which policies are constantly broken, which employees are repeat risks, where the real exposure is. Done right, it turns expense compliance from a post-mortem into a first line of defense that runs on every transaction, and shrinks the review pile to the exceptions that genuinely need a person.
• The business analyst. An agent that answers the questions an admin would otherwise hand to an analyst — asked in plain language against their real spend data — then learns which numbers they care about, turns those into reports their whole company can use, and eventually watches those numbers and tells them when something is off. It starts as a junior analyst who runs queries on request and grows into the analyst for the entire business.

Together they add up to a single answer for the customer: am I in control of my company's money?

As a PM on Finance Admin AI, you'd take point on one of them with real, end-to-end ownership — which one, and how much of the other you pick up, is something we'll shape around where you're strongest and what the team needs when you start.

Both are early and real — in customers' hands, with the gaps that implies. And no one has a finished playbook for this: what makes an agent's judgment about a policy trustworthy enough to act on, how you evaluate an answer that has no single correct value, how you keep quality moving up as the models shift underneath you. You'll invent much of that, with the ownership and pace of a startup and the backing of an established company behind you.

What you'll do

You'll work on our core AI team building agents for the finance admin and own one of these end to end. You'll set its product direction, decide which capabilities to build and how they should behave, and spend real time with controllers and finance teams to learn what they'll actually trust an agent to do on their behalf. From there you go as deep as it takes to make it real: shaping agent behavior, writing and iterating prompts, designing the evals that prove it works, and digging into traces when it doesn't. You'll partner closely with engineering, design, and research, and you'll set the quality bar for agents that customers rely on to keep their spend clean and their numbers right.

Where you'll work

This role will be based in either our San Francisco, Seattle, or New York City office. You must be willing to work in the office at least 3 days per week on Monday, Wednesday, and Thursday. Remote work is allowed up to 4 weeks per year, in increments of at least one week.

Responsibilities

• Own product direction for your area of the Finance Admin AI surface — the strategy, the roadmap, and the day-to-day calls on what ships.
• Translate messy, company-specific finance workflows — expense policies, review and approval chains, month-end close, the way each customer actually enforces its rules — into agent behavior that holds up in production.
• Define what "good judgment" means for an agent operating on someone else's money, and build the evals, prompting approaches, and feedback loops that get it there — then re-invent them as the underlying models and tooling shift.
• Get hands-on. Prototype, measure, read transcripts and traces, and iterate toward what actually works rather than what looks good in a demo.
• Earn autonomy incrementally: figure out where the agent should recommend, where it should act, and how a customer builds enough trust to let it move from one to the other.
• Work directly with finance admins at real companies, and let their usage — not assumptions — drive the next iteration.
• Set and defend a high quality bar for agents operating in a trust-critical, financial context.

Requirements

• You're an experienced product manager. You have a track record of owning complex, user-facing products end to end — setting direction and shipping to real users — with the scope and judgment this senior role calls for.
• You've built for company admins, and you've built it for customers. You've shipped software to the personas who keep a company running — finance, accounting, HR, payroll, IT, compliance, data — and you know how much this can vary from one company to the next. It matters that this was product you sold to other companies, not an internal system inside your own.
• Depth in company workflows or in data and reporting, and genuine interest in the other — the queues, approvals, and exceptions an admin works through, or how people get answers out of their own data and what it takes to model it so those answers are right.
• You have technical depth, and you've worked on systems without a single correct answer. Having built agentic software is a strong plus. Probabilistic, optimization-driven systems where "done" doesn't exist and the work is making it steadily better: AI/ML, search and ranking, marketplace matching, risk and fraud detection, game engines. You're comfortable when the spec is an outcome and a set of signals rather than a decision tree. Also, if you've shipped an agent that real users depended on — designed the evals, iterated the prompts, worked out where it should act on its own versus stop and ask, and lived through the failure modes — say so and tell us what you learned.
• You're comfortable with ambiguity and with systems you can't fully predict. You pull signal out of noise, make good intuitive bets, and ship and refine rather than holding out for a perfect answer. Perfect is the enemy of good here.
• Strong product taste and judgment about what's worth shipping to customers and what isn't.
• You move fast and go deep — comfortable prototyping, reading data, and getting into the details rather than working through layers.

Nice to have: hands-on depth in prompting or eval design; familiarity with fintech, expense management, or accounting workflows.

Compensation