What makes Brex easy or hard to automate
Brex publishes a set of REST APIs grouped by job: Transactions for card and cash activity, Expenses for expense records and receipts, Team for users, cards, departments and locations, Budgets for spend limits, and Payments for transfers. Everything is JSON over HTTPS with an OpenAPI definition, so reading data is simple from any platform that can make an HTTP call.
The harder part is that Brex already does a lot on its own. It captures receipts, enforces spend limits and syncs to accounting software natively. The automation question is rarely “how do we get data out of Brex” and more often “what judgment work still lands on finance every month”. That is where custom workflows earn their keep.
Writes deserve more care than reads. Creating cards, changing limits and sending transfers move real money or real access, so we put those behind an approval step and log every call.
The AI layer
Brex is a system of record for spend. The AI work happens around it:
- Reading context a rule cannot. A model can read a memo like “team dinner, client visit” next to a merchant and amount and judge whether it fits policy, which a static rule cannot.
- Chasing people politely. Missing receipts and memos are the main cause of slow closes. A model can write a short, specific nudge that names the charge, and parse the reply.
- Summarizing spend. Owners want “marketing spend rose because of two conference sponsorships”, not a CSV. A model can write that from grouped data you compute first.
The model proposes; code checks totals, dates and duplicates against the API before anything is written back.
Limits to plan around
- Token scope and ownership. User tokens are tied to the admin who created them. If that person leaves and the user is deactivated, integrations can break. Tokens also expire if they are not used for 90 days, so a seasonal workflow can find its token dead. Document who owns each token and rotate on a schedule.
- Pagination. List endpoints are cursor-paginated with a capped page size (Brex reduced the expenses list maximum to 100 in 2026), so a month of expenses for a large team is many calls. Build pagination and backoff in from day one.
- No documented sandbox for in-house tokens. Expect to test against live data. Start with read-only scopes and dry-run modes that write to a review sheet.
- Webhook coverage. As of 2026, Brex webhooks cover a short list of events (transfers, user updates and referrals), not card transactions or expenses, and only one endpoint can be registered per client. Plan scheduled polls for expense and transaction workflows.
How Lumeor typically connects it
We usually run n8n with an HTTP Request credential holding a read-scoped Brex token, an LLM step for review or writing, and a deterministic validation step in code. Reviewers see flagged expenses in Slack or a shared sheet with approve and reject buttons. For teams comparing platforms, our n8n vs Zapier comparison covers when Zapier’s Brex app is enough. The expense review guide walks through the review pattern in detail, and the same approach applies to Ramp. If you are not sure which finance workflow to start with, the AI readiness assessment is a short way to rank them.