On June 3, 2026, Ramp — newly valued at $44 billion after a $750 million Series F — did not launch another card or bill-pay feature. It launched an operating system. Ramp Stack, described by the company as an AI-native platform for accounting firms, deploys AI agents that reconcile cash, manage prepaid schedules, post journal entries, run flux analyses, and shepherd the monthly close, all with an audit trail. It is Ramp's first product built specifically for CPA firms, and it is the clearest signal yet that the month-end close is becoming an agent-led workflow.
Ramp is not alone in pushing AI into accounting, but the positioning matters: 92 of the top 100 CPA firms already have clients on Ramp, and the company partners with more than 4,500 firms. Stack is the monetization of that distribution — an AI layer that turns the firm's existing Ramp data into draft workpapers the firm can review. Here is what Stack automates, what it does not, and how small business clients should evaluate an AI-assisted close.
What Ramp Stack Actually Does
Stack is not a chatbot that answers accounting questions. It is a set of agents that execute the repetitive, document-heavy steps of the close:
- Reconciliations: bank and credit card reconciliations, intercompany, and balance-sheet account reconciliations, with supporting schedules tied to the GL
- Schedule updates: prepaid expense, fixed asset, and deferred revenue schedules rolled forward with the current period's activity
- Journal entries: recurring and reclass entries posted as drafts for approval, with source documents attached
- Flux analysis: variance explanations that compare current period to prior period or budget and draft the narrative for management review
- Autonomous close workflow: sequencing those tasks, assigning owners, and tracking status to close the books — Ramp's language is "from handling the monthly close to managing close transactions and post journal entries"
The company emphasizes two design choices that distinguish Stack from a generic LLM wrapper:
- Transparency and auditability. Agent actions are logged with the source data and the logic used, so a reviewer can trace a journal entry back to the bank statement line and the rule that generated it. That is the difference between an AI note and an AI workpaper.
- Deployment via Forward Deployed Engineers (FDEs). Ramp pairs product with FDEs — engineers embedded with the firm to contextualize the firm's chart of accounts, close calendar, and approval policies before the agents run. The model is Palantir-inspired: AI plus high-touch implementation rather than self-serve activation.
An expanded Visa partnership also lets AI agents execute corporate payments with real-time controls, closing the loop from close to cash movement — a reminder that the operating system is not just reporting the close but acting on it.
Why Accounting Firms Are the Beachhead
The CPA firm channel is the ideal wedge for an AI close product, for three reasons that have nothing to do with model quality:
- Distribution. Firms aggregate small business clients that already use Ramp for cards and bill pay. An agent that can see the client's transactions, receipts, and approvals from day one has the context generic tools lack.
- Trust and liability. Firms, not software vendors, sign the compilation or review opinion. A firm that reviews an agent's draft benefits from leverage; a business that lets an agent post directly bears the risk. Selling to the firm places a licensed professional in the review loop by design.
- Staffing pressure. Firm hiring has not kept pace with client growth. A workflow that detects staffing pressure and routes hot leads — one of Ramp Stack's GTM use cases — is aimed directly at capacity-constrained firms that would otherwise turn away small business clients.
That positioning also explains the $750 million raise's stated use: building Gen AI infrastructure, including token spend management for AI costs and real-time budget tracking across teams — the very infrastructure Stack itself consumes.
What Still Requires Human Review
Stack automates the draft, not the judgment. The same failure modes that plague AI bookkeeping generally apply to an AI close, at higher stakes:
-
Exercising professional judgment on estimates and accruals. An agent can propose a bad-debt reserve, a warranty accrual, or a revenue recognition adjustment, but it cannot weigh the qualitative factors — a customer's deteriorating payment history, a product defect trend — that inform the estimate. The reviewer must.
-
Evaluating nonstandard transactions. Related-party transfers, owner draws vs. distributions, and debt vs. equity classifications turn on documents the agent may not have: operating agreements, loan covenants, and board minutes. An agent that posts a distribution as compensation will create a payroll compliance issue that no reconciliation will catch.
-
Maintaining independence and documentation. If the firm both prepares and reviews the books via the same AI, independence considerations for attest clients require careful scoping. The workpaper must show who reviewed what, when, and what changed — a requirement Stack's audit trail is designed to satisfy, but only if the firm enforces the review.
The firms that benefit most will be those that treat Stack as a first-year associate that drafts quickly and needs diligent review, not as a replacement for the review.
What Small Business Clients Should Ask
If your CPA firm tells you they now run on Stack, ask three questions:
-
What did the agent draft and what did a person review? Request that the close package flags agent-drafted entries versus human-reviewed entries, with the reviewer's initials and date.
-
Where does the agent's context come from? An agent that sees only Ramp data will miss cash transactions, personal account commingling, and non-Ramp cards. Ensure the firm's data feed includes the full transaction set, not just the Ramp subset.
-
How are token and AI infrastructure costs passed through? Stack consumes AI tokens that the firm pays for and may bill through. Understand whether the agents are included in the fixed-fee close or billed as a separate technology charge, and how token spend scales with close complexity.
For the self-contained business that does its own books, Stack is not directly relevant — it is sold to firms, not to end users. But the underlying pattern — AI that drafts reconciliations and journal entries with a transparent workpaper — will reach direct small business products within quarters. The same review discipline will apply.
Keep Your Close Human-Led
Ramp Stack is a credible step toward an AI-assisted close because it couples agents with the two things generic tools lack: the transaction context from the firm's existing Ramp footprint and a human review workflow that preserves auditability. The $44 billion valuation and the FDE-heavy rollout reflect confidence that firms, not just businesses, will pay for that leverage.
Use the leverage, but keep the close human-led: let agents draft the reconciliations and schedules that consume the most hours, and spend the saved time on the judgments that determine whether the financials are actually right.
Simplify Your Financial Management
Whether your close is assisted by Ramp Stack or run on a spreadsheet, the underlying need is the same: every reconciliation, every journal entry, and every flux explanation traceable from source document to ledger. Beancount.io provides plain-text, version-controlled accounting where the close is transparent by construction — so an AI draft and a human review meet on the same auditable record. Get started for free and keep your monthly close as rigorous as your AI is fast.