What Finance does, with agents.

The situation

FMS pulls evidence from the operational systems over authenticated calls and keeps a fingerprinted record for every pull. Somebody still has to line that evidence up against the draft journals it should produce, and catch where source state or amounts do not agree.

How Arin helps

Arin is designed to read the persisted evidence records and the matching draft lines, confirm they balance and tie back to an approved source state, and flag the ones that do not. Where FMS fails closed on unapproved source state, Arin surfaces that as an exception rather than trying to work around it. It prepares the reconciliation and a recommendation, then routes the entry into the maker-checker queue. It never posts - there is deliberately no post-journal tool for an agent to call.

What changes

A maker sees clean, evidence-backed drafts and a short list of genuine exceptions instead of raw pulls. The pen stays with a person, and every posting still moves draft to submitted to approved to posted.

The situation

Period close is a checklist of stages that each need evidence and a decision. Chasing what is done, what is blocked and what is waiting on an approver eats a controller's month-end.

How Arin helps

Arin is designed to read the close checklist, report which stages are complete and which are blocked, and advance the ones that are ready via an approval-gated complete-stage tool. Applying the lock requires an approver, a reason and an idempotency key on every call - so the lock only lands when a named controller has signed off, and a retry never double-applies. Because there is no unlock tool in the catalog, Arin cannot reopen a locked period on its own.

What changes

The controller sees a live close status and gives one clear sign-off rather than reassembling state by hand. The lock is applied with a recorded approver and reason, fully audited.

The situation

Bank reconciliation and cash matching run end to end in a session today. Most lines match cleanly; the value is in isolating the handful that do not and explaining why.

How Arin helps

Arin is designed to walk the reconciliation, propose the obvious matches, and present the unmatched or ambiguous lines with the context behind each one. It prepares suggested pairings and a recommendation, but a person confirms the match and any resulting entry moves through maker-checker before it posts. This capability is in active build and runs in memory today, so the honest framing is assistance with judgement, not an autonomous books-closing loop.

What changes

A reconciler spends their time on real exceptions instead of ticking off matches. Nothing is treated as reconciled until a person confirms it.

The situation

Accounts payable matching compares invoices to purchase and receipt evidence pulled from the operational systems. Quantity and price mismatches need explaining and routing, and the temptation in most systems is to let automation quietly pay them.

How Arin helps

Arin is designed to read the two and three-way match results against the fingerprinted source evidence, group the exceptions by cause, and recommend a disposition for each. It routes approved items into the posting lifecycle for a maker and checker to handle. Crucially, there is no release-payment tool in the catalog, so there is no agent path to moving money - Arin prepares and recommends, a person authorizes.

What changes

AP staff get a triaged, evidence-linked exception list instead of a raw match report. Payment authority stays firmly with an accountable person.

The situation

The same operational events - an approved purchase order, a warehouse receipt - should reliably produce the same draft journals for review. Wiring that up as a dependable routine usually means engineering time.

How PECO helps

With PECO's in-app builder, a controller is designed to describe the flow in plain language: pull fresh evidence for the relevant source references, prepare the matching balanced draft lines, and drop them into the maker-checker queue. PECO renders that into a visual workflow the controller reviews and confirms before it ever runs, so the automation is explicit and repeatable rather than a black box. The routine stops at draft: it prepares and routes entries, and never posts, since posting stays governed behind separated human approval.

What changes

Recurring bookkeeping preparation becomes a defined, confirmed routine the finance team owns. Every entry it produces still lands as a reviewable draft, never a posted line.

The situation

An operator working inside FMS wants to ask an assistant plainly - what do the current statements show, is this spend within the budget guardrail, snapshot this for the record - without that assistant gaining any way around the ledger's controls.

How Semorel helps

Semorel is a client-side SDK that lets a trusted agent read a live app's semantics and invoke only authorized actions. Against FMS it is designed to read the same governed surface a person sees - financial statements, the operational pack, snapshots and budget constraints - and to invoke only approval-gated, idempotent tools such as create-snapshot or check-budget-restriction, each gated by ICE-issued scopes. The ledger-mutating tools are absent from the catalog, so no amount of assistant access reaches the pen. Every action it takes is scoped, idempotent and audited.

What changes

Operators get a live, conversational read on the books and honest budget guardrail checks in-app. The agent's reach is bounded by structure - it can explain and snapshot, but it can never post or release.