What an agent needs to close the books on the ledger you already use
An agent working from an export can prepare a close, but it cannot finish one. What access it really needs to close the books, and what two-way sync with your ledger should mean.
Manoj Tutika6 min read
Working on this yourself? Talk to Mark, our co-founder. We’re here to solve it with you.
On this page
An agent can run the month-end close on the accounting software you already use, but only with current access to everything the close touches and the right to write the result back. That means the full ledger with its history and open items, the bank and source documents, your booking rules, and permission to post finished entries. An agent working from an export or a one-way copy can prepare parts of a close. It cannot finish one, because it does not see the changes other people make and cannot put its own work where it counts.
Getting this right matters more every month. Controllers are trying general-purpose assistants on their trial balance. Accounting firms are testing agents on client administrations. Vendors promise a close in days on top of whatever ledger you use. Some of this works well, and some of it quietly breaks the moment two people and one agent touch the same books.
This post is for controllers, finance leads and accounting firms deciding how to bring agents into the close. It covers what the close actually needs, the five kinds of access an agent requires, why one-way access goes wrong, and what to ask about two-way sync.
What does closing the books actually involve?
Strip away the checklist templates and a monthly close is a set of tasks that each read some data and write something back:
- Bank reconciliation. Read every bank and payment provider line, match it to invoices or book it, and write the matches.
- Supplier invoices. Read the documents, check them against orders and earlier bookings, and post them with the right account and VAT.
- Accruals and prepayments. Read what was received but not invoiced, or paid in advance, and post the adjustments.
- Depreciation and revenue recognition. Read the asset register and the contracts, and post this month's share.
- VAT check. Read every VAT-coded line and compare the position with what you expect to file.
- Review and reporting. Read the result, explain the movements, and lock the period.
Every one of those tasks reads from more than one place, and most of them write to the ledger. That is the whole problem in one sentence.
What does an agent need access to?
1. The whole ledger, not a report
A profit and loss export tells an agent the totals. It does not tell it which invoices are still open, how this supplier was booked the last twelve times, or which customer has a credit note waiting. Those details live in journal lines, open items and master data. An agent that only sees reports will guess at all of them.
History matters most. The best predictor of how a business books a transaction is how it booked the same kind of transaction before. That history is in the ledger.
2. The source data
Bank lines, payment provider payouts, supplier invoices in an inbox, contracts. The ledger records what has been booked. The close is mostly about what has not been booked yet. An agent without the bank feed and the documents can only review, never complete.
3. Your rules
Every business has rules that are written down nowhere. Software subscriptions go to one account, the annual insurance is released monthly, invoices above a certain amount need the founder's approval. An agent needs those rules stored somewhere it can read them, and a way to learn new ones when someone corrects it.
4. Permission to write back
A close produces entries: matches, bookings, adjustments. If the agent cannot write them to the ledger, someone retypes them, and you have built an expensive way to make a to-do list. Write access should be scoped, limited by amount and logged, but it has to exist.
5. Freshness in both directions
This is the one most setups miss. Your ledger is not only touched by the agent. Your accountant posts a correction, a colleague matches a payment by hand, a customer pays an invoice that the bank feed has not delivered yet. The agent needs to see those changes quickly, or it will act on a picture of the books that is no longer true.
Why is one-way access not enough?
Here is an illustrative example. It is Friday 30 October, the last working day of the month, in a company where an agent works from a copy of the ledger that is only refreshed overnight.
At 09:02 a customer pays invoice 2026-0412 for €2,420. At 10:15 a colleague matches that payment by hand in the ledger. At 11:40 the accountant reverses an accrual of €3,500 for consultancy fees, because the actual invoice came in.
The agent working from last night's copy does not know any of this. At 14:00 it sends the customer a friendly reminder about an invoice they already paid. At 17:00 it prepares the month-end accruals and books the €3,500 consultancy accrual again, because in its copy neither the reversal nor the invoice exists. Both mistakes are small. Both are exactly the kind of thing that makes a finance lead stop trusting automation.
With a two-way sync, the agent sees the payment and the reversal within its next sync, skips the reminder and leaves the accrual alone. Its own work flows back to the ledger the same way, so the accountant sees it too.
What should two-way sync mean in practice?
"Two-way sync" is easy to write on a website. Ask what is behind it.
- What is copied? The full ledger, with journal lines, open items, master data and history, or only balances and reports?
- How fresh is it? How long does a change made in your ledger take to reach the agent, and the other way round?
- What is written back? Journal entries only, or also payment matches, invoice allocations and attached source documents?
- What stays the system of record? If your accountant files from your ledger, the ledger must remain the place where the truth lives, and the agent's copy must follow it.
- What happens at the limits? Ledger APIs differ in what they allow. When something cannot be written through the API, does the agent tell you, or does it fail silently?
- What about closed periods? Does the agent respect the lock in your ledger, and does it propose corrections in the open period rather than writing into a closed one?
Ledgers like Exact Online and Xero publish APIs for entries, invoices and payments, but not every object can be read or written in the same way. A good integration is honest about where those edges are.
How do agents change the close itself?
With the right access, the close stops being an event at the end of the month and becomes work that happens during it.
Bank lines are matched the day they arrive. Supplier invoices are booked when they land in the inbox. Accruals and depreciation are prepared before the month ends, for your approval. By the first working day, what is left is review: the exceptions an agent could not decide, the estimates that need judgement, and the final look before you lock the period.
So the agent does not have to replace your ledger to close your books. It does have to see everything your ledger sees, as it changes, and be able to write its work back.
How Thred Layer does it
Thred Layer runs finance agents on top of the accounting software you already use. Thred keeps a full copy of your ledger in sync and connects to your banks, payment providers, email and other systems over their official APIs. The agents do the work in Thred, within the rules you set, and push the finished entries back to your ledger through a two-way sync.
Your accounting software stays the system of record for your accountant and your filings. Agents read by default, write only with your permission and ask above the limits you set, and every entry is traceable to its source. More on that on our security page.
For accounting firms, the same layer works across all your clients' administrations, whatever ledger each client uses. See the accounting firms page.
If you want agents in your close without moving off your ledger, book a demo.
Frequently asked questions
Can ChatGPT or Claude close my books from an export?
A general assistant can analyse an export, explain movements and draft adjustments. It cannot complete a close from a file, because it does not see changes made after the export and cannot post its work back to the ledger.
What is two-way sync in accounting software?
Two-way sync means changes in your ledger flow to the connected system and finished work from that system flows back to the ledger. For a close, it should cover journal lines, open items, master data and payment matches, not only balances.
Why is a nightly copy of the ledger risky for agents?
Between refreshes, people keep working in the ledger. An agent on yesterday's copy can chase an invoice that was paid this morning or book an accrual your accountant already reversed. Fresh data in both directions prevents those errors.
Does the ledger stay the system of record when agents work on top of it?
It should. With a two-way sync, the existing accounting software remains where your accountant works and files from, and the agent's copy follows it. Check that any integration respects the ledger's locked periods too.
What access should an accounting agent have?
Read access to the ledger, bank and documents, and scoped write access for the entries it is allowed to post, with limits above which it asks for approval. Every entry it makes should be logged with its source.
Sources
Read next
- AI FP&A use cases: what AI does well in financial planning and analysis, and what it needs underneath
- One agent, many ledgers: mapping accounts, VAT codes and journals across Exact Online, Xero and Moneybird
- How an agent learns your booking rules without you writing them
Where Thred comes in
Finance agents running on top of your existing ledger. You set the rules, approve the work and handle the exceptions. No migration.



