A sub-ledger is a detailed ledger for one type of balance, such as customer accounts, supplier accounts, fixed assets, inventory or payroll, whose total feeds a single control account in the general ledger. The sub-ledger answers "what does this one customer owe?" The general ledger answers "what do all customers owe?" The two totals must always agree.
Why does a sub-ledger matter once agents act?
Most finance agents spend their time in sub-ledgers, not the general ledger. Applying cash, coding supplier invoices, updating asset registers: all of these change sub-ledger records, which then flow to the general ledger automatically. That is a safer place to work, because the sub-ledger's own process controls apply.
The risk is a shortcut. An agent fixing a receivables difference by posting a journal straight to the receivables control account will make the general ledger look right while the sub-ledger stays wrong. The customer's statement and the balance sheet then disagree. Blocking direct posting to control accounts is one of the simplest and most valuable bounds in AI agents for finance.
How does it work in practice?
- Detail in the sub-ledger: each invoice, receipt, credit note or asset sits in the sub-ledger with its own record.
- Summary to the general ledger: the sub-ledger posts totals, often daily or at period end, to a control account.
- Reconciliation: at each close, the sub-ledger total is compared with the control account balance. Any difference means something bypassed the sub-ledger or a feed failed.
- Access: agents get read access to both, and write access, where granted, only through the sub-ledger's process.
Read access to the general ledger is usually the first step and is described in how to give an AI agent read access to your general ledger. Before any write access, test on past periods, as described in how do you test an AI agent before it touches the general ledger?
What does that look like in an example?
A wholesaler's agent reconciles the payables sub-ledger to the general ledger at month end and finds the control account 2,500 higher. It traces the difference to a manual journal posted directly to the payables control account by someone correcting a supplier credit note. The sub-ledger never saw it. The agent drafts two entries for a person: reverse the direct journal, and post the credit note through the payables sub-ledger against the right supplier. Both ledgers then agree, and the supplier's statement is correct.
How is a sub-ledger different from a general ledger?
The general ledger is the complete record of the business, organised by account. A sub-ledger is a detailed record for one area that summarises into a single general ledger account. The general ledger also holds entries that have no sub-ledger, such as accruals and most manual journals. Granting an agent the right to post into one is not the same as the other: a sub-ledger posting goes through that area's process controls, while a general ledger journal goes straight to the books, which is why letting an AI agent post journals in your general ledger is a separate decision with stricter bounds.
Which related terms should you know?
- Control account: the single general ledger account a sub-ledger feeds.
- Accounts receivable and accounts payable: the most common sub-ledgers.
- Fixed asset register: the sub-ledger for long-term assets and their depreciation.
Questions people ask
Why does a sub-ledger not agree with the general ledger?
Usually because something was posted to the control account directly, a sub-ledger posting failed or was duplicated, or the two were run at different cut-off times. Reconciling them at every close, and blocking direct posts to control accounts, prevents most differences.
Can an AI agent post to a sub-ledger without a person?
For routine, certified work such as applying exact-match receipts, yes, within bounds the customer sets. Items involving judgement, such as write-offs, credit notes and asset disposals, should be drafted for a person. The sub-ledger's process controls apply either way.