Close is where accounting agents move fastest, and where the audit trail fails hardest. A variance flagger fires on day 2. A recon agent pulls the bank feed and proposes journal entries on day 3. A narrative drafter writes the MD&A summary on day 5. By day 7, no one can answer the question the audit committee actually asks, which is who decided what, on what evidence, and when.
The architectural answer is to stop treating the GL as a scratchpad. QuickBooks, Xero, NetSuite, and Sage Intacct stay the source of truth for posted entries. Dock holds the agent OUTPUT layer: reconciliation rows, variance flags, journal-entry drafts, close-narrative drafts. Each Dock row carries a platform-specific pointer (qbo_je_id, xero_je_id, ns_je_id, intacct_je_id) so the workspace knows exactly which GL artifact the row corresponds to. Agents read fresh from the platform API on every run, never from a stale Dock cache. The Dock consent gate fires the entry back to the GL ONLY after a controller approves it. The books are untouched until a human says yes.
The Dock table that runs the close looks like this:
| Account | QBO balance | Source balance | Variance | Agent flag | Recon | Reviewer | Decision | Audit |
|---|---|---|---|---|---|---|---|---|
| 1010 Cash, Chase ops | 4,218,440.12 | 4,218,440.12 | 0.00 | clean | recon-agent | M. Liao | approved | view |
| 2100 AP, NetSuite sub | 1,802,113.40 | 1,799,987.40 | 2,126.00 | timing, 3 invoices | recon-agent | M. Liao | accrue | view |
| 4100 Revenue, deferred | 982,400.00 | 1,041,200.00 | 58,800.00 | ASC 606 cutoff | revenue-agent | J. Park | hold | view |
Worked example: day-3 close, multi-subsidiary. The recon agent pulls bank feeds from three subs across QBO and NetSuite, posts proposed entries as Dock rows with qbo_je_id and ns_je_id pointers, and flags one variance as ASC 606 timing risk. The controller opens the row, sees the agent's evidence chain (bank statement line, customer contract clause, prior-period treatment), and either approves to fire back to the GL or sends it to revenue accounting for a second look. Every click is attributed. Every approval is a signed action with the human's identity, the agent's identity, and the GL pointer all bound together. The audit packet writes itself.
Why it matters: cycle time drops because the agent does the mechanical recon work in parallel across every account, while the controller's hours go to judgment calls instead of tying out spreadsheets. Audit committee defensibility improves because every entry has a named decider and a captured evidence chain, not a service-account ghost no one can interview. Multi-platform consolidation stops being a quarterly fire drill because the Dock workspace spans all four GLs with one consistent attribution model and one approval queue. Half of finance teams still take longer than five business days to close, and only about 18 percent close in three; the bottleneck is reconciliation and review, not posting, which is exactly the layer Dock owns.
Spin up a Dock workspace for your next close, point your recon agent at it, and keep the books exactly where they are.
FAQ
What's the autonomy threshold for a close agent? Read and propose, never post. Agents draft journal entries as Dock rows; the consent gate requires controller approval before the entry hits the GL. See the dangerous-ops contract.
How does multi-platform consolidation work? Each row carries a platform-specific pointer, so a single Dock close workspace can span a QBO parent, a NetSuite sub, and a Sage Intacct sub without forcing a common chart of accounts. The AP routing pattern uses the same shape.
Is the agent narrative quality good enough for the 10-Q? No. The narrative agent drafts a summary; the controller rewrites. Dock attributes both. The point is review speed, not autopilot.
What's the controller signoff cadence? Per row, not per close. Each approval is timestamped and bound to the GL pointer, which is what the audit packet needs. See also the bookkeeping and close primer.
Sources
- PCAOB AS 2201, An Audit of Internal Control Over Financial Reporting That Is Integrated with An Audit of Financial Statements. https://pcaobus.org/oversight/standards/auditing-standards/details/AS2201
- Ledge, The state of month-end close in 2026: finance team benchmarks (drawing on BlackLine, FloQast, and Numeric close survey data). https://www.ledge.co/content/month-end-close-benchmarks-for-2025