Autonomy & skills
Decide how far the agent acts on its own, task by task. Trusted, low-risk work runs automatically; anything that touches a figure or a client waits for you.
Filing is always human-gated
The autonomy ladder
Suggest
Proposes, never acts. You see the idea and do it yourself. For anything new or sensitive.
Approve
Does the work, waits for sign-off. The agent prepares; nothing takes effect until you approve.
Auto
Acts on its own, above the threshold. Routine, reversible work. Drops to “Approve” whenever confidence is low.
Always you
Locked to a human. Filing with IRD can never be automated, by design.
Auto-act confidence threshold
90%
507090100
Below this, even “Auto” skills stop and ask you. Raise it to be more cautious; lower it to let the agent do more. Currently: anything under 90% confidence is escalated.
Agent skills9 enabled · 2 in beta
Categorise transactions
Codes every transaction to the right account & GST treatment, learning your firm's conventions.
Auto above 90%96% avg
12.4k runs
Bank reconciliation
Matches ledger entries to the bank feed and flags anything that won't reconcile.
Auto above 92%99% match
8.1k runs
Chase missing invoices
Emails clients for tax invoices it needs to claim GST, using your pre-approved templates.
Pre-approved templates only
340 sent
Apply GST treatment
Decides zero-rated, exempt, apportioned or standard. Surfaces the edge cases.
Edge cases always escalated
5.6k runs
Prepare GST return
Assembles the full GST 101 with reconciliation checks and an evidence trail.
Always prepared for your review
1.2k runs
Detect anomalies beta
Watches for unusual movements, duplicates and outliers across the whole book.
Surfaces findings, never edits
surfacing
Draft IRD correspondence beta
Writes first-draft replies to IRD queries, grounded in the client's records.
You review & send
86 drafts
Provisional tax estimates
Keeps a running provisional-tax estimate as each GST cycle closes.
Estimates only · informational
live
Draft year-end reports
Builds annual report packs from the ledger and drafts narrative notes & policies for your review.
Always drafted for your approval
214 packs
API & MCP access
Lets external AI tools call agent skills over MCP or the API, under these same autonomy rules.
Inherits each skill's setting
2 keys
File with IRD
Lodges the approved return through the IRD gateway under your firm's authority.
Locked: cannot be automated
Always youhuman-gated
Recent autonomy changesFull log in audit trail →
JM
You set Detect anomalies from Suggest → Auto2 days agoJM
You raised the auto-act confidence threshold 85% → 90%5 days agoAgent auto-dropped Categorise to Approve for Wyndham Electrical , accuracy dipped below floor6 days ago
RP
Rua P. enabled Draft IRD correspondence beta1 wk ago