T:0 brings AI-native accounting to US startups. The real test is control, not speed.

What changed

T:0, an Airwallex subsidiary, announced on 1 September that its AI-native accounting platform had moved out of private beta. The company says it is currently available to startups operating in the United States and is designed to automate bookkeeping, revenue recognition and expense reconciliation while producing current financial information.

The product description is notable for how it divides work. T:0 says AI interprets ambiguous transactions and writes explicit accounting rules. A deterministic system then applies those rules, maintains an audit trail and sends matters requiring judgment for human review. This is different from asking a general chatbot to produce accounts from raw data.

Airwallex's launch announcement and independent trade reporting confirm the release, but the claims about accuracy, speed and operating improvement remain vendor claims. Availability is also limited. A company outside the stated US-startup scope, or one with complex regulatory and reporting requirements, should not assume the current product fits its jurisdiction or accounting framework.

• The launch is a product release, not proof of performance for every finance function.
• Explicit rules and an audit trail can make automation easier to inspect.
• Human review remains necessary where contracts, estimates or policy require judgment.

Why it matters to businesses

Startup accounting often sits between disconnected bank feeds, billing systems, expense tools, spreadsheets and external accountants. The cost is not only manual entry. It is the delay between an event and a trusted financial view. An AI-native layer may reduce that delay if it can resolve routine ambiguity while preserving the evidence required for review.

The central control question is whether every automated outcome can be explained. A finance lead should be able to see the source transaction, rule, version, resulting entry, exception status and reviewer action. If the system changes a rule, the team should know whether previous periods are locked, replayed or restated. Without that lineage, a polished dashboard can make errors faster rather than make finance better.

The architecture also offers a broader lesson for business AI. Use probabilistic models to interpret messy inputs, then convert the result into explicit rules or structured decisions wherever the process requires repeatability. Keep a human path for genuine judgment and retain evidence for audit, learning and recovery.

What to do next

Start with a shadow close using a completed historical month. Connect only the minimum required systems, preserve the existing books as the control and compare every material balance. Classify differences as data, policy, rule, timing or integration errors rather than collapsing them into one accuracy percentage.

Review the system with the controller or external accountant. Confirm how it handles revenue recognition, accruals, prepayments, refunds, foreign exchange, intercompany entries and unusual contracts. Ask how rules are approved, changed, locked and replayed, and how evidence is exported if the business changes vendors.

Move to current operations only after acceptance thresholds are met. Define which transactions can post automatically, which require review and which remain outside scope. Monitor exception age and correction rate after launch because performance can change as the business adds products, entities and contract types.

• Use a completed month as a controlled comparison before live reliance.
• Measure corrections and reviewer time alongside automation rate.
• Require an exportable audit trail and documented rule-change process.
• Keep qualified accounting oversight for policy, judgment and statutory obligations.