I built the approval gate, then put a price on it

In ledgerloop, a clean invoice under $1,000 processes without any human intervention. The creator of this automated system built the approval gate, then put a price on it. The company recently raised $30 million to eliminate the need for human involvement in agent payments. Despite the creator's background, they agree with critics who argue for the continued presence of human oversight, but only for amounts up to $1,000.

There are two conditions: any exception or a clean bill over $1,000. Below that threshold, a clean three-way match allows the invoice to post automatically, without any signature required. Based on test scenarios, a $730 clean invoice goes straight through, while a $9,360 clean invoice still stops because of a material discrepancy, regardless of the match result.

Similarly, a steel bar invoiced at 9% over the purchase order and a situation where 100 units were received but only 80 were invoiced both result in the invoice being stopped. These two scenarios form the foundation of the system. The actual processing of the invoice is determined by deterministic tested code, not the model. The agent reads the invoice, investigates, and proposes a course of action, but does not decide on the final amount.

The scenarios discussed are not representative of real production traffic. The main argument the creator hopes to convey is not about whether humans should approve payments, but rather that a human approval stops being a control when no one documents the specific payment thresholds. An unwritten limit is not a policy, but rather a habit that cannot be audited.

The exact number may be debatable and potentially incorrect for a particular business. However, having this limit coded into the system with a clear explanation is important. The entire process, from the agent deriving the workflow to the actual invoice routing, is demonstrated in the ledgerloop case study.