Developer builds AI invoice tool that routes payments under $1,000 without human approval

Software developer Dylan Merigaud built 'ledgerloop', an AI-assisted accounts payable tool that automatically posts clean invoices below $1,000 without human review, while flagging anything above that threshold or containing discrepancies for manual approval. The system triggers a manager review on two conditions: any exception in the invoice, or a clean bill exceeding $1,000, regardless of how well it matches the purchase order. Merigaud argues that the real control risk in automated payments is not whether humans approve transactions, but whether organisations have formally defined which transactions require approval. An unwritten approval threshold, he contends, is an unauditable habit rather than a policy, and the specific dollar limit matters less than having it explicitly coded with documented reasoning. The project was shared as a demo case study, with Merigaud urging AP teams to examine whether their own straight-through processing limits are formally established and accountable.