The hard part of an AI feature is knowing where NOT to use AI
A payment decision has to be exact and repeatable. So in the product I built, the money logic is deterministic code, and the agent only touches the parts where judgement is genuinely open-ended. Every AI demo right now is an agent doing everything. Point it at the problem, let it reason end to end, marvel at the trace. It demos beautifully. Then you try to put it in front of a real workflow with…
Deterministic code is essential for financial decisions because they must be exact, repeatable, auditable, and identical every time. However, adding an AI agent to the process is only effective in areas where the task is genuinely open-ended and cannot be solved with deterministic rules. For instance, interpreting a vendor invoice with varying layouts requires a vision model to extract structured data, while the downstream processing can be handled by deterministic code.
Similarly, deriving an approval workflow from an org chart is a challenging task that involves mapping titles to signing authority and resolving conflicts, which is better left to an AI agent rather than a human. Lastly, investigating flagged exceptions requires the agent to read and reason over unstructured context, such as notes, prior invoices, and vendor history, before making a recommendation.
Ultimately, the AI model serves as a tool to remove tedious tasks and hand off decision-making to the accountable person, rather than replacing the entire job.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
