Most businesses don't need a more autonomous AI agent. They need a less broken process. Agree or disagree?
retained text
Earlier this year a distribution client asked us for an autonomous agent to process their orders end to end. Before quoting we spent a day sitting next to the ops person who actually handles those orders. It was pretty clear they didn’t need what they’d asked for. About 90% of their orders were fine and needed nothing smart. They came in from Shopify and went straight into the ERP. She spent most of her day on the other 10%. Usually a missing PO number or an address that didn’t match between the two systems. Sometimes a customer on a contract price the ERP didn’t know about.. So I mostly agree with the title with one tweak. In most places I’ve seen, the exceptions are what’s broken. A few things I want to know from people here. Should you fix the whole workflow before adding AI or point agents at the exceptions normal automation can’t handle? I’d go exceptions with a human approving. When an order breaks, our agent pulls context from both systems, works out what’s probably wrong and drafts a fix and she approves it with one click. Exceptions used to take most of her day. Now it’s about an hour. Does standardising a workflow make it easier to automate but worse for customers? Sometimes yeah. So before I standardize anything I ask each person why they do it their way. If the reason has to do with the customer or with risk, like a contract that allows split shipments, I keep it and write it down as a rule. If the answer is “that’s how I was shown” it usually gets standardized. How do you handle those? and should you get rid of the workarounds people invented before you automate? Back to the sticky note, it listed 6 customers whose orders she always checked by hand. 2 years earlier one of them had been shipped the wrong pallet size and almost cancelled their contract. The “autonomous” version would’ve shipped straight past it. The rule I use now is that workarounds that just move data around are usually waste and those are the ones to automate. Workarounds that check something are usually safeguards and each one has a story you need to hear first. The only reason we kept that check was that I asked her why those six. I try to ask that before cutting anything now. If anyone’s found a better way of sorting them, I’m all ears. TLDR: the routine 90% usually runs fine and the time goes into exceptions. Automate the routine stuff plainly, let an agent prep fixes for the exceptions with a human approving and don’t remove a workaround until you know what went wrong the first time.
document fetched 2026-10-27 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1xfz5la/most_businesses_dont_need_a_more_autonomous_ai