Your Supplier Changed Last Month. Why Is Your Scheduled Agent Still Ordering From the Old One?

You switched suppliers last month. The new account is set up, the price is better, and deliveries arrive on time. Then a purchase order goes out to the old supplier anyway, sent by your scheduled agent in the middle of the night. Nobody told the agent to use the old supplier. Nobody told it not to, either. It simply ran the way it ran last month, and last month the old supplier was still the right answer.

This is the quiet way automation goes stale. Nothing errors. The workflow runs, the order form fills in, and the email sends. The failure only becomes visible when the old supplier replies, or when a delivery shows up at the wrong dock, or when finance asks why two suppliers were paid for the same stock. By then the run is long finished, and the agent has already forgotten it ever happened.

A scheduled agent does not keep up with your business unless something keeps its memory current. It wakes up, reads whatever it was given at build time, does the job, and goes quiet again. If the supplier changed in a meeting, an email thread, or a new contract, and none of that reached the agent's memory, the next run will order from the supplier it already knows.

The old supplier lives in more places than the prompt

The first place to look is the obvious one: the agent's instructions. A prompt written when the old supplier was current may name that supplier outright, or point to a catalog, a price list, or an ordering portal that still belongs to it. Rewriting the prompt fixes that one copy. It does not fix the other copies hiding in the workflow.

A second copy often lives in the workflow itself. A node may hard-code the supplier's email address, a lookup step may query the old supplier's catalog first, or a default value may fill in the old supplier whenever a field is blank. The agent follows the workflow, not your intentions, so a default you forgot about will quietly win over a correction you made in a chat.

A third copy lives in past runs. If the agent reuses the last order as a template, copies the last purchase order, or pulls "the usual supplier" from a summary of previous orders, it will keep choosing the supplier that appears most often in its history. History is a useful guide until the answer changes. After that, history is exactly what drags the old answer forward.

A supplier change is rarely one fact

Part of why this failure survives is that a supplier change is never just a new name. It is a bundle of facts that all have to change together, and a scheduled run that updates only one of them will still get the order wrong.

The contact and ordering channel change. One supplier takes orders by portal, another by email, and a third through a distributor. An agent that emails an order to a portal-only supplier will wait for a confirmation that never comes.

The price and terms change. Minimum order quantities, delivery windows, payment terms, and return rules all move with the supplier. An agent working from the old terms may order too little to qualify, or promise a delivery date the new supplier never offered.

Lead time changes, and that changes planning. A supplier that delivered in two days is not the same as one that needs two weeks. An agent that reorders on the old lead time will run out of stock before the new delivery arrives.

When any one of these facts stays stale, the run produces a plausible-looking order that is wrong in a way only a human notices, and only after the fact.

Give the agent a supplier memory it reads before it orders

The fix is not to watch every run by hand. The fix is to make the supplier facts part of the memory the agent reads at the start of every run, and to update that memory in the same step as the business decision, not weeks later.

Start with a single current-supplier record for each category you buy, holding the supplier name, ordering channel, contact, price list location, minimum order, lead time, payment terms, and approval rule. One record per category, kept in one place, so there is no second copy to go stale.

Next, date every supplier change, recording what changed, when it changed, and what the old value was. A run that reads "supplier changed on the 12th, old supplier retired" will not fall back to the old name just because it appears in older orders.

Then, retire the old supplier instead of deleting it. Marking it retired, with a note about its replacement, keeps the history readable while making it unusable for new orders. The agent should be able to look back without being able to order back.

Finally, make the agent check before it sends. Before any order goes out, the run should read the current-supplier record, confirm the supplier is still approved for that category, and confirm the order matches the current terms. If the record and the workflow disagree, the run should stop and ask, not pick the workflow's default and carry on. A stopped order is a small delay. A wrong order is a delivery, an invoice, and an apology.

Why a memory layer, not another spreadsheet row

A spreadsheet row can hold the current supplier, and for one agent running one workflow, that may be enough. The problem arrives when the same supplier facts matter in more than one place: the reordering agent, the delivery-chasing agent, and the invoice agent. A row updated in one workflow is still stale in the others.

That is the gap a shared memory layer closes. Vilix AI is a cloud-hosted memory layer, so there is nothing to install or maintain. The same memory follows the agent across tools over MCP, so a supplier change saved once is visible to every agent that reads it, whether it runs in an automation platform, a script, or a phone app. It stores full conversation history, not just extracted facts, so the reasoning behind a supplier switch, the email thread where the terms were agreed, stays retrievable instead of collapsing into a single row.

The terms stay simple: a free plan that is free forever, a 7-day Pro trial that needs no credit card, and your data stays yours, with export or delete available anytime.

A supplier change should change every future run

A scheduled agent is only as current as the memory it reads at the start of a run. When a supplier changes, that change has to land in memory before the next run fires, not in a document the agent never opens. Put the current supplier, its terms, and its approval in one place the agent checks first, retire the old supplier instead of leaving it as a default, and stop any order where memory and workflow disagree.

Do that, and the next supplier change will be a quiet update instead of a wrong order discovered after delivery.