The Supplier Agent
Forty lifecycle notices arrive in a week. Three matter. Working out which three takes a competent person most of a day, which is why the notices sit in a queue for three weeks and the answer arrives after the last-time-buy window has closed.
What it is, and what it is not
An agent scoped to the supply surface: component lifecycle data, the approved manufacturer list, supplier change notifications and the exposure each of them creates against products that actually ship.
It is not a procurement tool. It does not negotiate, does not rank suppliers commercially, does not place a last-time-buy order and does not reserve stock. Those are decisions made by people with a budget and a relationship, and an agent that attempted them would be substituting confidence for authority.
What it does is the join. A feed knows a manufacturer part is going end of life; it has no idea you buy it, under which internal number, in which three products, two of which are shipping, with eleven weeks of stock behind them. Assembling that is a traversal, and traversals are exactly what should be automated.
What it actually does
Triages the week. Forty notices become a ranked list: which affect parts on your approved manufacturer list, which of those are in resolved structures for shipping products, and how much runway the on-hand stock buys. Notices that touch nothing you build are reported as such rather than dropped silently.
Finds the alternate you already have. Before a redesign is scoped, the useful question is whether an equivalent is already qualified — on another product, by another team, two years ago. That is a graph traversal people reliably fail to run because it requires knowing what to look for.
Explains a risk score. Four steps only — ok, watch, at risk, critical — and the agent says which signals produced the step and where the feeds disagreed. Two sources reporting active and one reporting not-recommended is information, and averaging it away would be the wrong service.
Drafts the change. Where an alternate is already approved elsewhere, it prepares an ECR with the affected objects populated and the substitution proposed, at whatever permission tier you have set.
The failure it prevents
The manufacturer issued the notice in March. It reached a distributor mailbox that forwards to somebody who left in January. Nobody was careless; a routing that used to work stopped working.
In November the line stops. The replacement has a sixteen-week lead time and a different footprint, so the board needs a revision, which needs a qualification, which needs a customer notification. Two hours of work in March becomes a quarter in November.
The agent does not prevent the notice being missed — the feeds do that. What it prevents is the notice arriving as an undifferentiated status change in a list of forty, where the three that matter are indistinguishable from the thirty-seven that do not until somebody spends a day separating them.
How it meets the rest of the product
Exposure is computed through resolved structures, so a part appearing only in a superseded revision is not reported as a live risk. Systems that report everything train people to skim, which is precisely how the real one gets missed.
Matching runs on manufacturer part number against AML entries, never on descriptions or internal numbers. A house number with no AML entry matches nothing and is surfaced as unmatched rather than matched to something plausible — a wrong match is worse than a missing one because nobody goes back to check it.
Every claim cites the objects and revisions behind it: the notice, the manufacturer part, the AML entry, the products from the resolved structures. A reviewer can click through and re-run the query, and because the underlying traversal is deterministic it returns the same result.
Where a substitution is proposed, it lands as a draft change through the ordinary process — impact set computed, reviewers recommended, effectivity left to a person. Nothing reroutes on a released structure without a change.
How it meets your ERP
Runway is the number people act on and it is entirely your ERP's. On-hand, committed, open purchase orders and lead time read back from NetSuite, SAP, Dynamics 365, Epicor or Acumatica and appear on every ranked item, so a recommendation reads as “three shipping products, eleven weeks of stock, no qualified alternate” rather than as a severity label.
The supplier record masters in your ERP — terms, pricing, the vendor part number a buyer transacts against — and is read-only to the agent for the same reason it is read-only to people. What Manufacturing PLM owns is which manufacturer parts engineering has approved and qualified.
The agent will not place an order. Purchasing is transactional and nothing transactional is written in either direction. It produces the exposure that makes a last-time-buy decision well-informed and stops there, where the buyer already works.
Where the boundary is
It cannot see risk in parts it has no data for. A component bought under a house number with no approved manufacturer entry has nothing to match against, and the honest report is that nothing was found rather than that nothing exists. The unmatched list is the concrete, finite place to start.
It also does not forecast. Lifecycle data is what four commercial feeds report, disagreements included. An agent that predicted obsolescence would be selling confidence rather than information, and the first wrong prediction would cost more than every right one saved.
Facts
| Scope | Lifecycle feeds · AML · supplier notices · exposure |
| Feeds | SiliconExpert · Z2Data · Accuris · Octopart |
| Matching | Manufacturer part number only — never descriptions |
| Exposure | Through resolved structures — shipping, not historical |
| Risk scale | Four steps, with the contributing signals named |
| Alternates | Finds ones already qualified on other products |
| Runway | On-hand, committed and open orders, from your ERP |
| Never | Places orders, reserves stock, negotiates, or forecasts |
Frequently asked
What does it do that the feeds do not?
The join. A feed knows a manufacturer part is going end of life. It has no idea you buy it, under which internal number, in which three products, two of which are shipping, with eleven weeks of stock behind them. Assembling that is the day of work.
Will it flag parts we no longer ship?
No. Exposure runs through resolved structures, so a component present only in a superseded revision is not reported as live risk. Systems that report everything train people to skim the list, which is exactly how the notice that mattered gets missed.
Can it find an alternate we already approved?
Yes, and it is the question people most reliably fail to ask. One traversal finds equivalents qualified on other products, the qualifications that referenced the part, and whether a change is already open — each hop cited to the revision it was read at.
How does it handle feeds that disagree?
By showing the disagreement rather than averaging it. Two sources reporting active and one reporting not-recommended-for-new-designs is real information, and collapsing it into a single score discards the part a buyer would actually act on when deciding what to do.
Can it raise a purchase order?
No. Purchasing is transactional and nothing transactional is written in either direction. The agent produces the exposure that makes a last-time-buy decision well-informed — products affected, runway, whether an alternate exists — and the buyer places the order where they already work.
Does it predict which parts will go obsolete?
No, and an agent that did would be selling confidence rather than information. What you get is what four commercial feeds report, disagreements visible, joined to the parts you buy and the products that ship. The first wrong prediction would cost more than every right one saved.
What about parts with no AML entry?
They match nothing, and the agent says so rather than guessing at a similar part. That is an argument for maintaining the approved manufacturer list, and the unmatched list gives you a concrete and finite place to start rather than a vague instruction to tidy up.