Inventory
Monday, the aisle and the open POs. An agent ranks what to reorder and what has gone dead. It drafts the purchase order. A buyer still sends it.
Last updated: September 22, 2026
The work today
Buyers still rebuild "what needs a PO" from a spreadsheet, last month's velocity, and a promo calendar in someone's head. Auto-send is worse: a flash-sale day becomes the run rate, reserved units look like a stock-out, and you stack a PO on inbound that lands next week. A reorder-point formula that already posts the right qty does not need an agent wrapped around it.
What the agent does
An inventory agent ranks SKUs as reorder, wait, stock-out risk, or excess from on-hand, reserved, inbound, lead time, and open POs. It drafts a purchase order. A named buyer sends it. Demand forecasting and PO construction are part of this family. Merchandising and margin sit here as a section — not a sixth top-level SKU.
Workflow
Reads, tools, then a stop. Skip the stop and you have a demo.
On-hand, reserved, and inbound separately. Treating reserved as available is how you advertise a SKU you already promised.
Velocity, lead time, open POs, and the promo calendar you can actually attach. Rank reorder vs wait vs excess.
Vendor, qty, and the evidence the buyer needs. The agent does not submit the order.
A human with buying authority approves. Tokens that are not buyer or admin cannot create a purchase order.
Systems
Named tools only. Anything undeclared is unreachable, regardless of the prompt.
Live position with as-of timestamps. A nightly dump without as-of is how the brief lies after 10 a.m.
Preferred vendor, lead time, and open POs so the agent does not stack a second order on inbound.
Sales velocity and any forecast you trust enough to put behind a named tool — not a model guessing from the PDP.
PO APPROVAL
Every purchase order is a proposal. Cedar should deny createPurchaseOrder for any token that is not a buyer or admin. If you cannot name the person who must approve every PO, do not build this agent yet.
Reference bounds
Level 2 — reference architecture. Risk critical. Oversight: human control.
Rank what to reorder, what will stock out, and what is dead, then draft a purchase order.
Explore, then act, then confirm, then verify. Explore gathers the context for the inventory agent. Act stays reversible. Confirm stops before an irreversible step. Verify is a separate check of the outcome.
This is the reference architecture for the page, not a published production deployment. The shared contract is the AWS store-agent architecture. Permissions and data boundaries are in securing store agents.
What good looks like
Qualitative on purpose. We do not have published agent case studies, so we will not invent a percentage.
Buyers see reorder / wait / excess with evidence. The formula you already trust can keep posting; the agent covers the cases the formula cannot see.
The brief fails if it collapses position into one "stock" number. That is a product rule, not a hope.
Dead stock and giveaway pricing show up here as a section and a linked guide.
Monday, the aisle looks fine until someone asks what to reorder today. Ads are still spending. An open PO is already inbound. Excess is sitting another season. The answers are in position, lead time, and purchasing — not in last month’s report.
This week the agent ranks reorder, wait, stock-out risk, or excess from live position. It drafts the purchase order. It does not send it.
A person still signs every purchase order, price markdown, and preferred-vendor switch.
Skip it if a reorder-point formula already posts the right qty to the preferred vendor and nobody is joining promo calendars or open POs by hand. Also skip it if stock is only a weekly CSV with no as-of. That is a fine outcome. You do not need an agent to restate a rule.
This page is the commercial summary. Reorder ranking, purchase-order drafts, and demand inputs each have a field-guide post. This page tells you what we sell and what we will not automate unsupervised.
Which SKUs are quietly destroying margin, and which need a merchandising pass, is the same live inventory and price picture. We did not add /ai-agents/merchandising/. Read the margin intelligence guide, then come back here or to operations if the pain is the morning triage rather than the buy.
There is no native Shopify AgentCore inventory connector. Shopify, Magento, and custom stacks look the same at the tool boundary — we attach to your WMS or admin APIs.
Name the buyer. Expose on-hand / reserved / inbound as separate fields. Then the eCommerce AI Agents engagement. If join keys across SKU, warehouse, and open PO are fiction, start with the knowledge agent. AWS is how it stays up. It is not the first question we ask you.
These posts are the long-form canonicals. This page does not replace them, and they are not redirected here.
AI Inventory Agent: What Should We Reorder Today? (2026)
Monday, the aisle and the open POs. An agent ranks reorder, wait, stockout, or excess. Sample cover at 5 days is a fixture. A buyer still sends the PO.
AI Purchase Order Agent for eCommerce (2026)
Monday, the unsigned PO packet. An agent drafts qty, vendor, and need-by. Qty over 2× 28-day velocity stays HITL. A buyer still sends it.
AI Demand Forecasting Agent for eCommerce (2026)
Monday, the planning table. An agent explains the number a stats model already owns. Sample horizon is 28 days. A buyer still signs the PO.
AI Margin Intelligence Agent: Which eCommerce Products Are Actually Profitable? (2026)
Merch features last week's revenue heroes. ERP cost is empty. Fill the margin contract first — unknown if cost is missing, not a ranked profit list. The ~$791/mo platform cost floor at 50,000 sessions is planning cost, not margin found.
What Should an Inventory Agent Reorder? (2026)
Ask for a risk brief, not a purchase order. A sample SKU at 5 days of cover is a fixture, not a client result. A buyer still sends the PO. Auto-send is an unattended purchaser.
AI Agent vs Workflow Automation: What Should eCommerce Businesses Use? (2026)
Do not replace Shopify Flow or the order-system cancel window with a chat bot. Hybrid is the default: rules own money; the agent returns a decision. Support-shaped pilots ran about 3× platform spend when a browser stayed on.
Human-in-the-Loop AI Agents for eCommerce: Where Automation Should Stop (2026)
Human approval is a queue with order id and tool trace — not a prompt. The public worksheet uses a $75 refund ceiling above which a person must click Approve. Never auto-approve when the SLA expires.
The eCommerce AI Agent Readiness Assessment: Is Your Business Ready? (2026)
Score org-wide readiness out of 30. Below 16 out of 30, do not let it change orders. McKinsey found 62% experimenting and 23% scaling — still a minority per function.
eCommerce AI Agents on AWS
Production AI agents for eCommerce on Amazon Bedrock AgentCore — support and WISMO, inventory, merchandising, margin, returns and B2B. Tool boundaries, evals before launch, and a human on anything that moves money.
Agentic Commerce Readiness
Make your store sellable to AI shopping agents. ACP and UCP protocol surfaces, an MCP server over your catalog, and product data that survives an agent comparison — built on AWS.
Family page → industry or decide tree → the engagement. AWS cases on this site are not agent results.
Other families
The same WISMO, tracking, and returns questions every day. An agent looks up the order, answers from live data, and hands off the moment it is out of its depth.
A B2B AI sales agent that prepares quotes, reorders, and account answers from customer-specific price — not the public catalog. Sales still owns the relationship.
The morning pile: exceptions, stalled orders, and the work nobody wants to triage. An agent ranks what needs a human today — it does not silently rewrite the warehouse.
If orders, customers, products, and inventory cannot be joined, an agent will guess. This is the data layer that makes every other agent honest.
Common questions
Tell us how POs get made today. If you cannot name the buyer who must approve every one, we will stop there.