--- title: Inventory Agent description: An agent that ranks what to reorder, what will stock out, and what has gone dead — from live position, not last month's report. It drafts the PO. A buyer sends it. url: https://www.joinhorizon.co/ai-agents/inventory/ publishDate: 2026-09-22 updateDate: 2026-09-22 --- # Inventory Agent > 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. 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. ## Merchandising and margin 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](https://www.joinhorizon.co/blog/ai-margin-intelligence-ecommerce-2026/) guide, then come back here or to [operations](https://www.joinhorizon.co/ai-agents/operations/) if the pain is the morning triage rather than the buy. ## What this is not - An agent that auto-sends POs. - A single "stock" number the model is allowed to treat as truth. - A replacement for a buyer who understands vendor politics and promo risk. 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. ## How we start Name the buyer. Expose on-hand / reserved / inbound as separate fields. Then the [eCommerce AI Agents](https://www.joinhorizon.co/services/ecommerce-ai-agents/) engagement. If join keys across SKU, warehouse, and open PO are fiction, start with the [knowledge agent](https://www.joinhorizon.co/ai-agents/knowledge/). AWS is how it stays up. It is not the first question we ask you. ## Problem 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. ## Agent 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 ### Read position On-hand, reserved, and inbound separately. Treating reserved as available is how you advertise a SKU you already promised. ### Score the risk Velocity, lead time, open POs, and the promo calendar you can actually attach. Rank reorder vs wait vs excess. ### Draft the PO Vendor, qty, and the evidence the buyer needs. The agent does not submit the order. ### Buyer sends A human with buying authority approves. Tokens that are not buyer or admin cannot create a purchase order. ## Systems ### WMS / inventory API Live position with as-of timestamps. A nightly dump without as-of is how the brief lies after 10 a.m. ### Purchasing and vendors Preferred vendor, lead time, and open POs so the agent does not stack a second order on inbound. ### Demand inputs Sales velocity and any forecast you trust enough to put behind a named tool — not a model guessing from the PDP. ## 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. ## How the inventory agent stays bounded Rank what to reorder, what will stock out, and what is dead, then draft a purchase order. - **Maturity:** Level 2 — reference architecture - **Risk:** critical - **Oversight:** human-control - **Starts when:** A buyer asks what needs a purchase order. - **Tools:** A write goes through a router, a permission check, a policy check, and a budget check. The model does not commit it. An approval token lives in tool context, not in the user message. - **Stops for a person:** Every purchase order is a proposal. Tokens that are not buyer or admin cannot create one. - **Checked by:** On-hand, reserved, and inbound stay separate. The draft matches those figures. - **Untrusted data:** Spreadsheets and promo notes that try to set demand. - **If it fails:** If the inventory agent stops, name the reason: completed, budget_exceeded, timed_out, cancelled, guardrail_blocked, approval_required, tool_failure, verification_failed, or partial_completion. Retry a timeout at most twice. Back off on rate limits. After repeated verification failure, escalate. Stop when the budget is exhausted or a permission is denied. No unbounded loop. Every purchase order is a proposal. Tokens that are not buyer or admin cannot create one. - **Lifecycle:** 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. Reference architecture, not a published production deployment. The shared contract is the [AWS store-agent architecture](https://www.joinhorizon.co/blog/aws-ai-agents-for-ecommerce-factualminds-2026/#operating-contract). ## Outcomes ### A risk brief, not an auto-PO Buyers see reorder / wait / excess with evidence. The formula you already trust can keep posting; the agent covers the cases the formula cannot see. ### Reserved and inbound stay visible The brief fails if it collapses position into one "stock" number. That is a product rule, not a hope. ### Margin is in the same picture Dead stock and giveaway pricing show up here as a section and a linked guide. ## Field-guide posts - /blog/ai-inventory-agent-reorder-ecommerce-2026/ - /blog/ai-purchase-order-agent-ecommerce-2026/ - /blog/ai-demand-forecasting-agent-ecommerce-2026/ - /blog/ai-margin-intelligence-ecommerce-2026/ - /blog/ai-agent-vs-workflow-automation-ecommerce-2026/ - /blog/human-in-the-loop-ai-agents-ecommerce-2026/ - /blog/ecommerce-ai-agent-readiness-assessment-2026/ ## FAQ ### Will the agent send purchase orders? No. It drafts. A buyer with authority sends. An agent without a named approver is an unattended purchaser — we will not scope that. ### When should we not build this? When a reorder-point formula already posts the correct qty to the preferred vendor and nobody is reading promo calendars or open POs by hand. Do not wrap a working formula in an agent for the slide. Also skip it if stock is only a weekly CSV with no as-of. ### Where does merchandising and margin live? As a section on this page, plus the margin-intelligence field-guide post. Heavy ops triage stays on the operations family. ### Do you have inventory-agent case studies? Not yet. The reorder, PO, demand, and margin field-guide posts are the published depth. Do not treat AWS cost or IoT cases as inventory-agent results. We have no published agent case study that invents fill rate or turns-of-cover. --- *Source: https://www.joinhorizon.co/ai-agents/inventory/*