IFS, Infor and SAP Put ERP’s AI Agents Under the Same Rules as Users

Jessica Nicole

October 18, 2026

IFS used its Unleashed conference in Orlando on Oct. 13 to launch IFS Intelligence, an AI layer that answers plain-language questions about live operational data and then acts on the answer inside IFS Cloud and connected systems. CEO Mark Moffat said industrial companies often run “at around 60% of their potential,” and that the missing 40% sits in control rooms, night shifts and data spread across systems. It works three ways: Cowork prepares actions that run only after a person approves them, Loops hands defined workflows to background “digital workers” that escalate when judgment is needed, and Runway speeds up migrating and configuring IFS itself.

The part worth reading twice is the Industrial AI Harness underneath it. IFS says agents run through the customer’s existing identity, permissions, business rules, approvals and audit controls, and that the AI “cannot and will not be able to see or do something the signed-in user is not authorized to see or do.” Every decision and action is logged, and IFS says no raw data goes to outside AI providers. Pricing is an enterprise-wide license with no feature tiers and a fair-use cap on AI executions instead of credits. Sims Crane & Equipment, an early user, says its people can now ask which cranes in the fleet best fit a given customer job.

Two men working at the consoles of a large 1960s industrial control room with a wall-mounted mimic panel and gauges. — IFS says much of a plant’s lost capacity is knowledge that already sits in the control room. Photo: Missouri State Archives, Commerce and Industrial Development Photograph Collection, ca. 1960s, no known copyright restrictions, via Wikimedia Commons.

Infor answered the same week with an expanded Industry AI architecture and a new release of its Velocity Suite. Its agents work from a shared semantic layer called Infor IQ, which gives every CloudSuite application the same definition of an item, a location or a customer, plus the industry detail underneath: a food maker’s agent has to understand the Brix reading on a sugar shipment, and an auto parts agent has to understand a VIN. Infor also exposes business-level process APIs, so an agent can create a purchase order in one call instead of chaining dozens of technical ones, which Infor says cuts errors and token costs.

Infor’s governance controls sit in the platform itself: human-approval workflows, agent permission structures, explainable logic and immutable logs of every agent action. Its blog argues agents should get zero standing privileges, with access granted for one task and removed when it ends. A YouGov survey of 2,111 business decision-makers, run for Infor in August, shows why that matters: 73% of manufacturing respondents said off-the-shelf AI does not fit their industry, respondents estimated that 44% of AI output still needs expert review, and 15% said no one in their company clearly owns AI governance.

Tall orange and blue pallet racks loaded with goods inside a Sainsbury's regional distribution centre. — An agent that can create a purchase order in one call can also fill these racks with the wrong stock in one call. Photo: Nick Saltmarsh, CC BY 2.0, via Wikimedia Commons.

SAP landed in the same place the same week. SAP Cloud ERP Private 2025 FPS02, released Oct. 13, adds agents for finance close, working capital and order-to-cash, and uses Joule with SAP Document AI in quality management to extract and match supplier certificate data, flag exceptions and pre-fill defects before inspection release. A practical test for any of the three: take one purchase order an agent created and, from the log alone, show who approved it, which permission it used and what data it read. If that takes a support ticket, the governance is a slide, not a feature.

N23D is building ERP in Rust on the same record as PLM, MES and QMS. The design treats any automated actor, whether a script or an agent, as a named user under the same roles and approval steps as a person, with every change kept in the same history as an engineering change order. That way “who released this PO, and why” has one answer, in one place, without stitching logs from three products together.

Sources

Featured photo: A large crawler crane at a Chuo Shinkansen construction site in Kasugai, Aichi, Japan, by KKPCW, CC BY-SA 4.0, via Wikimedia Commons.