Autonomy With Approval Limits: SAP Agents in Transportation

A carrier calls in a truck failure. Normally a planner opens each freight order, checks who else serves the lane, re-tenders and updates the customer. In a scenario SAP has described to the trade press, the planner simply tells Joule about the breakdown and asks it to switch carriers, and Joule makes the change across every affected freight order4.

The software stops answering questions and starts changing transactions. That raises the question every transportation leader should settle before the fourth quarter: which changes may happen without a person signing off?

What SAP has put on the table

At Sapphire in Orlando on 12 May 2026, SAP reorganized its AI strategy around agents. Its plan calls for more than 50 Joule Assistants, each coordinating a subset of more than 200 specialized agents, with a Knowledge Graph that maps business entities and processes so agents have context to work from1. SAP's chief executive framed the bar for this work plainly.

At SAP Connect in Las Vegas in October, the supply chain detail arrived. SAP expanded six supply chain assistants. The Logistics Assistant gains capabilities covering warehousing, outbound optimization, fulfillment and transportation dispatching. SAP says general availability for these new capabilities is planned for Q4 20262. SAP also announced an AI Agent Hub in SAP LeanIX as a central inventory of a company's agents3.

For transportation management, Logistics Viewpoints, reporting on an SAP briefing, lists planned agents for predictive logistics insights, consignment-order processing, freight invoice analysis, and tendering and subcontracting optimization. These are roadmap items, not shipped features4. The same report describes the intended operating model as moving away from events that land on a dashboard for a planner, toward a chain that runs from event through context, decision and recommendation to execution, with people kept in the loop when the agent's confidence is low or a transaction is consequential4.

One early customer result comes from SAP's own announcement: Doehler's CIO said its logistics issue resolution agent finds the root cause and recommends a fix, and that the team has recovered roughly a third of the time it used to lose to troubleshooting2.

Notice the wording: a one-click decision. A person still clicks.

On cost, we could not find a single published SAP rate card. Independent licensing advisers report that agents are metered in SAP's AI Units, drawing down for each step an agent takes67. Their figures differ, so treat consumption as something to forecast and negotiate, not assume.

A hard year to test new autonomy

The Strait of Hormuz has been heavily restricted since early March 2026. Haver Analytics, working from IMF PortWatch data, described it as effectively closed as of 30 August, after a partial reopening from 17 June to 14 July9. Within days of the closure, logistics visibility firm project44 reported container shipment diversions up more than 360%10. On 20 July, Yemen's Houthi movement announced a maritime blockade of Saudi Arabia, and a growing number of ships began routing away from the Bab el-Mandeb strait119.

Sensing a disruption and re-planning around it is exactly what these agents are meant to do. As of October 2026, we have not seen published evidence of how they perform in conditions like these.

The calendar adds pressure. Mainstream maintenance for SAP Business Suite 7, which includes ECC, runs through 31 December 2027, with optional extended maintenance through 20308. Our read is that for many shippers, the AI roadmap and the migration plan have become one plan.

Gartner expects spending on supply chain software with agentic AI to grow from under $2 billion in 2025 to $53 billion by 2030, and it expects enterprise deployments to lag behind vendor availability1213. Our read: that kind of lag is rarely about the software. It is about the operating model around it.

Where agents help, and where they need a limit

Our read: the best early work for agents is the work planners already find repetitive and well defined. Five areas stand out.

• Exception handling. An agent can gather the freight order, the affected deliveries, the customer commitment and the options before a planner even looks.
• Event deviations. Event Management already compares expected milestones with reported ones. An agent turns a late pickup or a missed check-in from an alert on a screen into a proposed response.
• Carrier selection and tendering. Re-tendering inside contracted carriers and agreed rates is rule-bound and fast to check. Going to the spot market is a different decision.
• Freight settlement disputes. Matching a carrier invoice against the settlement document, the agreement and the event timestamps behind a detention charge is tedious work an agent can prepare.
• Disruption re-planning. Rerouting around a closed corridor touches cost, lead time and customer promises at once. Here an agent should recommend, not execute, until it earns a track record.

What we tell clients: write approval limits the way finance writes signing authority. For each action, decide whether the agent may act alone, act and notify, recommend and wait, or never act. Set thresholds on what matters: cost increase over the contracted rate, service-level impact, customer tier, hazardous goods, cross-border moves and any change that affects a payment. A carrier swap within the same rate band on a domestic lane might be act-and-notify. A spot tender at twice the contract rate, or a short-pay on a carrier invoice, should wait for a person.

Fig. 01Art · Supply chain

Auditability is the other half. SAP says its assistants work within existing authorizations and audit trails, and that customers control the level of autonomy3. Use that. Give each agent its own technical identity, with a role no broader than a junior planner's, so the freight order history shows who acted. Log the alternatives it considered and the rule that let it act. Prefer actions that reverse cleanly.

ERP Today's reporting on SAP Connect notes that SAP describes procurement agents with tighter limits than logistics agents: a supplier-risk agent flags and a buyer decides, while manufacturing agents are described as making autonomous corrections5. Expect to tune autonomy domain by domain, not once for the whole suite.

The data an agent stands on

In our view, four foundations decide whether a transportation pilot succeeds.

First, carrier and location master data. Carrier capabilities, equipment types, location opening hours and geocodes have to be right, or the agent will confidently tender to a carrier that cannot serve the lane. Second, freight agreements and rates that live in the system rather than in spreadsheets, so the agent can tell a contracted move from a spot move. Third, event quality. If carriers report milestones late, with inconsistent codes or local times without time zones, the agent will chase phantom delays. Fourth, carrier connectivity. SAP's Business Network for Logistics supports tendering, freight invoices and milestone events through APIs and EDI for larger carriers and web portals for smaller ones4. Where a carrier is still on email, the agent has nothing to read.

None of this is new advice. What is new is the cost of ignoring it: a planner works around bad data, an agent acts on it.

What to do now

• Pick one exception type on a few lanes for a first pilot, such as carrier no-shows or late pickups, where the decision is frequent, bounded and easy to check.
• Run the agent in shadow mode first. Compare its recommendations with what planners actually did, and track acceptance rate, time to resolution, cost change and how often a decision had to be reversed.
• Write the approval matrix before go-live. List every action the agent can take and assign it to act, act-and-notify, recommend or never, with cost and service thresholds signed off by operations and finance.
• Audit your master data and event feeds for the pilot lanes: carrier capabilities, locations, agreements in the system, and the share of milestones that arrive on time with consistent codes.
• Forecast AI Unit consumption for the pilot and get unit rates and caps in writing before your next SAP renewal.
• Put the agent roadmap and your S/4HANA plan on one page, anchored on 31 December 2027.

Olympas works in SAP Transportation Management and Event Management, and this is the order we would follow: data and limits first, autonomy second.