Operations are usually where AI proves itself first in a physical business - and where the temptation to simply speed up the old routine is strongest. Here is what functional transformation looks like across the supply chain, and where the value sits.

Why operations are where this lands first

In a business that makes and moves physical products, operations and logistics is the function most exposed to complexity and volatility: demand that swings by season and promotion, supply that arrives late, inventory spread across brands, regions and channels. That exposure is exactly why it tends to be the first place agentic AI earns its keep. The industries feeling the fastest impact in 2026 are the ones with the most volatile, most interconnected supply chains - consumer goods and retail among them.

There is a useful nuance in where the gains appear. The early wins tend to come in execution rather than planning: detecting a disruption and responding to it in real time, before it cascades. Planning improves too, but it improves second, once day-to-day execution is generating clean, live data to plan from. For a team deciding where to start, that ordering matters.

The pattern today: coordination as a full-time job

Walk any operations floor and much of the effort is coordination, not decision-making. Someone notices a shipment is running late. They check what it affects. They email planning. Planning checks stock across systems. Someone works out whether to expedite, substitute or reallocate. Someone updates the customer. Each of these is a handoff, and between each one the problem sits and grows. By the time the chain has run, the disruption has often already caused the very stock-out or missed delivery everyone was trying to prevent.

This is firefighting, and it is expensive in a way that rarely shows up on a single line of the P&L. It shows up as expedited freight, as safety stock held “just in case”, as service-level penalties, and as skilled people spending their days chasing rather than improving.

Three places AI changes the work first

Across the sector a consistent pattern is emerging in where operations teams get the most value. It is worth naming the three, because they are concrete and they build on one another.

1. Demand sensing and forecasting. Agents draw on sales, seasonality and external signals to sharpen the forecast continuously, rather than in a monthly cycle. McKinsey has estimated that AI can cut supply-chain forecasting errors materially - by up to half in some analyses - which flows directly into fewer stock-outs and less dead inventory.

2. Inventory and allocation. With a sharper forecast, agents can rebalance stock across brands, regions and channels - deciding what goes where, and flagging where a human should choose between competing demands.

3. Exception-based execution. This is the shift that changes the day. Rather than a person watching for problems, agents monitor the flow continuously and act on exceptions within agreed limits: rerouting a shipment around a delay, adjusting a carrier booking, switching to an alternative source when it is cheaper or faster - and escalating the calls that need a human.

A redesigned operation

Put those together and the disruption we started with looks different. A delay is detected the moment it happens, not when someone notices. In parallel, the effect on every affected order is calculated, viable responses are assembled with their cost and service implications, and the routine responses are actioned inside pre-agreed thresholds. What reaches the operations lead is not “there is a problem, please investigate” but “here is the problem, here is what has already been handled, and here are the two decisions that need you.”

“The operation moves from reacting to a problem after it has spread, to catching it while it is still small.”

This is the sequential-to-parallel shift from our previous article, applied to the supply chain: an orchestrator coordinating specialist agents - one watching logistics, one watching inventory, one watching supply - with people supervising and owning the outcomes.

On the systems you already run

None of this requires ripping out the ERP. The platforms most operations run on have added exactly this capability. SAP’s Joule platform now offers agents aimed at supply-chain and logistics processes and a protocol for them to coordinate; Google Cloud’s Gemini Enterprise Agent Platform provides a managed environment to build and orchestrate operations agents. Both support open protocols, including Model Context Protocol, that let agents read from and act on the systems that already hold your orders, stock and master data. Individual capabilities are at different stages of release - some generally available, some in preview - so confirm the specifics at the time of reading. The redesign runs on the estate you already have; the work is in the process, not a platform migration.

Guardrails matter more here than almost anywhere

Operations is not a low-stakes place to let software act. A wrong reroute, an over-eager reallocation or a procurement commitment made in error has immediate financial and service consequences. This is why the credible designs use controlled autonomy: agents operate within defined thresholds, high-impact actions require human approval, and changes can be simulated before they are committed. The aim is not maximum automation; it is stable, trustworthy operation without the “plan churn” that comes from a system reacting to every wobble. Governance is not a constraint bolted on afterwards - it is part of what makes the redesign safe enough to trust.

Where the value shows up

Framed properly, this is a margin and resilience story, not a cost-cutting one. McKinsey’s analyses have put the logistics-cost reduction from embedding AI in the supply chain at several percentage points, rising towards a quarter in the most complex networks, alongside meaningful improvements in forecast accuracy. But the figure that tends to move leadership is not the headline percentage; it is the compound effect: less expedited freight, lower safety stock, fewer penalties, and skilled people freed from chasing to spend their time improving the operation rather than rescuing it.

Start with one flow

As with any functional transformation, you do not do the whole supply chain at once. Pick one flow that visibly suffers from firefighting - a particular fulfilment path, a category prone to stock-outs, a lane that always seems to need expediting. Map it end to end, separate the decisions from the handoffs, and redesign it so agents carry the coordination and monitoring while your people own the decisions and the exceptions. Prove it there, and the pattern transfers.

In our first two articles we argued for redesigning work rather than speeding it up, and showed what a parallel, agent-supported workflow looks like. Operations is where that argument tends to pay off first. If you would like to see it mapped onto one of your own supply-chain flows, that is the kind of session we would begin with.

Connect with us now.