AI produces confident answers. Confidence is a belief, not a check against reality. Before you trust AI with your supply chain, you need to know the systems underneath it actually work.
I have spent more than 20 years in supply chain execution. In that time, I have watched our industry chase a long list of buzzwords. Most of them came and went without changing much.
AI is the first one I have watched cross over from the conference stage to the dinner table. I suspect its impact on the supply chain will be unlike anything we’ve seen before.
We Have Been Here Before
Supply chain loves a buzzword. Cloud Native. DevOps. Digital Twin. IoT. Control Tower. Blockchain. Microservices. Low-code. Each arrived with promises to reshape how the industry works. Most never made it past the hype.
Blockchain is a clean example. In 2018, Maersk and IBM launched TradeLens to digitize global shipping on a shared ledger and signed up more than 150 companies. Four years later, Maersk announced it was shutting TradeLens down because it never reached commercial viability.
The pattern repeats. A technology shows up with a big promise. Vendors rebrand around it. A few pilots launch. Most quietly fade. The ones that stick usually become plumbing nobody talks about anymore.
Others were very real. They just took longer to reach the warehouse. DevOps, microservices, and cloud-native architecture reshaped enterprise IT over the last decade and a half. Execution systems are only now catching up. Manhattan launched Manhattan Active Warehouse Management, its cloud-native WMS, in May 2020. Blue Yonder and Infios are also bringing modern cloud architectures and AI capabilities deeper into warehouse execution. In WMS, the gap between a technology trend and its widespread operational adoption has often been measured in years.
If you have been in this industry a while, you have earned some skepticism.
This Time It Is At the Dinner Table
AI is different, and you can feel it at home just as much as you see it at work.
I do not recall a single dinner conversation about digital twins. Nobody at my table ever asked how a control tower could help with homework. AI comes up constantly. Many families I know use it. Odds are yours does too.
The numbers back that up. OpenAI said in July 2026 that its models now reach more than 1 billion active users, less than four years after ChatGPT launched. Pew Research Center found the share of U.S. teens using ChatGPT for schoolwork doubled in a single year, from 13% to 26%.
Blockchain never had that. No other past supply chain buzzword did that I’m aware of. When a technology is already in the hands of a billion people, the question is not whether it reaches your operation but rather how and when.
Some of it will be slop. Some of it will change everything.
AI is going to drive more innovation in supply chain than anything we have seen. Most of it will not survive contact with a real warehouse.
Merriam-Webster named “slop” its 2025 Word of the Year, defined as low-quality digital content produced in quantity by AI. Enterprise software is not immune. Gartner predicts more than 40% of agentic AI projects will be canceled by the end of 2027 due to rising costs, unclear value, or weak risk controls. It also estimates that only about 130 of the thousands of vendors claiming agentic AI are the real thing.
The same Gartner research projects that by 2028, at least 15% of day-to-day work decisions will be made autonomously by AI agents, up from zero in 2024.
Some of the winners will be real, and they will move fast. Your vendors, your partners, and your competitors are all building with it right now. Even if you ignore every AI pitch that crosses your desk, the systems you depend on are changing underneath you.
Our Industry is Built to Avoid Change
For most of my career, the smart move in supply chain execution was to change as little as possible.
You went live. You survived. Or didn’t. Then it was left alone for five to seven years. You absorbed one massive, high-risk upgrade and started the clock again. In a 2024 Modern Materials Handling survey, 24% of respondents said they had been using their current WMS for 15 years or more.
That aversion was rational. Change is where things break. In 1999, Hershey rushed a major ERP and supply chain implementation to go live before Y2K, compressing testing and training along the way. The result was a system that struggled heading into its busiest season, leaving the company unable to fill $100 million in Halloween orders despite having inventory on hand.
When you only change once every several years, you can afford to make testing a project. You staff up, run hundreds of manual tests, sign off, and put the team back on their day jobs. Testing becomes an event, not a habit. That strategy has never been a great one for regression testing. It was an all-out effort to hit some go-live deadline. That is not a scalable strategy. It’s a cry for help.
Change is Now the Default
That model no longer holds. You cannot go from avoiding change and testing once every few years to changing constantly without testing far more often.
The pressure is coming from two directions.
The world outside the warehouse. COVID, the Suez blockage, the Red Sea reroutes, Hurricane Helene. McKinsey Global Institute research found companies can expect a supply chain disruption lasting a month or longer every 3.7 years on average, costing about 45% of one year’s EBITDA over a decade. Every disruption forces process and configuration changes.
The systems inside it. Modern platforms are evolving continuously. Upgrades used to be major projects. Now vendors deliver them on increasingly frequent schedules: quarterly, monthly, and sometimes more often. AI will only speed that up, because your vendors are using it to write and ship code faster than ever. Every release can touch your customizations, your integrations, and your data.
If a full manual regression takes six weeks and a release lands every four, you are behind before you start. Testing can no longer be an event. It has to run every time.
So let AI test it? Not quite.
It may seem like an obvious answer: just point AI at the problem. More change, more testing, let the machine handle it.
AI absolutely belongs in testing. It just does not belong in the verdict.
Modern models reason, look things up, and take action. The line is belief versus verification. When an AI alone says something works, that is its judgment. A well-designed test has no opinion. It checks a defined condition against how the real system actually behaves, and the answer is true or false. What it cannot tell you is whether you thought to check everything that mattered.
Supply chain execution is especially unforgiving here because the software does not operate in isolation. A WMS can tell you where inventory should be, but the outcome depends on routing, replenishment, labor, integrations, and the actual state of the floor, often across systems owned by different vendors. A change that looks perfectly reasonable in one system can produce an entirely different result across the operation.
Two examples from recent work in a live warehouse environment show where that matters.
What the data model misses. A configuration table controls automatic replenishment. Nothing in the schema stops you from adding a row for any item. When asked to validate replenishment configuration, an AI reading that table would be confident and wrong. Whether replenishment actually fires depends on location-level inventory settings, a valid sourcing route in a separate routing system, and release timing that is not captured in the replenishment configuration itself. Several variables have to agree. The table shows only one.
A passing test that breaks something else. An AI agent asks whether its insert succeeded and its assertion passed. It does not know to ask what else reads that table. In one real run, the trigger command was not scoped to the inserted row. It swept the entire warehouse for every eligible configuration. It cancelled a live, unrelated pick in progress and permanently moved real inventory, with no undo. The test passed. This is why test execution has to respect operational boundaries, not just confirm that individual commands succeeded.
A green checkmark tells you one thing worked. It tells you nothing about what else you touched.
Use AI everywhere around the test. It can turn your documents, recordings, and configurations into executable scenarios, build automated tests with real, representative data, spot gaps in coverage, and speed up troubleshooting. It makes continuous testing practical at a scale that was not possible before. But the verdict still has to come from the system itself. AI can help formulate the question. A deterministic test verifies the answer.
Confidence Comes from Testing
This buzzword is not going away. It is in your kids’ homework and your vendor’s release notes. Some of what it produces will be slop. Some of it will change how your operation runs. You will not always know which is which until it hits production.
The risk rarely sits in the vendor’s core code. It sits in your customizations, your data migrations, and your integrations. That is where continuous change does its damage, and that is where you need proof, not predictions.
The alternative to big-bang upgrades is not fewer upgrades. It is continuous ones, small enough that nobody needs a six-month validation cycle to trust them.
That is the problem we work on at Cycle Labs: helping supply chain teams turn constant change into something they can manage with confidence. Not by slowing innovation down, but by making testing a tool that lets you make the changes you want, when you want. Proactively, not reactively.
AI gives you more freedom to innovate. It also gives you more responsibility to validate. Use it to move faster. Then let a test tell you what is true.