Agentic-Washing: How to Tell a Real Agentic TMS From a Rebranded Rules Engine in 2026

What Agentic-Washing is

Every technology category eventually attracts a label rush, and “agentic” is having its moment. As agentic AI becomes the thing enterprise buyers want, vendors whose products are fundamentally rules-based automation are discovering that they, too, are “agentic,” often with nothing changed but the marketing. That is agentic-washing: presenting a system that executes preset logic as one that makes autonomous decisions.

For a logistics buyer this is not a harmless bit of positioning. The whole reason to want an agentic TMS is that it decides and adapts on its own, which is a genuinely different capability from a system that automates the rules a human configured. Buy the label and get the rules engine, and you have paid for autonomy you do not have, while still doing the decision-making the software was supposed to take off your plate. The defense is to stop evaluating the word and start evaluating the capability. This piece gives you the line that actually separates the two, seven red flags that expose agentic-washing, and the questions to put to any vendor claiming the term.

Gartner explicitly identifies “agent washing” — rebranding RPA, chatbots, and assistants as “agentic” — and estimates only about 130 of thousands of agentic AI vendors are genuine.

The Difference That Defines “Agentic”

A rules-based system executes logic a person defines in advance: if this condition, then that action. It can be sophisticated and useful, but every decision it makes was really made by whoever wrote the rules; the software just applies them. An agentic system makes the decision itself. It senses the situation, decides what to do within guardrails it was given, executes, and learns from the outcome, and it keeps doing this in real time as conditions change.

Four properties follow from that, and all four have to be present for “agentic” to be true rather than decorative. Decisions are AI-initiated, the system acts, it does not just surface options for a human to choose. Reallocation is real-time, the system re-decides as the day changes rather than running a fixed plan. The loop is closed, outcomes feed back and the system improves. And autonomy is governed, the system acts on its own within explicit oversight, explainability, and control. A product missing any of these is doing something short of agentic, whatever the datasheet says.

Also Read: How to Evaluate a Modern TMS in 2026: A Practical RFP Framework for US Enterprises

Seven Red Flags of Agentic-Washing

1. You Can Enumerate the Rules

Ask the vendor to explain how a decision gets made, and if the answer is a set of if-then rules you could write down, the system is executing logic, not deciding. Real agentic decisioning weighs many factors at once in ways that cannot be reduced to a rule list a person maintains.

2. It Suggests, but Never Decides

If the product surfaces recommendations and waits for a human to approve every one, it is assistive, not agentic. That can be valuable, but it is a decision-support tool. An agentic system acts within its guardrails and reserves human involvement for the cases that genuinely need it.

3. It Plans Once and Cannot Re-Decide

If the system builds a plan at the start of the shift and then simply executes it, with changes requiring a manual re-run, it is a planning tool. Agentic systems re-decide continuously as conditions change, which is where most of the real-world value sits.

4. It Never Learns

If the product behaves identically on day 500 as on day one, the loop is open: it is not learning from outcomes. A genuinely agentic system closes the loop, feeding actual results back so its decisions improve for the specific operation over time.

5. “Agentic” Is Just a Chatbot on Top

A conversational interface is not the same as autonomous decisioning. If the “agentic” claim rests on a chatbot layered over the same rules engine, with no autonomous operating decisions underneath, the intelligence is in the interface, not the operation. Ask what decides, not what you can talk to.

6. It Cannot Explain or Trace Its Decisions

A system that “just decides” with no explainability or audit trail is either not really deciding, or not safe to trust with decisions. Real agentic autonomy comes with governance: the ability to see why a decision was made, trace it, and control the level of autonomy. No governance is a red flag in both directions.

In a Deloitte survey of 3,235 leaders, only 21% of organizations have a mature governance model for agentic AI.

7. It Automates One Task Instead of Orchestrating the Operation

Automating a single function, routing, for instance, and calling the product agentic overstates it. Agentic operation means coordinating decisions across the whole operation, dispatch, allocation, exceptions, carriers, not optimising one step while the rest stays manual.

Also Read: TMS-WMS-ERP Integration Architecture: A 2026 Guide

The Questions to Ask a Vendor

Turn the red flags into a short interrogation for any “agentic” claim:

• When a decision gets made, who or what makes it, and can you show me the logic? (If it is an enumerable rule set, it is not agentic.)
• Does the system act on its own within guardrails, or does it wait for approval on every decision?
• What happens mid-shift when conditions change, does it re-decide automatically, or does someone re-run the plan?
• How does the system improve over time, and what does it learn from?
• Can it explain and trace a specific decision, and can I set how much autonomy it has?
• Does it coordinate decisions across the operation, or automate a single task?

Straight answers to these separate a genuine agentic TMS from a rebranded rules engine faster than any datasheet.

Deloitte finds only about 11% of organizations have agents in production despite 38% piloting them.

What Genuinely Agentic Looks Like

Locus is the world’s first agentic TMS, and it is useful here as a concrete reference for what the real thing looks like against each test. Decisions are AI-initiated: specialised agents assign, sequence, route, and reallocate work rather than surfacing suggestions for a human to approve. Reallocation is real-time: the system re-optimises continuously across 250+ real-world constraints as conditions change. The loop is closed: it senses, decides, executes, and learns, improving with the operation rather than running static logic. And autonomy is governed: explainability, traceability, evaluation, defined autonomy levels, an execution sandbox, and human-in-the-loop controls mean the system acts on its own within explicit oversight. It orchestrates across the operation through coordinating agents rather than automating a single task, and any conversational interface sits on top of that decisioning, not in place of it. That combination, AI-initiated, real-time, closed-loop, governed, and orchestrated, is the standard a genuinely agentic TMS meets.

Also Read: What is an Agentic TMS? A Practical Guide for Enterprise Logistics Leaders in 2026

Why Holding the Line Matters

Categories are shaped by whoever defines their standards. If buyers accept “agentic” as a label any rules engine can claim, the term degrades into meaninglessness and the genuinely different capability gets lost in the noise. If buyers hold the line, decide or execute, act or suggest, learn or repeat, then “agentic” keeps meaning something, and the vendors who actually built it are the ones who benefit. For a logistics leader evaluating the market in 2026, that is the practical takeaway: do not buy the word. Test for the capability, and let the answers, not the marketing, tell you whether a TMS is agentic.

Learn more, visit locus.sh.