Natural-language interfaces are becoming inexpensive to add to robot and warehouse software. A prompt can describe an objective, but it does not prove that the objective can be executed safely or economically.
The hard system starts after the prompt: grounding intent, resolving ambiguity, checking constraints, simulating consequences, issuing bounded commands and recovering when reality diverges from the plan.
Interface novelty will compress quickly. Defensibility is more likely to come from the warehouse world model, operational history, integrations, control policies and validated execution loops.
Vendors should disclose which decisions are generated, which are deterministic, what can be simulated and how unsafe or infeasible requests are rejected.
Evaluate the execution chain, not the demo dialogue. Ask what happens when inventory, traffic or station capacity changes mid-mission.