INSIGHTS
Perspectives for decisions. Methods for action.
Manufacturing operations and controlled agents. Titles and summaries are in English; full articles are the Chinese Edition.
Rules before tools: agent levels L0–L4 and a registry
Decide what an agent is allowed to do, who signs off and how far data may travel - before choosing any architecture.
A prompt is not control: the deterministic kernel and three red lines
Any number or judgment that reaches a customer, an audit or a label must come from a traceable deterministic kernel. The model explains; it does not manufacture truth.
Three questions that decide agent selection
Source of truth, consequence and step determinism decide the engineering form - not the other way round.
Why a digital project has to start from a business problem
When a project starts from a system shortlist rather than a business problem, there is no way to decide afterwards whether it worked.
The five most common breakpoints from order to delivery
Losing control of delivery is rarely a scheduling problem. It is the absence of one operating thread across order, demand, material, production and exception handling.
Four conditions to meet before AI enters real operations
The value of AI is not decided by a model demo. It depends on whether data, process, accountability and measurement can carry a recommendation into real action.
Use 30 minutes to align your stage and smallest starting scope
The first discussion is not a pitch for an all-in-one package. Bring a delivery, readiness, inventory or legacy-integration problem. Together we identify your stage and an initial scope you can verify.
Book a focused discussion →