Gleb Tsipursky, PhD, behavioural scientist and author of The Psychology of AI Adoption at Work reveals why confectionery manufacturers need clear limits on the authority given to AI agents as automation becomes increasingly autonomous.
How is AI changing the way confectionery manufacturers approach automation?
Confectionery manufacturers already use automation to improve consistency, throughput, quality control, packaging, maintenance and planning. AI is adding another layer.
The next practical challenge arrives when an AI system moves beyond analysis and begins taking actions across production and business workflows.
International Confectionery has followed the industry’s use of AI for process optimisation, and its awards programme continues to recognise technology and innovation across the sector. As those capabilities mature, manufacturers need to decide how much authority an AI agent should receive before a person has to approve its next move.
Why has this become such an important issue now?
KPMG’s September 24 AI Pulse survey shows why this question deserves attention. Sixty-two per cent of surveyed organisations are building, developing or deploying AI agents, while nearly half have already defined high-risk uses where agents cannot make autonomous decisions.
That suggests organisations are increasingly moving beyond AI systems that simply provide information or recommendations. They are beginning to consider systems capable of acting on that information.
For confectionery manufacturers, that makes it increasingly important to establish boundaries around exactly what an AI agent should and should not be able to do.
What do you mean by an ‘authority budget’ for AI?
Confectionery manufacturers can translate that principle into a simple operating tool: an authority budget for every agent.
An authority budget defines the actions an agent may take independently, actions requiring human approval and actions it may never take. The idea is to establish these boundaries before an AI agent starts operating rather than trying to determine them once the system is already interacting with production or business processes. It gives manufacturers a practical way of deciding where automation is appropriate and where human judgement must remain part of the process.
How might an authority budget work in a confectionery factory?
A maintenance agent might summarise machine history and draft a work order independently. It could require a planner’s approval before changing a preventive-maintenance schedule. It should stop when an action could affect worker safety, food safety, allergen controls, product quality or a critical production asset.
The same logic applies elsewhere. A scheduling agent could identify production conflicts but should require approval before moving a customer-critical run. A procurement agent may compare suppliers while lacking authority to commit spending above an agreed limit.
Similarly, a quality agent could flag anomalies but should not independently release product or alter a regulated record.
