How it differs from a chatbot
A chatbot responds to a prompt and stops. An agent works towards a goal. It can read data from several sources, decide which steps to take, call tools such as creating a record or drafting a report, and check the result. In a plant, that might mean reviewing overnight downtime, linking it to maintenance history and preparing a summary with suggested actions before the morning meeting.
Practical uses
- Summarising long activity histories, checklist results and reports for tier meetings
- Highlighting trends and repeat issues across lines, shifts and sites
- Drafting fishbone diagrams, root cause action plans and checklists for people to refine
- Answering frontline questions from approved procedures and manuals
- Creating or updating records, such as actions, when instructed
The data foundation comes first
Agents are only as good as the data and processes they work with. Before they add value, an organisation needs consistent records of issues, actions and checks; clear ownership of work; and connections to systems such as ERP, MES, CMMS and HR. If problems are tracked in spreadsheets and email, an agent has little reliable context to work from.
Governance and human-in-the-loop
Manufacturing decisions affect safety, quality and compliance, so control matters:
- Define agents centrally so every team gets the same structured output, rather than everyone writing their own prompts.
- Decide what can run automatically and what needs human review.
- Keep accountability with people. An agent can suggest a countermeasure, but a named owner decides and verifies it.
- Limit data access to what each agent needs, and keep an audit trail of what it did.
- Improve agents with PDCA, like any other standard.
Risks to manage
- Confident but wrong outputs, especially when source data is incomplete
- Over-reliance, where people stop thinking critically about suggestions
- Inconsistent results from unmanaged, ad hoc prompting
- AI output that sits outside the management system and is never acted on
The most useful agentic AI feeds its output into the existing operating rhythm, such as tier meetings, action lists and standard work, rather than creating a parallel channel.