A small manufacturer does not always have a CMMS, quality software or a full ERP. Often, maintenance is tracked in a spreadsheet, non-conformities in a workbook, stock in an inventory tool and late suppliers in the buyer's inbox. It works as long as the people who maintain those files are around and available. It seizes up as soon as there are holidays, a production peak or someone leaves.
AI agents do not replace an industrial information system. They fill the gaps between your tools: they read what has already been entered, apply your rules, alert the right people and prepare the documents. This article reviews the three areas where they help most: maintenance, quality and supply.
Why manufacturing suits AI agents#
Industrial work is highly structured. There are maintenance intervals, stock thresholds, supplier lead times, quality procedures. In other words: rules. And wherever there are written rules, an agent can apply them consistently.
What the agent brings is not spectacular artificial intelligence, it is consistency:
- it checks every day what a person checks when they remember to;
- it follows up tirelessly and without forgetting;
- it turns scattered information into a document the production manager or the owner can read.
If the concept is new to you, the page what is an AI agent explains how it works.
Preventive maintenance: no more missed deadlines#
Preventive maintenance is driven by due dates: a number of operating hours, a number of cycles, a calendar interval. Across a few dozen machines, tracking those dates by hand quickly becomes a job in itself.
A maintenance agent reads your machine list and counters in a spreadsheet, calculates upcoming due dates, creates work orders in your task tool (Monday, Trello, Asana or another connectable app), checks that the required parts are available and sends the schedule to the technicians every Monday. That is the Preventive maintenance template.
Every Monday at 6 am, read the "Equipment" tab of my Google Sheet (machine, type of due date, interval, last service, current counter). For each job due within two weeks, create a work order in Monday with the machine, the maintenance routine and the planned parts. Check in the "Parts" tab that stock covers the needs and flag anything missing. Then send the week's schedule to the technicians on Teams, and a summary to me.
With counters read by hand#
If your counters are not reported automatically, that is not a blocker. A technician can enter readings in a simple form; the agent reads the form. What matters is that the information lives in one stable place.
Quality: every non-conformity followed through#
The problem with non-conformities is not reporting them, it is following them through to an effective corrective action. Many stay open because nobody chased the person responsible for the action.
A quality agent records each report (shop-floor form, customer complaint received by email), pre-fills an initial root-cause analysis using the method you use (5 Whys, Ishikawa, 8D), creates the corrective action with an owner and a deadline, chases it until it is closed and produces a monthly quality dashboard. That is the Non-conformity tracking template.
| Step | Who does what |
|---|---|
| Report | The operator fills in the form, the agent records and numbers it |
| Initial analysis | The agent proposes an analysis outline, the quality manager corrects it |
| Corrective action | The quality manager approves, the agent creates the action and the deadline |
| Follow-ups | The agent chases the action owner as the deadline approaches |
| Closure and review | The agent consolidates the month's indicators in a report |
The root-cause analysis the agent proposes is a working outline, not a conclusion. It saves you from the blank page, but the quality team makes the call. To keep up with the standards and regulations that affect you, the Industrial regulatory watch template summarises what is new every week.
Supply: ordering at the right time#
A component shortage stops a line; overstock ties up cash. In between, you need to know your consumption, your supplier lead times and your actual stock, and run the numbers often.
A replenishment agent compares your raw material and component stock every morning with planned consumption and supplier lead times, calculates the reorder point using the rule you give it and prepares purchase requests for approval. That is the Materials replenishment template.
Monday to Friday at 6 am, read stock levels and consumption in Zoho Inventory. For each item, calculate the reorder point as follows: average consumption over the last four weeks multiplied by the supplier lead time, plus the safety stock shown on the item sheet. If available stock drops below that point, prepare a purchase request with the usual order quantity and send everything to me for approval before 8 am.
The formula has to come from you: the agent applies it, it does not choose it. If you have no formal rule, start with a simple one and adjust it after a few weeks.
Late suppliers#
The second part is tracking orders already placed. The Late suppliers template compares confirmed delivery dates with actual receipts every day, chases late suppliers, flags the impact on production and calculates a monthly service rate per supplier. That figure, calculated from your own data, becomes a concrete argument in annual negotiations.
The weekly production report#
Many owners of small manufacturing firms spend part of Friday compiling the week's figures. The Weekly production report template gathers volumes, downtime and variances against forecast into a PDF report and an Excel file, sent by email. You read it; you no longer compile it.
Linking maintenance and purchasing#
Agents on an account can work as a chain. The most telling example in manufacturing: one agent schedules maintenance and writes the required parts into a shared table; a second agent reads those needs and prepares parts orders in time. That is the Maintenance and parts purchasing chain. The second agent only sees new rows, so it never orders the same part twice.
Agents also keep a memory: a supplier who is usually late, a temperamental machine, a preference of the production manager. You can give them context files such as maintenance routines or quality procedures. More details on the features page.
Scenario: a precision machining shop#
Take a 25-person SME that machines parts for the aerospace and medical sectors. The production manager tracks maintenance in a spreadsheet, the quality manager handles non-conformities in a shared workbook, and the buyer tracks orders in his inbox.
The company sets up three agents, one per person, starting with the simplest:
- Week 1: late suppliers. The buyer describes the task and connects his order spreadsheet and email. Every morning, the agent prepares the list of late orders and a draft follow-up per supplier. The buyer approves them in ten minutes instead of sorting through everything himself.
- Week 3: preventive maintenance. The production manager first completes his spreadsheet (intervals, last services), then launches the agent. The first work orders reveal two machines whose servicing had slipped.
- Week 6: non-conformities. The quality manager replaces her workbook with a shop-floor form; the agent numbers, tracks and chases. The monthly dashboard is produced without any compiling.
The order is deliberate: start with a follow-up whose source is already clean, then move on to those that need a bit of tidying first.
Setting up the first agent#
- Choose a follow-up that currently depends on a single person. That is where the risk is highest and the gain most visible.
- Check the source: is the spreadsheet or tool up to date? An agent that reads wrong data produces wrong alerts.
- Write the rule in one or two sentences: when to alert, whom to alert, what threshold.
- Describe the task in the chat or start from a template in the manufacturing library.
- Connect the tools, approve the sample and the monthly price announced.
- Start in "proposal" mode: the agent prepares, a person approves. Automate the steps that never raise questions.
Agents cost from €3/month, most between €9 and €49 depending on complexity, and the Studio plan suits teams that run many of them. See the pricing page.
What not to expect#
An AI agent does not run a machine, does not replace a PLC and does not make safety decisions. It works on information: it reads, calculates, alerts, prepares. That is already a lot when that information is currently scattered. For building firms, the article AI agents for construction covers job sites and subcontractors, and the manufacturing use cases page offers more scenarios.