Insights
AI for small manufacturers: where it helps and what to check
AI can write a first draft of a CNC program, a quote or a cut list. A person still has to check it before anything runs or ships.
If you run a workshop or a small factory in Stoke, Richmond or around Port Nelson, you know the office side takes as long as the floor. Programs to write, quotes to price, drawings to read, records to keep. AI can help with some of the repetitive office work, which may leave machinists more time for making things.
Here are six areas where AI helps a small manufacturer, and where you need to be careful. Michael has thirty years as an analyst, CTO and project manager. Our work includes the business software around the shop floor, and we can discuss machine code such as G-code, or PLC logic where the controller allows it, as part of a project.
G-code and CNC programs
G-code is the list of instructions a CNC machine follows: move here, cut at this speed, change tool. Writing it by hand for one simple part is routine. Writing it for many variants of the same part is tedious. Given a description of the part, the material, the tooling and your machine's conventions, AI can produce a first draft.
This is where care matters most. A person always checks machine code before it runs. A wrong number can crash a tool into a fixture, scrap a part or hurt someone.
A person always checks machine code before it runs.
Checks for AI-written machine code
- Read the program line by line, or have an experienced programmer do it.
- Run it through your simulator or backplot before it goes near the machine.
- Check tool numbers, offsets, feeds, speeds and safe retract heights against your own setup.
- Prove it out like any new program: single block, reduced rapids, hand on the stop.
- Keep your existing approval process.
AI saves the typing and the first pass. Your programmer's experience does the checking. The text-to-G-code demo shows the idea, but its output isn't something to run unchecked.
Quoting job-shop work
Job shops quote a lot of work they never win. Each quote means reading a drawing, estimating setup and cycle time, pricing material, adding outsourced processes and typing it up. AI can read an enquiry, pull out the quantities, material and tolerances, and draft a quote using your own rates and rules of thumb.
It doesn't know your shop. It won't know that a particular grade of stainless eats tools, or that your five-axis machine is booked for a fortnight. Your estimator reviews and adjusts every quote. The time saved is in the reading, the sums and the formatting. The AI quote builder demo shows a version of this.
Drawings and cut lists
AI can read a drawing or a parts list, pull out dimensions and quantities, and turn them into a cut list, a bill of materials or a spreadsheet. For fabrication, joinery and wood processing, a cut list that would otherwise be retyped by hand can take less time.
Reading drawings is where AI makes mistakes. It can misread a dimension, miss a revision or overlook a note. Treat the extracted list as a draft and check critical dimensions against the original. Where tolerances matter, a person signs off.
Production planning
Planning means deciding what runs on which machine, in what order, with what material. Small shops usually do this in someone's head, on a whiteboard or in a spreadsheet. AI can build a schedule from open orders, due dates, machine availability and material on hand, and rebuild it when a rush job lands or a machine goes down.
It works best when your data is tidy: accurate cycle times, up-to-date stock, honest due dates. If the inputs are guesses, so is the plan. Your planner knows which customers are flexible and which aren't, so treat the output as a starting point. The job scheduling demo shows how a plan gets produced and adjusted.
Quality records
Inspection sheets, certificates, batch records and non-conformance reports are tedious and, in many industries, required. AI can turn the measurements and notes your team already captures into a consistent format. It can also spot patterns, such as a dimension drifting toward the limit across a batch.
The record has to be true. AI must never fill in a measurement nobody took. Set it up so measurements come from people or instruments, and AI only formats and summarises them. If a customer or regulator requires specific records, check what they ask for rather than assuming a tool covers it.
Maintenance logs
Maintenance notes end up scattered across paper, texts and memory. AI can turn a quick note like "spindle noisy, topped up coolant" into a structured log entry, remind you when servicing is due, and show which machine has had the most trouble. It won't predict failures, but a tidy history makes problems easier to see.
What works and what needs care
| Area | Realistic use | What needs care |
|---|---|---|
| G-code and CNC | First drafts, repetitive variants | A person verifies every program before it runs |
| Quoting | Reading enquiries, drafting quotes | Your estimator checks pricing and capacity |
| Drawings and cut lists | Extracting lists and quantities | Misread dimensions, so check critical ones |
| Planning | Draft schedules, quick re-plans | Only as good as your cycle times and stock data |
| Quality records | Formatting and summarising measurements | Records must reflect real measurements |
| Maintenance | Structured logs and reminders | Does not replace a proper service schedule |
Where to start with AI for manufacturing
Pick one task your team groans about. Quoting is a common first choice because the gain is easy to measure: hours per quote, quotes per week. The ROI calculator gives you a rough estimate of the value of that time, which is not the same as cash saved. We'll tell you if AI is the wrong tool for the job.
We do this work through AI automation and custom software development. The manufacturing page covers what we do for local workshops and factories. Software like this usually needs adjusting as the shop changes, and ongoing support can be agreed as part of the scope.
Next step: see the engineering and manufacturing page for a workshop walkthrough and demos.
Michael ZwienerThirty years as a business analyst, CTO and founder. Leads every build at DXBE in Nelson. About us
Common questions
Manufacturing and AI: common questions
Is it safe to run G-code written by AI?
Only after a person has checked it. Read it, simulate it, and prove it out on the machine like any new program. AI does the draft, and your programmer stays responsible for what runs.
Will AI replace my machinists or estimators?
That is not the aim. The tools help with typing and routine checking, and the judgement stays with your machinists and estimators.
Can you work with my machine and controller?
It depends on the controller and how you use it. Tell me what you run and I'll tell you before we start whether it's a good fit.
What about my drawings and customer data?
Customer drawings are often confidential, so we're careful with them and I'll go through where the data goes before anything is set up. My article on AI and privacy in New Zealand covers the general principles.