A machine shop can use AI beyond quoting for five jobs: sorting incoming RFQs, drafting work instructions from setup notes, formatting inspection paperwork, writing scheduling notes and sending customer status updates. All five produce text, which suits AI. None should create measurements, programs or decisions that a person hasn't checked.
Quoting gets the attention because it's visible, but it is only one of several office jobs that eat a shop owner's evenings. The other five are costed below in hours and dollars for an illustrative 12-person shop, with the line between a draft and a decision marked for each.
The shop and the rules before any job
Take an illustrative job shop: 12 people, 14 machines (mills, lathes and a couple of finishing stations), about 35 RFQs a week, 60 open jobs, and a mix of repeat parts and one-offs. The office is the owner, an estimator who also does scheduling, and a part-time quality person. All the figures below are made up for modelling. Replace them with your own before you decide.
Three rules apply to every job, so they are stated once:
- Drawings and customer data are restricted until cleared. Read the NDA and any export-control terms. If you cannot clear a drawing for a tool, give the AI the RFQ text and your own notes instead.
- Use a business plan. ChatGPT Business and Claude Team are $20 a seat billed annually or $25 monthly with a two-seat minimum, and Microsoft 365 Copilot Business is a $21-a-seat add-on to a Microsoft 365 business plan on annual billing ($25.20 monthly). Business plans don't train on your content by default. Check each vendor's page before buying.
- AI drafts, a named person decides. Every output goes to the person responsible for that record: the estimator, the programmer, the inspector or the owner.
Three of the five jobs get a full costing below: RFQ triage, work instructions and inspection paperwork. Scheduling notes and customer updates are lighter and shown at the end. Quoting itself is covered in what AI quoting software costs a machine shop.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowJob 1: RFQ triage before anyone prices anything
A request for quote arrives as an email with attachments, a spreadsheet or a portal message. Most of them are not worth a full estimate: wrong material, unrealistic lead time, quantities below your minimum, or a customer you've already decided not to serve. Triage is the 10 minutes of reading before a quote is started, and it is where an estimator's morning disappears.
The AI job is a first-pass summary and score. Paste the text of the RFQ (not the drawing, if it is restricted) and ask the assistant to pull out the facts a person would look for, check them against your shop's rules, and say what is missing.
You are helping an estimator at a small machine shop. Below is an RFQ.
Our shop rules:
- Materials we cut: aluminium, mild and stainless steel, brass, plastics (acetal, nylon)
- Max part size: 600 x 400 x 300 mm
- Tolerances tighter than +/-0.02 mm need a quality manager review
- Minimum order value $400
- Standard lead time 15 working days
Extract: part count, material, quantity, tolerances, finish, delivery date,
certification required, and any missing information. Then give a verdict:
QUOTE, ASK FIRST (list questions), or DECLINE (give the rule it breaks).
Do not guess missing values. Say "not stated".
[paste RFQ text]An illustrative output for a typical request: "Parts: 3. Material: 316 stainless for two parts, 'brass' for one (grade not stated). Quantity: 50, 50, 200. Tolerances: +/-0.01 mm on one bore. Finish: passivate. Delivery: 10 working days. Certification: material certs. Verdict: ASK FIRST. Questions: brass grade? Is the 10-day date firm for all three parts? Is the +/-0.01 mm on the bore critical to function? Note: tolerance below +/-0.02 mm needs quality review, and 10 days is below the standard 15." The estimator reads that in 30 seconds, instead of opening four attachments to find out.
What triage costs and saves
| Line | Before AI | With AI first pass |
|---|---|---|
| RFQs a week | 35 | 35 |
| Minutes per RFQ to read and sort | 12 | 4 (read the summary, check the odd ones) |
| Estimator hours a week | 7.0 | 2.3 |
| Hours saved a week | - | 4.7 |
| Value at $38 an hour (loaded) | - | about $178 a week |
| Tool cost | - | $25 a month per seat, two seats about $50 |
On those numbers the saving is around $770 a month against $50 of tools, but the figure that matters more is faster replies. A customer who gets a question back within an hour is likelier to stay in the running than one who waits two days. The check on the AI is simple: once a week, take ten RFQs it marked DECLINE and have the estimator confirm they were right to decline. If more than one is wrong, fix the shop rules in the prompt. A wrongly declined repeat customer costs more than the time saved. For the next step, turning RFQ emails into draft quotes builds on the same summary.
Job 2: work instructions from what people already know
Every shop has parts that only one machinist knows how to run properly: the order of operations, which jaws, where the part tends to move, what the fixture needs. When that person is off, the job slows or scraps. Writing it down is the fix, and the reason it doesn't happen is that nobody has the 45 minutes.
The AI job is turning rough material into a clean sheet. The input can be the setup sheet, the operator's voice memo talking through the job, a few phone photos of the fixture and a list of tools. The output is a one-page instruction with numbered steps, the tool list, the critical dimensions to check at each stage and a "watch out" box. A fuller picture of the photo-and-video approach is in AI work instructions from photos and short videos of the line.
Below is a transcript of our machinist describing how he runs part
4417 (aluminium bracket, two operations) and the tool list from the
setup sheet. Write a one-page work instruction:
1. Header: part number, material, machine, fixture
2. Numbered steps for each operation, one action per step
3. Tools and offsets table
4. "Check before you continue" points with the dimension and tolerance
exactly as stated in the transcript
5. "Watch out" box with anything he said goes wrong
If something is unclear or missing, write "CONFIRM WITH [name]" in the
step instead of filling it in.
[paste transcript and tool list]Illustrative output for step 6: "Flip the part and locate on the soft jaws (bored to 38.00 mm). CONFIRM WITH Dev: torque for the vise." The machinist reads it, answers the confirm points, and signs it. Total time: about 10 minutes for the AI draft plus 10 minutes of the machinist's review, against roughly 45 minutes to write the same sheet from scratch.
The cost of 40 recurring parts
If 40 repeat parts need a sheet, writing them by hand costs about 30 hours, or around $1,140 at $38 an hour. With AI drafting and machinist review, it is about 13 hours, around $500, with a better chance of finishing it. The drawing stays out of the AI tool if it is restricted; the transcript and tool list contain nothing the customer owns. Practical note: voice memos on a phone work better than asking someone to type, and a machinist talking while standing at the machine says more than the same person typing at a desk.
The risk is that a sheet looks authoritative. A made-up torque or a wrong offset in a clean layout is worse than a rough note. The "CONFIRM WITH" rule is the control, plus a revision stamp and the name of the person who proved the instruction on a real job.
Job 3: inspection paperwork without inventing results
Inspection records, first-article reports and control plans are where AI is tempting and where the risk is highest. The rule is clear: AI can structure and cross-check, but the numbers come from a person with a gauge.
Three uses are safe. First, building the blank report from the drawing's dimension list: the inspector types or reads the balloon numbers, features, nominal values and tolerances into the prompt, and the AI produces a table with a row for each, an empty "measured" column and a pass/fail column. Second, cross-checking a completed report: paste the table and ask it to list any row with a blank measured value, any measured value outside its tolerance, and any feature on the drawing missing from the report. Third, writing the customer cover note summarising a completed report.
An illustrative check: a 22-feature report is pasted back with the instruction "list rows with blank measurements, rows outside tolerance and any inconsistency between nominal, tolerance and result." The AI returns "Feature 14: measured 12.043 against 12.000 +/-0.025: outside tolerance, marked PASS. Feature 19: measured value blank." That catches a wrongly marked pass that a tired inspector at 5 p.m. missed. The decision on what to do next is the inspector's and the quality manager's.
Cost and time
Say the shop issues 12 first-article reports a week at 25 minutes of formatting and cross-checking each, five hours in all. With AI building the template and cross-checking, that falls to 10 minutes each, two hours a week, a saving of three hours or about $114. Add a monthly sample: the quality manager takes five finished reports and re-measures three features on each part, to prove the records match reality. That costs about an hour a month and is the safeguard that makes the whole job defensible. Never let the AI "fill in" a blank result, and never ask it to estimate a measurement from a photograph. Where a customer or standard requires specific record formats, copy them into the prompt exactly, and check with the customer before changing the layout. If you are weighing camera-based checks, AI visual inspection cost covers that separate, more expensive route.
Job 4: scheduling notes the floor can follow
The shop's schedule lives in the estimator's head and a whiteboard. The AI job here is small and useful: turn the week's job list into plain-language notes for each machine and each person. Paste the list of open jobs with due dates, operations remaining and machine, plus the known constraints (a lathe in for service on Wednesday, a supplier delivery of bar stock on Thursday). Ask for a one-page plan per machine for the next three days, with a list of jobs at risk of missing their date and the reason.
The assistant doesn't schedule better than an experienced planner; what it does is spot conflicts and write the notes. An illustrative flag: "Job 2231 needs the 5-axis for 6 hours on Wednesday, but the machine is booked for service 08:00-12:00. Job 2231 ships Friday. Options: move service to Thursday, or move operation 2 to the second mill and accept 2 more hours." The planner makes the call. Time cost: about 15 minutes to paste and review each morning, against 30 minutes of writing up the plan by hand. Use it with the capacity planning check for taking an order so that the quoting side and the floor side agree.
Job 5: customer updates before the customer asks
Customers ring when they don't know where their parts are. A short status message, sent at three points (material arrived, in production, shipped), removes most of those calls. The AI job is to write the message from the job record: "Order 2231, 200 brackets: material arrived Monday, machining started Tuesday, we expect to ship Friday. We'll message again when it leaves." A person approves the batch, and the same template can be used by anyone in the office.
The one rule is that updates state facts and dates, not promises. An update that says "on track" when a part is two days behind is worse than silence. If the plan slips, the message should say so, give the new date and, where useful, offer a split shipment. Cost: about two minutes per job to approve, compared with the 10 minutes that a phone chase costs when it happens, and most shops with 60 open jobs see a handful of chase calls a day.
Writing the shop rules page that triage depends on
The triage prompt is only as good as the rules in it, and most shops have never written theirs down. They live in the estimator's instincts: "we don't do titanium", "never quote a weekend delivery", "that customer pays late". Writing a page of them is useful even before AI reads it, because the owner and estimator find out where they disagree.
A workable page has five parts. Capabilities: materials, maximum and minimum part sizes, processes in-house and processes you send out. Limits: tolerance thresholds that need a quality review, minimum order value, minimum and maximum quantities. Lead times: standard times by job type and the surcharge for rush work. Customer rules: accounts on hold, accounts that need a deposit, and industries or certifications you don't serve. Questions: the five things you always have to ask when an RFQ lacks them, such as material grade, surface finish and whether the tolerance is functional. Keep it under one page and put a date on the version, so that when a rule changes the prompt changes with it.
Start with wording that is easy for a machine to apply. "Aluminium alloys 6061 and 7075 only" is easier to follow than "mostly aluminium". "Decline under $400 unless a repeat customer" is better than "small orders aren't worth it". Where a rule has an exception, write the exception next to it. When the AI declines something the estimator would have quoted, the fault is nearly always an unwritten exception, and the fix is a single added line.
All five jobs on one page
| Job | Illustrative time saved a month | Value at $38 an hour | Main risk | Named checker |
|---|---|---|---|---|
| RFQ triage | about 20 hours | about $770 | Wrongly declining good work | Estimator |
| Work instructions | about 17 hours, one-off for 40 parts | about $630 once | Confident wrong detail | Machinist who ran the job |
| Inspection paperwork | about 13 hours | about $490 | Results nobody measured | Inspector and quality manager |
| Scheduling notes | about 5 hours | about $200 | Missing a constraint not in the list | Planner |
| Customer updates | count your chase calls first | depends on volume | Promising a date that slips | Office lead |
Add the first, third and fourth lines and the recurring saving is around 38 hours a month, about $1,450, against a tool bill near $50 to $100. That is an estimate from invented volumes, so don't quote it to your bank. Do take from it the order of magnitude: these jobs are worth doing, but they will not change the shop's fortunes alone. They free the estimator and quality person for the work that does, such as chasing the repeat customers and tightening the process capability on your hardest parts.
The programmed shortcut that scrapped a batch
An illustrative mistake: a programmer, short of time, asks an assistant to adjust a feed rate in an existing program for a tougher batch of stainless. The suggestion looks sensible and he runs it after a quick read. The change interacts with a tool change earlier in the program, and the spindle overloads and breaks an end mill. Two parts are scrapped, and the lost time is about three hours.
The fix is a rule that drafts of anything that runs on a machine go through the same prove-out as hand-written code: simulation, a dry run and a first-off check. That is why this list leaves programs out. If you want to try it anyway, read what a machine shop must check before trusting AI-written G-code first.
Putting a first month together
- Week 1: write the shop rules (materials, size limits, tolerance thresholds, minimum order, lead time) in a single page, and clear the tool with the customers whose terms you need to check.
- Week 2: run RFQ triage in parallel with the estimator's normal read. Compare verdicts for a week without acting on the AI's.
- Week 3: start work instructions with the three repeat parts that cause the most trouble when the regular operator is away.
- Week 4: add the inspection cross-check on one customer's reports and the three-point customer updates on jobs over a week long.
Measure hours per week on each job before and after, the share of triage verdicts the estimator changed, and any scrap or rework that traces back to a document. If you'd like help choosing among these for your own shop, that is what my AI implementation consultation covers.
Questions machine shop owners ask about AI
Is it safe to upload customer drawings to an AI tool?
Not by default. Many drawings sit under a non-disclosure agreement, and some are controlled by export rules. Read the customer's terms first, use a business plan that doesn't train on your content, and keep restricted drawings out of any tool you haven't cleared. Where in doubt, give the AI the text of the RFQ and your own notes instead of the drawing.
Can AI write the CNC program?
Treat it as a drafting aid at most. A program that looks right can crash a spindle or scrap a part, and nothing in an assistant's output has been simulated on your machine and tooling. If you try it, simulate every line and have a programmer prove it out, as covered in the G-code tutorial linked in this series.
Which of the five jobs should I start with?
Start with RFQ triage. It uses text you already receive, a mistake is caught before anyone cuts metal, and the time saving shows within a week. Work instructions come second because they need an operator to verify. Inspection paperwork comes third because records must be accurate and signed by the right person.
Does AI inspection paperwork replace the inspector?
No. AI can format reports, cross-check that every dimension on the drawing has a row and flag blanks. The measurements, the pass or fail decision and the signature belong to the inspector. A report that contains numbers no one measured is a quality failure and potentially worse.
Further reads
- AI for a Laser Cutting Business: Price by Material, Time and Waste — Pricing by material, time and waste for a cutting business.
- Wholesale RFQ Response Emails: AI Drafts From Your Price List — Answering request-for-quote emails from a price list.
- Factory Energy Audit With AI: Find Peak Loads and Idle Waste — Find idle waste and peak loads on the shop floor.
- How to Spot Patterns in Complaints, Returns, and Defects With AI — Spot patterns in complaints, returns and defects.
- AI Automation Setup for Small Businesses: DIY or Hire It Done? — Whether to set up automation yourself or hire it done.
- WIP Tracking for a Job Shop: A Spreadsheet With AI Status Updates — A two-tab spreadsheet for jobs and operations, the formulas that flag late work, and prompts that draft customer updates from the daily export.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI, Anthropic and Microsoft business plan pages; general machine shop and quality-records practice; illustrative figures throughout.