When a Repeat Job Outlives the Machinist Who Set It Up

A customer reorders a part the shop last ran fourteen months ago. The program is somewhere on the network, probably. The print is in the job folder. What nobody can find is the rest of it: which vise the part was held in, which soft jaws were cut for it, why the second operation runs at a lower feed than the CAM output says, and which dimension drifts if the coolant is low. The machinist who knew all of that is on vacation, or retired last spring. A setup that took an hour the first time takes most of a shift, the first piece fails inspection, and the job that was quoted on last year's cycle time loses money.

Most owners of small CNC job shops know some version of this. It rarely appears as a line on the income statement; it appears as long setups, scrapped first articles, late deliveries on work the shop has already done successfully, and an owner who cannot take a week off because too much of the shop's process lives in their own memory and in that of two or three senior people.

Most of what a shop knows was never written down

The problem is old, but the arithmetic around it is getting worse. Machinists are older than the workforce as a whole: Bureau of Labor Statistics survey data put their median age at 45.7 in 2025, against 42.1 for all employed people, and nearly a third of the roughly 309,000 machinists counted were 55 or older. There are about as many machinists in the decade before typical retirement age as in their late thirties and early forties, while the youngest age bands are thin.

The BLS occupational outlook for machinists and tool and die makers projects about 30,400 openings a year over the next decade, even as total employment edges down by 1%. All of those openings come from replacing people who leave the occupation or retire, not from new positions. For a small shop, that means the question is not whether its most experienced people will leave, but whether what they know will still be in the building when they do.

Part of the reason so little is written down is that the knowledge is genuinely hard to articulate. A CAM program captures toolpaths, but not how hard to clamp a thin-walled part, which indicator reading means the fixture is seated, or the habit of touching off a particular tool twice because its holder has a little runout. Much of it was learned by fixing a bad part, and it is stored as instinct rather than as a procedure. The people who hold it are also the shop's most productive, so every hour they spend writing is an hour a spindle sits idle.

The work mix makes it worse. Job shops run hundreds of part numbers in small lots, many repeating only once or twice a year. Gaps that long are where memory fails, but no single job feels important enough to stop and record properly.

Setup sheets, ERP travelers and the problem of upkeep

Shops have tried to fix this for decades, and the tools are not the weak point. Most CAM packages generate setup sheets with tool lists, stick-out lengths and offsets. Shop forums have traded setup sheet templates, macros and scripts since the early 2000s, and owners setting up new shops still post asking others what their setup sheets and inspection forms look like. Job-shop ERP systems add digital travelers and a place to attach notes to each operation.

Where these approaches tend to break down is upkeep and discipline. A generated setup sheet records what the programmer intended, not what the setup person changed at the machine to make the part come out right. Those changes get scribbled on the printout, which goes in a drawer or the recycling bin, and the next run starts from the uncorrected version. Forum posters who describe documentation efforts that stalled tend to point to discipline and management buy-in rather than software. ERP rollouts carry their own cost, because someone has to enter years of existing jobs before the system is useful for repeat work, usually the same senior person who has no time.

A Modern Machine Shop profile of Precision Profiles, a 34-person shop in Pennsylvania and a 2026 Top Shops honoree, describes the starting point plainly. When the current CEO took over, roughly half of the shop's manufacturing knowledge was in employees' heads, and much of the rest was on paper tablets, legal pads and graph paper on the floor. The shop rebuilt around an ERP system with tablets at each machine showing drawings, program numbers and setup information, plus more than 150 in-house fixtures identified by part number so proven setups could be reinstalled rather than rebuilt. It reported an average CNC setup time of about 10 minutes in the 2026 survey, against a median of about 90 minutes. Those changes accumulated over several years after the current owner bought the business in 2018, and that kind of sustained management attention is exactly what most shops of that size struggle to find.

Capturing the setup as it happens

What has changed is the cost of turning informal knowledge into usable records. The bottleneck has always been the writing, not the storage. A senior machinist will rarely sit down to type a procedure, but most will talk through a setup while doing it, take a few photos of the fixture on a phone, or mark up a printout. Until recently, that material stayed unstructured: voice memos nobody replayed and photo folders nobody could search.

Current speech recognition and language models can take a narrated walkthrough, the photos and the existing CAM setup sheet, and draft a structured record from them: workholding, tool notes, offsets, inspection watch points and the reasons behind deviations from the program. The machinist corrects the draft rather than writing it from scratch, which is a few minutes of work instead of an hour. A newer operator can then ask plain-language questions of those records at the machine, such as what went wrong the last time a part ran, without interrupting anyone. Putting a tablet or phone at the machine no longer requires a large software project to justify.

Records still have to be reviewed, tied to the right revision and updated when a setup changes. But the effort per job drops enough that capture can happen during normal work rather than as a separate project that never gets scheduled.

Fitting capture to how a particular shop already works

For a small shop, the useful version of this is usually narrow and specific. It means a tool that fits the shop's existing CAM output, file structure and job numbers, prompts the setup person for the handful of things that actually matter when a job closes out, and attaches the result to the part so it surfaces automatically when the reorder arrives. Built that way, custom AI automation does not replace a senior machinist's judgment; it gives the shop a realistic way to keep a record of it before the next retirement.