In brief
We were promised that artificial intelligence would free up our time for what matters. In 2026, look where it settled first: in conversations. Chatbots answering customers, voice systems screening calls, generated emails replying to other generated emails. Meanwhile the paperwork — the re-typing, the reports, the filing — is still waiting for someone to deal with it.
This article defends a simple idea: we are automating the wrong end of work — the relationship instead of the paperwork — and for a small manufacturer, the right move is exactly the opposite.
Every job has two ends
Take any trade — mine for fourteen years, quality in small manufacturing companies, but also sales, purchasing, planning. There are always two ends.
There's the paperwork end. Re-typing the dimensions from a drawing into a spreadsheet. Reassembling a file for an audit. Filing reports. Re-entering into one system what you just wrote in another. That end is tedious, invisible, and nobody chose their trade for it. It's time that produces nothing: it carries information from one place to another without adding anything to it.
And there's the relationship end. The customer calling because a part worries them. The supplier you call back to understand a delay. The machinist you go talk to on the floor instead of sending an email. That end is the exact opposite of the first: it's what builds trust, and trust is what makes contracts come back. I spent fourteen years verifying it — a quality problem announced in person, early, with a plan, almost never loses a customer. Silence does.
If you had asked me ten years ago which of the two ends a machine should take over, the answer would have taken three seconds. Nobody would have hesitated. And yet.
What we did instead
Look around you, as a customer this time. You call your insurer: a voice system asks you to describe your problem "in a few words", makes you repeat it, then offers three options that don't match your situation. You write to a software vendor's support: a conversational agent replies instantly, politely, at length — without having understood the question. You receive a prospecting email that was obviously generated; your own tool offers to answer it with a generated reply. Somewhere, two pieces of software are corresponding on your behalf, and yours summarizes the exchange for you.
I'm not naming anyone, because it isn't anyone's fault in particular. It happened for an almost mechanical reason: conversation is the easy end to automate. Text goes in, text comes out. No drawing to decipher, no tolerance to check, no system to connect. Language models do that better than anything else — so that's what we gave them to do. We didn't automate what weighed the most. We automated what was simplest to plug in.
We didn't hand the machine the chore. We handed it the customer.
And let's be honest about the result: nobody likes it. Not the customer, who can tell within three sentences that they're talking to a system, and whose attachment to the company drops a notch. Not the employee, who inherits the conversations the robot failed — meaning the worst ones, with someone already irritated. Not even the company, which saved a few salaries at the counter but pays the bill elsewhere, in a column no report shows: the slow erosion of the relationship.
What it costs a small company, precisely
For a large corporation, that erosion may be a rational trade-off. When you serve millions of customers, the individual relationship died long ago; the chatbot kills nothing, it replaces one waiting line with another.
For a small manufacturer, the math runs the other way, and that's the heart of this article. A thirty-person shop has no global brand, no advertising budget, no "customer success" department. What it has is something else: when a customer calls, a person picks up. A person who knows the file, who may have walked past the part that very morning, who can say "let me check and I'll call you back before noon" — and who calls back. The large buyers — in hydroelectricity, steel, transport — work with local shops precisely for that. It's not a charming leftover from the past. It's the competitive advantage. Often the only one that can't be copied.
Automating the relationship, for a shop, is therefore not an efficiency gain. It's dismantling, with your own hands, the one thing that sets you apart from a bigger, cheaper competitor. It's fighting on the big players' ground — volume, anonymity, self-service — after abandoning your own.
The right end
The paradox is that the other end — the paperwork — is where AI does remarkable work, on one condition: that it's framed. Not a general-purpose AI you chat with, but an AI locked into a precise task, bounded by purpose-built software that checks what it produces. I've written elsewhere about what I mean by framed AI, and what it can and cannot do in quality; I won't repeat it in detail.
But I can give the measure of what it changes, because I see it every week. In machine shops, the dimensional inspection report is the perfect example of the paperwork end: read a drawing, extract the dimensions, build the table, enter the measurements. Forty-five minutes per report, on average, of machinist time — a skilled person's time, spent carrying numbers from a sheet of paper into a table. With automatic dimension extraction verified by a human, that same report takes about seven minutes. In a shop producing a hundred reports a month, do the math: that's more than sixty hours given back every month. Not theoretical hours — hours belonging to specific people, who were hunched over a spreadsheet and no longer are.
That is AI's real dividend. Not a press release, not a conversational agent: hours. The question that matters — and almost nobody asks it — is what you do with them.
What you do with the time you win back
Because there are two ways to spend that dividend, and they lead to two very different companies.
The first: reinvest it in the machine. You saved time through automation, so you automate something else — and step by step, you end up reaching the emails, the phone, the relationship. That's the natural slope, the one the industry is sliding down right now. It produces companies that are very efficient and perfectly interchangeable.
The second: reinvest it in people. The machinist who no longer spends forty-five minutes on a report can spend ten more minutes on the part itself. The quality manager who no longer reassembles audit files by hand can call the customer before the customer calls — announce a non-conformity in person, with a plan, instead of letting them discover it in a shipment. The owner who no longer chases numbers can walk the floor. None of this shows up in a monthly indicator. All of it shows up in the only statistic that matters at the five-year mark: the customers who are still there.
That is, at bottom, the position I've defended since the start of this blog, and the third of my seven principles: technology handles the proof, the human keeps the decision — and I'll now add: the relationship. Well-placed AI replaces no one. It gives people back the time to do the part of their trade a machine will never do well.
The question to ask before automating anything
If you run a shop and someone offers you AI — someone does, these days, every week — I suggest a simple test. Ask yourself: does this automation bring me closer to my customers, or does it put something between them and me?
Extracting dimensions from a drawing, pre-filling a report, filing documents, catching a drift in measurements: all of that brings you closer, because it frees the hours you'll then spend in real conversations. A chatbot on your website, a smart answering system, generated follow-up emails: all of that gets in between. Even done well. Even politely. The rule fits in one sentence — automate what carries the information, never what carries the relationship.
And I'll leave you with the opening question: in your shop, today, where does the time go that your best people wish they could get back? Into the paperwork, or into the customers? If the answer is the paperwork — and it almost always is — then you know which end to start with. It's not the one being sold to you the loudest.
The convictions defended in this article are the ones that guided the development of Asterion Solutions, a suite of purpose-built software for small manufacturers: automation is reserved for the paperwork — extraction, reports, traceability — so the time won back goes where it counts, to people.