Atoms Need a Database

The whiteboard is the system

Walk into most small American manufacturers, fabricators, or fleets and you’ll find the same operating system. A whiteboard. A clipboard. A shared spreadsheet that somebody rescued after the last time it broke. And one person who knows where everything is.

That person is usually excellent. They’re also the single point of failure. They can’t be in two places, they can’t take a real vacation, and when they retire the company loses twenty years of knowledge that never existed anywhere except in their head.

That isn’t a technology problem, exactly. It’s that nobody ever wrote it down in a form a computer could read.

Where the software went

It’s worth being fair about why. Software went where the margins were: ads, retail, media, finance, subscription tools. Those are businesses that already live in bits. There’s nothing to model, because a click is already a record. The data is a byproduct of the business happening.

Atoms are harder. A weld, a pallet, a truck, a shift, a run of parts. Somebody has to sit down and decide what a “job” is at your company, and what has to be true about one, and then walk the floor and get it wrong twice before it’s right. That work was expensive back when producing software was expensive. So outside the big manufacturers who could afford seven-figure ERP programs, it mostly didn’t happen. Everyone else got a spreadsheet and made it work.

What actually changed

Not robots. Three unglamorous things, all of which happened in the last few years.

Producing software got a lot cheaper. A system that only makes sense for one company at one plant used to be impossible to justify. Now it’s a few weeks of work.

Sensors and cameras got cheap enough to put on a line without a capital project and a budget cycle.

And models can finally read the messy stuff. The scanned packing slip, the handwritten tag, the PDF spec, the photo of a part number. That was the real wall. Most industrial data isn’t missing. It’s just sitting in formats nothing could parse, in a filing cabinet or a shared drive or an inbox.

Put those together and the thing that was uneconomical, modeling one company’s actual operation instead of making them fit somebody else’s software, is now ordinary work.

Records, an API, a surface

Underneath the vocabulary, almost every system worth having is the same three things.

Records. Somewhere the facts live. A job, a part, a run, a truck, a customer, and what has to be true about each one. This is the hard part, and it’s not really a software question. It’s a question about how your business actually works, which is why it has to be answered on the floor and not in a conference room.

An API. An agreed way for other software to read and write those records. Boring, and it’s the thing that means you’re never trapped. Your accounting system, your customer’s portal, a camera on the line, and whatever you want next year can all reach the same facts without anyone rekeying them.

A surface. Wherever a person or a machine touches it. A tablet at the line, a phone in a truck, a dashboard, a camera making a pass or fail call, a page your customer can check without calling you.

A dispatch board, a quoting tool, a maintenance log, a QA station. Same shape underneath. The surfaces are the part everyone gets excited about, and they’re also the cheapest part to change. The records are what decide whether any of it works.

Get the records right and you can change the screens whenever you like. Get them wrong and a beautiful screen doesn’t save you.

What it looks like when a plant can see itself

None of the results here are futuristic. They’re the kind of thing that shows up in the first quarter.

The quote that took three days takes an hour, because cost history is something you can query instead of something you remember. You know your real capacity this week rather than your capacity in theory. When a customer asks where their order is, somebody can answer without walking out to the floor. The knowledge stops living in one person’s head, so the guy who knows everything can finally take a week off.

And once the records exist, everything you might want later has something to stand on. A camera on the line, a forecast, an agent that drafts the purchase order. People usually try to start there. Without records underneath, none of it has anywhere to land.

Now multiply it

There’s a lot of capital and attention going into making things in America again. Most of that conversation is about hard assets: new plants, new tooling, tariffs, chips, defense capacity. All of it real.

But capacity nobody can see is capacity nobody can buy. Today, if a prime needs a supplier who can do a specific process at a specific tolerance by a specific date, finding that supplier is phone calls and somebody’s rolodex. Not because the capacity doesn’t exist. Because it isn’t written down anywhere a computer can read.

Run the same pattern across ten thousand small manufacturers and you get something more valuable than any single new factory: a supply base you can actually query. Who can do this, at what tolerance, with what lead time, with what certifications, right now. That’s the part I think is underrated. The constraint on reindustrializing isn’t only steel and machines and people. Some of it is that the sector is illegible to itself.

You don’t fix that with a national platform. Nobody wants one and it wouldn’t work. You fix it one company at a time, each of them owning their own records, with agreed ways to share the slice they choose to share.

The honest limits

This isn’t a robot story. Most of the value shows up well before anything gets automated, and some of the best outcomes never automate anything at all.

It doesn’t start with a platform either. It starts with the one workflow everybody in the shop complains about. That workflow is where the records reveal themselves, because you can’t fix it without deciding what the things in it actually are.

And it isn’t fast because of AI. It’s fast because somebody spent real time on the floor working out what a job is at your company. The tools just mean that answer turns into a working system in weeks instead of quarters.

If this is your operation

If you run something physical, and the honest answer to “where does that live” is a spreadsheet and Dave, that’s the starting point, not something to be embarrassed about. Almost everyone is there. It’s the most common setup in American industry and it’s also the reason so much of it is invisible.

We do this work one workflow at a time, on site, and you own the code and the records at the end of it. If you’re not sure what half the words in a proposal mean, we wrote a glossary for that too.