I went to Printing United in September to find out where AI actually is in the workflow tools, so far its Software that doesn’t remember.

I stopped at a booth selling print workflow software to the paper printing industry. The rep walked me through it.

What it does

The software sits between the customer’s file and your equipment. Files come in through a storefront, direct, or FTP. The software preflights them. The rep said it checks around 140 criteria. It handles color management, it adds bleed and crop boxes and it builds a print-ready PDF. It can impose and nest, it connects to finishing equipment through connectors, and it is equipment-agnostic and it runs on XML data.

Half of the common mistakes in screen making start with a file nobody checked.

Where it lines up with what I build

More than I expected.

Structured data first. The rep said it plainly: as long as you have structured data, you can set it up. That is the same foundation under the press log. Validate before release. Their preflight passes or fails a file before it moves on. Our rotation pipeline does the same with print order. The art room approves it. The floor only sees approved work. Agnostic to the equipment. Their software does not care what the finishing device is. The data is the product. The machine is just where it lands.

Where it splits

I asked whether the system records anything about how the job was printed. From a historical perspective.

The rep said, “It won’t.” That is the whole difference.

It checks the file. It does not learn the print. Quality check in their system means preflight at the front and warehousing and shipping at the back. Everything in between is the press. The press is where the job actually succeeds or fails. That part is not recorded.

It runs on rules, not history. I asked if it was based on what you train it. The answer was that it is rules-based. You program it. That works for a file. A file has a spec. A print has a press, a mesh, an ink, an operator, and a humid Tuesday. You cannot write a rule for what you have never recorded.

That last part is why I built logging into the floor. The setup time work records what setup actually cost. The screen room logs mesh per screen, in screen order, because mesh is a decision, not a default. The press tablet logs whether the floor ran the released rotation and, if not, what it changed. Order, flash, or both. That override is the history the vendor’s system does not keep. And maybe this is the difference between screen printing variables and paper printing variables. I did paper printing really early in my career, and the variable count on screen printing is exceeding when you get to press. On paper printing, it was exceeding in pre-press.

And history has to be honest to be useful. We learned that the hard way in Which Number Is Lying. A forgotten timer can fake a day of work. Recording is not enough. You have to close the failure modes underneath it.

Rules and learning are not competitors

A rules engine is the right tool for a file. Files should be boring and predictable. If a rule can catch it, a rule should catch it.

A rules engine only knows what someone already told it. The press knows things nobody has written down yet. That is where AI earns its place. Not in replacing preflight. In capturing the EOM, the exposure, the override, and the setup minutes, so the next prediction starts from what your shop actually did.

That is the point I keep making in °FAI: Human in the loop AI, Keep the Failure. The failure is where the learning is. A system that only checks the input never sees the failure.

What I took home

The workflow software at the show is good at the part of the job that is already understood. It is silent on the part that is not.

Most of what makes a shop good lives in that silent part. It is also the part that is hardest to replace when a good printer retires.