# What lean manufacturing gets right about AI

By Tommy Magee

Type: Writing
Canonical: https://tommymagee.com/library/information-factory

Some of my best AI lessons came from moms on YouTube.

Expanded website edition of Tommy’s public LinkedIn post. Added examples are illustrative. The source links to the original.

A surprising amount of my AI philosophy comes from moms on YouTube and Japanese lean manufacturing.

The first one is legit. People running households are constantly coordinating information, juggling constraints, delegating, and dealing with exceptions. Some of the clearest AI tutorials I’ve seen come from people trying to make real life a little easier. They know what the job is. The tool has to earn its place.

That’s a useful starting point for a company too. Before asking which agent platform to buy, understand the work people are already trying to get through.

## Watch the work

One lean principle I keep coming back to is Gemba: understand the work where it happens. A process diagram tells you what’s supposed to happen. Watching someone do the job tells you where they wait, re-enter information, chase a decision, or work around the system.

For a construction or real estate team, that could mean following a single decision from the person who spots a problem to the person who can resolve it. Where does the context get lost? Why does it need another meeting? Is there actually a decision to make, or is somebody just looking for information?

Those are useful questions whether the answer involves AI or not.

## Keep responsibility visible

I also think about Jidoka, which I translate into my AI work as automating while keeping people responsible for quality. An agent finishing a task doesn’t tell us whether it did the right thing.

We need to decide what a good result looks like, who can check it, and when the system should stop and ask. That doesn’t mean reviewing every small action forever. It means being deliberate about which decisions we’re comfortable delegating and what evidence would change that judgment.

Then there’s Yokoten: spread what works. Once a team finds a useful approach, share the reasoning and the conditions that made it work. Copying a prompt without the surrounding context is a pretty thin form of knowledge transfer.

## Think about the whole system

Japanese manufacturers didn’t simply add technology to an unchanged factory. They rethought how the work moved. I see a similar opportunity for organizations that move information and decisions all day.

The phrase I use is information factories, with both human and digital workers. The interesting part is how they work together: who has the context, who owns the decision, and how the next person knows it’s their turn.

I’d start with one real workflow, learn from the people doing it, and make the smallest change that improves it. The tools will keep changing. Understanding the work is still the part that matters.

Original LinkedIn post: August 2026 (2026-08-05)

Source: https://www.linkedin.com/feed/update/urn:li:activity:7490994059873001472/

Topics: Systems thinking, Applied AI