
Three takeaways from the AI panel at AWI National.
If you run a millwork shop, you have probably told yourself AI is a next-year problem. Something to look at once the backlog clears.
I would have said the same thing a year ago. Then I moderated a session at AWI National where three member companies showed what they had actually built, and the room was the fullest of the convention. About 200 of the people there were owners, executives, and the people doing the work. Several told me afterward that this session was why they came.
Three things I took away.
1. More shops are already doing this than any of us realized
I expected a room of curious skeptics. That is not what I got.
One owner described how deep he already is: building with his team, not experimenting. Another told me he had hired a plant manager who showed up with his own AI tools already built and started using them in his first week. It did not arrive as a budget line or a software demo. It walked in with a new hire.
The same thing happened in my own company and I did not notice for months. My estimating guys were already building their own tools, on their own time, on personal accounts. When I found out, I gave them company-paid accounts and told them to build it for real. I was not leading that. I was catching up to it.
If you think nobody at your shop is doing this yet, you probably have not asked.
2. None of them went shopping for AI
The three companies looked nothing alike: 45 people, 165 people, and 600 people, with budgets to match. One owner built his himself on nights and weekends. One spent over a million dollars across three years with an outside firm. They had almost nothing in common except how they started.
Every one of them began with a bottleneck that was already costing money.
One had crews going home thinking the day’s work was done while $36,000 of ready work sat in the warehouse and the box line ran dry. Nobody could see it. One had every drawing waiting seven days in a manual QC queue before it could move. One had a full-time person checking work behind an eighteen-person engineering team and never catching up.
Not one of them bought a tool and went looking for a use. They named the thing that was already hurting and built the smallest thing that would fix it. One panelist put it bluntly: someone told him he was going to build software to replace a $4,000 tool, and he said don’t waste your time. Go find something big.
3. The useful distinction is AI in the tool versus AI that built the tool
There is AI inside a tool: software that calls a model while it runs, reading a contract or checking a drawing against a manifest. That is what most people picture.
Then there is AI that built the tool. You describe what you need, it writes the software, and the finished program runs on its own afterward. No model involved. Nothing to subscribe to.
The clearest example anyone gave was a shop-floor tool built over one weekend that costs nine dollars a month to run and has no AI in it at all.
Most of what got shown was that second kind. Which means custom software stopped being a capital project. The thing that used to require a developer on staff or a vendor contract now takes an afternoon and some stubbornness.
What the room actually asked
We ran out of time in Q&A and people followed the panel out of the room to keep asking. Two questions came up more than any others.
How do I protect my data? Paid business accounts, so your work stays yours. A written policy and some training, even an imperfect one. As one panelist put it, it needs better evolution, but you have to have something to stand on. Private repositories, always.
Do I hire an outside firm or do this internally? All three had done both. Outsource when you have nobody internal, the scope is clear, and you need something fast. Build internally when the scope is fuzzy, you need constant input from the people doing the work, and it is core to how you run.
Three things to do before the backlog clears
None of them require you to write a line of code.
Find out who is already doing it. Someone in your building is using AI on a personal account right now, with your business data. You would rather know.
Ask your people three questions. What do you use it for, what do you wish it could do, and where did it fall short. One panelist did exactly this across his whole team, and the wish list became his build list. It was the most-copied idea of the session.
Connect before you build. Hook it up to one system you already pay for. Ask it questions. Notice what you ask over and over. That is the thing worth building.
The shops on that stage are not ahead because they are technical. One of them had been in millwork six months. Another will tell you flatly he is not a programmer.
They are ahead because they looked at the place where work was already piling up and refused to keep paying for it.
You already know where that place is in your shop. For a lot of you it is the same one it has been for years: the work is sold, the floor can build it, and it is stuck on a desk waiting to become a drawing set.
You won the work. We get it to your floor.

Jacob Edmond
CEO of DuckWorks and chairs the AWI Technology Task Force.