AI Isn't the Problem. The Box Between Our Ears Is.
For years, I’ve had more software ideas than I could possibly afford to build. Even with a team of 5, 6, or 10 developers, there was never enough time or money.
Well, that game has changed. And I mean REALLY changed.
I’ve spent a lot of time building an AI Software Factory, using GitHub, Claude, ChatGPT Codex, and Microsoft Teams.
Here's how it works.
I have an idea. I type it into Teams. AI starts asking questions. And I mean GOOD questions. Sometimes it’s uncanny. It asks about things I hadn’t even considered.
We go back and forth. We move things around. We argue a little. We build a specification together.
A complex project that might have taken six months in the past? I can now get the specification done in two days and much of the coding done in three hours.
Not six months. THREE HOURS.
And here's where I think the association industry is headed.
Imagine taking that Excel spreadsheet with 500 requirements you prepared for your next AMS vendor demo and feeding it to AI.
If your requirements are good, AI asks a few questions. If your requirements are “I want a new AMS,” you better get comfortable. You’re going to be answering a LOT of questions.
But eventually, you could have a working AMS, with documentation, tests, and the features YOU wanted.
Want another feature? Describe it. Answer a few questions. Send it through the factory.
No waiting six months for a vendor’s next release. No endless consulting engagements. No giant annual SaaS licensing bill for software you could potentially own yourself.
And support? Increasingly AI-driven.
Now, I keep hearing, “AI just lets you arrive at disaster faster.”
I couldn’t disagree more.
Today, if I find a mistake, I can describe it to AI and often have it fixed in an hour.
In the old world? An hour writing the vendor an email. A day waiting for a response. Two weeks getting into their queue. Three more days coding, testing, documenting, and hopefully getting the fix back.
Give me the first option!
The factory does more than write code.
My software factory uses Claude to write much of the code, and Codex to review it. I’ve found Claude often writes better code, while Codex catches design and logic issues Claude misses.
They challenge each other. And believe me, they find things.
The factory runs hundreds or thousands of automated tests. It examines dependencies and edge cases we would never have had the time or budget to test manually.
And AI models? There are plenty to choose from. You don’t need Albert Einstein to tell you what 3 + 4 equals. But when you’re designing something complex, spending a few more dollars on a smarter model is a bargain.
It took me about four days to get my factory running and push the first simple feature through it.
The first two days were practically spent arguing with AI about permissions.
“Can I do this?” YES.
“Are you sure?” YES!
“Can I modify that file?” FOR THE LOVE OF GOD, YES!
But that’s actually a good thing. You establish boundaries. It shouldn’t casually change security, create users, or delete things you didn’t authorize.
And with GitHub, I have an audit trail. If something goes sideways, I can review the changes and roll them back.
Now here's what really bothers me.
I’ve attended several AI seminars in our industry, and so much of the conversation is about putting boxes around AI. Restricting it. Controlling it. Worrying about what might happen.
I’m not saying ignore security or throw common sense out the window.
But I think we’re missing the entire point.
The box isn’t around AI. The box is between our ears.
We’re still thinking about software development like it’s 2023.
We’re still assuming we need a large development team, a giant budget, a lengthy project, and a vendor to tell us what’s possible.
We don’t.
I’m not suggesting you start tomorrow by replacing a mission-critical AMS. Start small. Build something useful. Learn how the factory works. Then build something bigger.
Stop panicking. Stop listening to every vendor telling you that you need six months of consulting before you can even touch AI.
Start experimenting. Make mistakes. Fix them. Learn.
Build your own software factory.
Because I believe the next generation of association software isn’t going to come from another vendor demo.
It’s going to come from people who understand their business, know what they want, and finally have the tools to build it.
Get out of the box. Get moving!
BTW, I use AI to generate images and organize content, but my thoughts are my own.