I did something in two weeks that they told me was going to take up to a year. I was working with a YC company in a highly regulated industry. They came to me and said they needed to get licensed in as many states as possible by demo day.
So I did what I do for everything these days. I decided to build an agent out of it.
We had 33 states. All of them required different paperwork, different methods of submission, different portals. What made the agent useful wasn't that every state was the same. It was that the operating loop was the same even when the requirements changed:
Humans kept the judgment and legal review. The agent handled the repeated coordination.
The agent had its own email address. It had its own 1Password account. It created all the portal logins that we needed, because even that itself was something like 19 or 20 different government portals. Such a nightmare.
And some of them required a notary. This agent actually booked a mobile notary here in the California office, and also in Texas with our pharmacist. The coordination wasn't something I had to touch.
It had been four weeks since we started. We already had two licenses issued, and then we had eight. Now we actually have a process we can use with the next entity. If you're building a similar AI loop for repeated tasks, this is exactly the pattern that works.
At the time of filming, two licenses had become eight in four weeks. Since then, we've submitted in all 33 states.
This is the perfect use case of AI. It takes so much time and so much minutia. I have ADHD, which means I'm not great at that kind of repetitive tracking. But if you give an AI agent a memory system and point it at a fragmented process, it will figure out how to solve this.
That's what AI is actually good at. Expensive, fragmented processes that repeat with different fine print. The requirements change, the forms change, the portals change. But the loop stays the same.
Build the agent once, and it handles the coordination forever.
