So if deterministic logic, AI agents, and humans each have different strengths, what does the ideal intelligent workflow look like?

It's a combination of all three, and you need to be able to ask yourself, like, hey, how predictable is this task? What's my tolerance for risk or error? How fast does it need to be? How much money am I prepared to spend on it? And if you ask yourself those questions, you'll figure out at that point in time whether or not you should be using an agent, whether or not you should be using deterministic logic, or whether or not you should be using a human.

So if you've got something that has to be correct all the time, and you know there's a correct way of doing it, you should be using deterministic logic.

If you've got something that has to be done exceptionally quickly, you should probably be using deterministic logic.

If you've got something that needs company-specific judgment, You should probably be using a human being to make that decision.

But if you've got something that like there's very broadly a right way of doing it, or it's an unknown process that you may not have seen before.

Agents are incredibly good at doing that and you should be using an agent.

I found that fascinating.