What Building an AI Executive Producer Taught Me About Designing AI Systems

You may know I host a weekly radio program on KSVY 91.3 fm on Monday mornings. It’s a talk show averaging 4-6 guests per week. While it isn’t part of my client work, it’s become one of my favorite AI testing grounds.

Producing a weekly two-hour radio show takes far more than two hours. Recruiting guests, researching them, scheduling appearances, and managing years of records easily consumes several times the airtime. (I won’t share the crazy number of hours it takes to put it all together!)

When modern AI tools became practical for everyday work, I wondered whether they could manage something this complicated. But I couldn’t resist the challenge…

The AI Morning Show Agent

In building my AI-powered Executive Producer app, in a matter of minutes it processed and compiled:

• More than 17,000 emails over ten years of show records.
• A history of every show’s guest appearances and those scheduled into the future.
• A database of the show schedule, past-present-future, that’s searchable, shuffle-able, and re-schedule-able.

The app pulls together nearly 1,000 active guest records and more than 2,800 guest appearances from a disparate array of email records, spreadsheets and documents into one repository to manage the show.

Instead of hunting through folders trying to remember the name of someone who appeared before COVID, I simply ask my AI producer. Seconds later, I have their appearance history, contact information, and notes.

The dashboard shows me at-a-glance who’s scheduled for the following week’s show:

A pivot to the Schedule module provides a look forward at a month’s worth of shows:

The Guest module contains contact information, bios, show history and details:

There’s much more it can do… But while it took AI minutes to build my new AI-powered Executive Producer database, CRM, and scheduling manager, it took hours of careful design and preparation on the human end. My hours!

Here’s a small snippet of a lengthy prompt I used to kick off the project:

Objective/Goal: Become the Executive Producer of my weekly Monday morning radio show, 8-10am PT, on KSVY 91.3fm (ksvy.org).

1. Take over as many of the repetitive show preparation tasks as possible (as you learn them) so that I may secure better and better guests, and improve the show.
2. Seek out opportunities and strategy to help me grow the audience and reach of the show.
3. Default to always ask me to confirm steps, processes and particularly rules. (There are many exceptions to my rules because this show is run by humans!)
4. Do not jump ahead with major agent tasks without asking me if I’m ready for you to proceed…

Evaluating AI Capabilities

How did my app build go?

If I told you, “Smooth as a whistle,” you should be very suspicious!

Oh, no. True to the old saying that if something can go wrong, it absolutely will, my AI platform borked upon building my “producer” app. Not once, but several times. I learned the hard way because I discovered some very real limits in today’s AI platforms. (“Houston, we have a problem!”)

What kept your Cornucopian from a mental breakdown and excessive tears when the ship was sinking?

Experience and Careful Planning saved the day: Not only had I instructed the AI platform to create quality checks, changelogs, and redundancies to verify completion of tasks, but this built-in method ensured my new “producer” didn’t end up like C-3PO each time “he” was re-assembled. (“I’m an interpreter. I’m not supposed to know a power socket from a computer terminal.” I still have visions of “him” strapped to Chewbacca’s back in convoluted parts.)

The Long AI Road Ahead

Have I arrived at a fully-formed, Aaron Sorkin-worthy AI-powered Executive Producer?

Absolutely not.

And I didn’t expect to have my AI producer ready to act independently right away. The next step is having it (or another agent) manage the human communications, a far more complex undertaking. The next step will be:

• Evaluating inbound guest pitches and email
• Recommending possible show dates to prospects
• Sending follow up emails

Right now it’s more like a heavily supervised intern: it can write email replies, but it has to keep them in the Drafts folder for my review and enhancement. No sending allowed yet!

What Are the Benefits?

Once it’s fully trained, I expect it to cut my weekly show preparation by more than half. But isn’t that the hope for all good employees?

Once it’s fully-functional and accurate, I expect my AI producer to reduce my time on show management and preparation by more than fifty percent. Multiply that across fifty shows a year. You see where I’m headed with this?

In addition to basic administration management of my show, I expect my AI producer to manage social media and promotion of the show. With nearly 1,000 guest contacts recorded in its CRM, the producer can spot potential trends and propose show themes from the myriad guests I don’t remember from years back, improving show quality.

I hope to be asking my custom AI producer, “Review our guest history and let me know if we have an expert we could schedule for this XYZ emerging story?” or “Do you see an opportunity for two themed shows next quarter, calling upon our past guests? What do you think would make a compelling and dynamic set of shows?”

Will AI Be Taking Over Hosting Duties?

Absolutely not!

Just like your business, I’m looking for ways to improve the quality and humanity of my work with the show, not replace it with AI. Every business has repetitive work that steals time from the work only humans can do. That’s the part AI interests me most—not replacing people, but giving them more time to be people.

I’m not Jeannie.

I don’t play one on TV.

(And I don’t have her flat belly!)

But I am looking to produce ways to improve my services using the AI tool.

Building this system reminded me that successful AI projects don’t begin with AI. They begin with understanding how people work. Only then do you decide what AI should do, what humans should continue doing, and how the two work together.

If you’ve got a process that’s consuming too much time—or simply feels more complicated than it should—I’d love to explore whether an AI-powered system could help.