AI Agents, Innovation Antibodies, and PB&J with Brian Ardinger and Robyn Bolton

On this week's episode of Inside Outside Innovation, we talk about why AI agents should speak Latin, how to avoid the agentic convergence trap, and the power of the peanut butter and jelly sandwich. Let's get started.

Inside Outside Innovation is the podcast to help innovation leaders navigate what's next. Each week, we'll give you a front row seat into what it takes to grow and thrive in a world of hyper-uncertainty and accelerating change. Join me, Brian Ardinger, and Mile Zero's Robyn Bolton as we discuss the latest tools, tactics, and trends for creating innovations with impact. Let's get started

Podcast Transcript with Brian Ardinger and Robyn Bolton 

[00:00:30] Brian Ardinger: Welcome to another episode of Inside Outside Innovation. I'm your host, Brian Ardinger, and with me I have Robyn Bolton. Robyn, hello, how are you today? 

[00:00:47] Robyn Bolton: I am great today. How are you? 

[00:00:49] Brian Ardinger: I am doing well. We are coming off our all directors meeting this week for Nelnet, where we brought in about 250 of our directors from all over the world to talk about AI and other things that are going on in the business, and it's been quite eye-opening. So that's where I've been. I hear you're out on a beach. 

[00:01:05] Robyn Bolton: Yeah, I'm kind of doing the opposite of your highly productive week. I'm being highly unproductive and am on vacation here in Turks and Caicos. So, I just dragged myself in from laying poolside so that we could talk about innovation, because there's some really, really cool stuff going on.

Why AI Agents Speak Latin

[00:01:21] Brian Ardinger: We'll get you back to the beach very shortly, but I always appreciate your insights. We will jump right in. The first article on the list is from Tristan Kromer. In his blog, he talks about why my agents speak Latin. I thought this was a very interesting take by Tristan. Tristan, if you've been following the podcast, he actually came and spoke at the IO Summit most recently.

He was talking about, in this blog post, about how he is doing a lot more coding and creating AI agents, and he has kind of a unique way to see if the agent is going off the rails. So, what he's done is he creates these personas for his agents, and one of them he calls Cicero, and has it speak to him in Latin phrase to give it a personality.

And what he's found, by giving his agents particular personalities and quirks, that they actually come back, and over time we do know that AIs tend to lose the context. You know, they tend to drift over multiple rounds of dialogue. He says by incorporating this little tick of a personality into his agent, he can tell when the agent is drifting to the normal way because it stops speaking in Latin to him, or it's, it drops phrases, or does a couple of other things that make it known that the context is getting too full and the agent's going off the rails.

What I found so fascinating about that, not only the fact that it actually goes off the rails, but the fact that it was a way to give the layperson a way to understand if the AI was not doing what he had originally programmed it for. I just thought it was a great insight into it.

Using AI Personas to Detect Context Drift

[00:02:47] Robyn Bolton: Fascinating. It's funny, because Tristan and I were actually talking about this at the IO conference over breakfast the morning of the conference, and he mentioned that his chief of staff spoke Latin. And we didn't have a chance to get into why that was, so it was great to read this article, and it was also incredibly timely 'cause a couple of weeks ago I had decided to make the jump into using Claude Code and VS Studio and trying to get into the vibe coding.

And I told Claude Chat, "Hey, you are an expert in blah, blah, blah," and pretty quickly I noticed the chat lost the thread. And was just being chat and not being the expert and the coach I had asked for, and which then led me to typing in all caps as I usually, is usually how our chats end up. And it was just interesting to read in this article the research about why that was happening, what it is kind of in the background that causes any of the chats to lose that thread and then this kind of little safety net, we put it in there. So great, super actionable article, and now I'm thinking of all the personas my little agents can have. 

[00:04:00] Brian Ardinger: He was saying that, you know, when the tick disappears from the output, you know, the persona has left the building as well. So you'd kind of treat your quirk as your instrumentation panel and your dashboard. Quite clever. 

[00:04:11] Robyn Bolton: Yes, very clever.

Beware the Agentic Convergence Trap

[00:04:13] Brian Ardinger: All right, the second article I want to talk about today is from HBR, and it's talking about beware the agentic convergence trap. And in this case, they talk about three case studies that have recently, in 2024, a federal class action named six major hotel chains, Hilton, and Hyatt, and Marriott, and the gang, of alleging that they had shared AI pricing platform that had produced coordinated room rates across competing properties.

The hotels had never communicated directly, but the DOJ and the FTC called it price coordination. A second example of this kind of convergence is a regional grocery store chain replaced its human promotions planners with an AI system and trained it to look at all the same public market signals every competitor uses, and within two quarters of this promotional calendar had converged with the market leaders, and shoppers could no longer articulate why they preferred one chain over the other.

And then a third example, they talked about a national landlord had adopted an AI rent optimization platform and used by competing property managers, and it raised the rents in step with rivals it had never spoken to before. And so the US Department of Justice was looking into that as well. What I found fascinating, obviously, about the individual case studies, but the fact that relying heavily on AI as your determination of your strategy has some additional potential downsides if AI is converging the thought and converging the output across particular industries.

Why Human Judgment Creates Strategic Diversity

[00:05:38] Robyn Bolton: Two big thoughts as I was reading through this. One was I just loved this quote from the article. You know, it's kind of asking why. Why is this convergence happening, and why haven't we seen this before? Because everyone's using all the same data. And, you know, the quote, "Human judgment introduces natural variation. One manager overrides the suggested action. One team moves slowly. One regional leader reads the market differently. Those frictions, often regarded as inefficiencies, are the mechanism that produces strategic diversity."

And I think this, as we've talked about in past episodes, is why we need human in the loop, why humans are still valuable, and it made me think of so many companies I've heard that are using AI to brainstorm new ideas.

And, you know, the caution is, like, if you want radical innovation, if you want something radically new, AI isn't going to help you. But this is also a caution for you're going to get the same ideas as everybody else because you're sourcing from the same kind of data set, and it's th...