Intelligent Relations Separates AI Agents From Bots

Hello, and welcome to the latest edition of the Techstrong. Leadership Insight series. I'm your host, Mike Vizard.

Today we're with Stamatis N. Astra, who's the chief business officer for Intelligent Relations, and we're having a little chat here about, well, what is an AI agent and what is not? Because it turns out there's a lot of AI washing going on.

Stamatis, welcome to the show. Thank you. A pleasure to be here.

Thank you very much. In your mind, what is exactly an AI agent? What should it be considered as?

Because to the point, we are seeing the term tossed around a lot a bit, but not all these AI agents are, shall we say, fully autonomous, and a lot of them just seem to be kind of bots that somebody created, but is that really an AI agent? No. The answer is no, and I think the word that we're looking for is exactly what you said, autonomous.

We have evolved on the AI thinking. A couple of years ago when ChatGPT started, it was generation of text. You put a prompt, and then you gave generation of text, and you have all these wonderful things and images and creation based on prompts.

Then it became workflows, so you started kind of connecting prompts together. Very simply put, you had different workflows, different steps, and in a sequence, one agent, as they start calling them, or one prompt gave what the outcome was to the next prompt, and as a sequence, it completed those tasks. Helpful, very good, and all that stuff.

And now we are in the next frontier, which is setting goals, and instead of telling them a prompt, write a poem, read a book, give me a summary, analyze the spreadsheet, and then instead of just giving that prompt and then take them, analyze the spreadsheet, and go to my CRM and send emails, instead of that stuff, you give them specific goals, that they're based on the skills and the memory that you have built on the AI. And those goals can be lofty, can be important goals, depending what your business is. It can be business development goals, it can be research development, it can be research, or it can be any other type of goal your company has.

And the goal of the goal is to autonomously, continuously work on these goals based on the information on the internet or on the other systems you have into your company, and give you results and give you specific outcomes. And it's up to you to decide, okay, act on those outcomes, or wait and approve or change direction or do something like that. So one of the issues you keep hearing about is that a lot of the tasks being assigned are supposed to be deterministic, and they're pretty much done the same way every time.

And then we have a bunch of AI agents that are based on probabilistic technologies that never do anything the same way twice. So how do we kind of meld these things into these workflows that we're trying to accomplish so that the output is the same every time? That's the fear, and that's the problem, and I think this is the next generation of hallucination, what you just called.

Before, it was AI will make something up. AI's job is to answer your question. If it didn't have the answer to your prompt, it just made something up.

Right now, you give those goals, and the problem is exactly what you said. It doesn't mean they're going to complete those goals based on the common sense, but based on what they can or they cannot do, and change the outcome. That's why a lot of people are worried, and the guardrails and governors is still very high.

Most of the people that are using those to the full extent as it is right now, they still want to approve everything. They want to make sure that whatever happens is based on what they thought. And it also feels like the AI agents are designed to aggressively accomplish their mission, and so they will ignore guardrails, or they'll find some other way to do something, even though I may have put controls in place, or at least I thought I did, and then it turns out the AI agent found a way around those controls, so- Yeah ...

do we kind of need to revisit the way that we have architected the IT environment and the data access and everything and design it from the ground up for an AI agent? That's very important. That's a great point.

Again, the fundamental issue of the LLM, of any LLM is to finish, to do what they have to do. So that's the importance of putting enough guardrails. " That's a common thing that will people put because LLMs, they tend to flatter the user and say, "This is great input," and all that kind of crap.

It's kind of the same problem that LLMs will continue working around. If you have a task that says, "Don't follow up on this prospect because don't follow up on this prospect because they have a specific request, they will try to find a reason to do it. The bigger question that you say in the whole architecture of the IT, this is where everybody in the software-as-a-service industry is running around the building, and they're trying to figure out how to deal with autonomous LLMs.

A good example paradigm is, I'm old enough to remember 20 years ago, you remember we all had Excel files, and we have different files, and Word file, and then we're passing information from one file to another, and we felt super productive. And then the software-as-a-service industry started, and it kind of started automating all of those things. Salesforce and all of those huge companies now, at the end of the day, they automated the spreadsheets, and that created a whole new environment for security, for infrastructure, for storage, for memory that came with that.

It will definitely be a completely new infrastructure and development on those very, very tactical levels where the rubber meets the road, because from chasing spreadsheets around, we are now chasing chats around. I have 30, 20, 50 chats, and then I'm going to have 30, 20, 50 agents in my operations, and then I have to chase them around. So that will happen.

Is this at the root of the somewhat disenchantment with AI right now that exists in a lot of enterprises because they are discovering all these issues and discovering the fact that the data fundamentally is a mess, and while the computer science may be good, the reality is somewhat different, and we're just having a disconnect? There is a disconnect, and I think there's a pendulum. So right now, we are on the one end.

" And that's a fact. Everybody who was a vendor, or they wanted to sell a software to any big enterprise, the enterprise said, "Thank you. Great idea.

" And the reason was people are thinking, they thought, and they're still thinking that the build versus buy argument became different. It's like, I can build it so fast now. I have these tools that I can build anything so fast, so there's no reason to buy it.

So the de facto answer for the enterprise is, I'm not going to hire any junior person for now, make an agent, do an agent, Mr. Manager, and show me an agent cannot do it, and then you may hire someone. That's where we are today.

And the same thing with software vendors. Don't renew the annual contract. Don't change anything unless you prove me you cannot build it yourself.

And if you cannot build it yourself, get an agent to do it for you. So that's where we are. Those are big enterprises with still a lot of management in their gross margin, bottom line, every dollar counts.

Those are serious people. There's a lot of FOMO, fear of missing out, going around. Everybody's hearing what the other guy is doing on AI, and there's a lot of anecdotal evidence, and the anecdotes are not data.

And there's a lot of fear right now that they are missing the boat on the productivity side, not to mention the boat of the hardware side. So I expect that it's going to take, again, at least another six months. The other thing we've got to point out is the timelines have changed.

It used to be for any enterprise, six months to a year to make decisions, to make strategic changes, to make all these things. All that has changed. People are making changes in two, three weeks, big changes, because they're seeing it move so fast.

Claude Design came up, and everybody who was in the space was terrified, and they're afraid what is the next one. Do you think we're looking at maybe the beginnings of some massive digital business transformation? And I ask the question because a lot of the SaaS apps we have, or whatever it is, are built around a function that was created around the software and the way the software worked.

But maybe in the age of AI, the software will bend more to the way we want to work, and this whole notion of an order-to-cash process needs to go through 10 different applications will fade away in favor of something that is an AI-driven order-to-cash process. It will. That's where we're going.

The question is who's going to be the winner and who's going to be the loser. Technology changes. There is going to be some people that are going to be on the bleeding edge, and there's going to be some winners 20 years from now.

So again, I tend to look in the past. We used to have the search industry that was going to change everything. Remember Lycos?

Remember Yahoo? Mm-hmm. Then Google came around, and then after Google, everybody says, "That's it.

That's the end. " And then Facebook came up. And then after Facebook, the next version of blockchain and things like that came up.

So now we are in the AI phase, and 20 years from now is going to be some players that we haven't heard, and it's going to be some players that were admire now probably are not going to be here or they're going to be bought. Nobody knows, but one thing we know, the traditional software business is going to change, and it's a fundamental thing. You used to need a software engineer to give them product requirements, and the people started to code.

And right now, with AI, you just tell the system what you want, and now you give them the goals, and it's going to evolve to be much more powerful. But you just tell it what you want, and it does it. And that fundamental is going to stay.

That is not going to change. How the business is going to adapt, and obviously you cannot have $20 a month ChatGPT products and sustain trillions of investments all the way to-- So there's a lot of question marks, both on the hardware side and obviously the software side, and we'll see where it is. But again, you look at the trajectory of technology.

I have to go back in history when barcodes came. Nobody expected the barcode to be such a huge productivity boost in the GDP of the economy. " Again, that was a long time ago, and it changed everything.

I think we are on the same type of change in the economy. The growth and the productivity we're going to see from AI, and on the GDP accelerator of productivity is going to be tremendous. So to your point, we're going to describe our intent, and the agents will go and execute said intent.

But what happens when we have intents that are diametrically opposed? So let's say I am the seller of something, and my intent is to sell that at the highest margin possible, and I am the opposite of that as a buyer, whose intent it is to acquire that thing at the least amount of cost possible. So will these AI agents at some point battle it out somewhere, or will they eventually just cancel each other out and call us for help?

That's the question of open source. Are we going to have agents on the open source so they can eventually talk to each other? It's going to be an internet type of open source of agents, so if you're selling widgets in Kansas, and I am in Europe, can I see that?

Or it's going to be closed systems like we are today? Nobody knows. I'm sure it's going to be some in the middle, and we'll see what happens.

But I think at the end of the day, survival is what tends to win, and I think companies and businesses are going to opt for survival than aspirational and humanitarian open-source goals. What's your best advice to folks, or what do you see organizations doing today that just makes you shake your head a little bit and say, "Folks, we need to be a little bit smarter than that"? I think my best advice is don't fear the agent.

You can't. You cannot fight it. You can't.

Think again how you're going to use it. You're terrified about security, you can be terrified about autonomous mistakes, that things are going to happen, and all that, but don't fear the agent. All right.

Well, folks, you heard it here. AI agents are here to stay, so now it's a question of how we're going to live with them more than how it is they're going to live with us. Hey, Stamatis, thanks for being on the show.

Thank you. A great pleasure. Thank you.

Excellent questions. Excellent time. I'm thinking about it.

Thank you. AI Leadership Insight Series. You can find this episode and others on our website.

We invite you to check those all out. Until then, we'll see you next time.