The Perception Layer for AI Agents

Hey everyone. Welcome back here to Techstrong TV. My next guest is a woman, Slavena Haristova, and Slavena is with a company called ABBYY, A-B-B-Y-Y, where she is director of product marketing for something called Document AI, and we're going to talk about that in a minute.

But first, let's meet Slavena. Slavena, welcome to Techstrong TV. It's great to have you on here.

Thank you, Alan. Happy to be here, and looking forward to this discussion. All right.

Before we discuss ABBYY and Document AI, let's discuss you a little bit. Sure. Give us a sense of your journey, your career path.

Sure. Actually, my entire career path has been with ABBYY. I started about 22 years ago.

Really? Yeah, I was actually still in university and was very drawn by maybe a little bit the retail side of marketing and what people think when they hear marketing, but then transitioned quite quickly into product marketing, which is a very different space. And what I think is really exciting is this intersection between marketing and actually translating technology, especially for tech companies, translating technology into something people understand and into customer value.

So that quickly converted into this B2B space that we work in today with Document AI. So, the company has made a significant development since I joined, so it's been a really exciting journey. And it's been great to see how this technology evolved and what possibilities exist today that 22 years ago, probably nobody imagined that would be possible.

Now, 20 years ago, who would've thought? Who would've thought, right? Yeah.

Look, here in the US, it's almost unheard of for someone to be at the same company 20-plus years. Yeah. So congratulations on that to you.

Let's talk more about ABBYY, though. First of all, how do people find out more about ABBYY? What's their website?

com, so pretty straightforward, and you can read all about us there. But what we are about, when I joined 22 years ago, it was all about OCR, optical character recognition and- Sure ... at that time it was almost like magic, converting documents into something that machines can read, particularly images, scans.

At some point it became very heavily focused on mobile phone images and poor-quality mobile phone images from the early days of mobile phones and not so good cameras that there were. And it transitioned into capture, extracting data out of these documents, not just converting the text into bits and bytes, basically, but extracting value and meaning out of these documents, extracting invoice numbers, extracting totals, extracting tax data. And there was a big push, if you want, when robotic process automation, or RPA, came around, because people really quickly realized that just automating specific tasks is, at some point in business context, hits a document, because documents are behind every business process, basically.

No matter how much we talk about digital transformation and digitization, there are still a lot of documents that move these processes. And we're experiencing a quite similar wave today with AI. It's a little bit from a different angle.

I have a feeling that a lot of people discovered that documents need to be processed thanks to AI, and a lot of people think that AI is beyond the thing that can do, and I say AI here in this modern sense of AI or generative AI, the ChatGPTs of this world. But we tend to say here that we've been doing AI long before it was sexy, because optical character recognition also is AI- Sure ... in a more traditional form.

And with the progression and development of our technology, we have implemented a lot of the modern type AI into our product, because it extends the art of the possible with our technology and makes this level of automation that people have been imagining much more tangible than ever before. And then agent is sort of the new frontier. It was intelligent automation, and now we have this autonomous automation of these machines that can basically think for themselves and make decisions for themselves.

But how this is related to documents is documents contain all the data. So if the agent, similar to the RPA bots, if they cannot access the data and access it in a reliable manner, either the process stops and they cannot be autonomous, so they need to call a human, or they act on data that is incorrect, so they basically automate mistakes, and create a lot of risks at scale, which we find kind of dangerous. So our mission is to help companies overcome that barrier.

Absolutely. So you're right. I remember the OCR days, and how it was sort of automagical.

Take a picture of a doc, and it would convert it to text for you. Today, with Google Translate and stuff like that, and the gen AIs, those things are getting easier. But as you're talking about, as we move from generative AI to AI agents, we're seeing yet another revolution, another change, in what seemed almost magical become every day.

Yeah. Right? And now, as a result of this, at some level, whether it's the C-level, the CIO or CTO, or even just at the buying office, where it's making a huge impact.

What programs do we buy? What things do we buy? What do we use?

How do we spend? How do we allocate budget? Now, ABBYY is obviously a company that's continually reshaping itself and reinventing itself to stay in line with the latest technology out there.

How are AI agents reshaping CIO buying dynamics? Well, for one, from what the experience is, as I mentioned earlier, it seems a lot like people have now discovered and think that LLMs are the only thing that can process documents. And it's a little bit of they're trying to move all this document processing into LLMs and to generative AI.

So, that works to some extent. It works very well with high-end structured documents, for example. That's something that digital document processing wasn't able to do until now.

But, on the other side, there is a lot of talk now about token maxing and companies wasting their budgets already in the middle of the year, and really not being able to stay in control of their cost with AI. So, one of the things that we see is that there are technologies that already existed that are able to solve certain tasks in a very efficient way, and instead of moving it all through LLMs. And it is a very similar thing with agents.

But another aspect that we see with agents, they're supposed to or expected to be autonomous, as I said. So when a buying decision is made, currently, it is a group of people who sit down, develop requirements, put budget, evaluate five, six solutions, make POCs, and so on. But with agents, it is thinkable that at some point, an agent would be able to discover different tools and services that it needs in the process as soon as it encounters a problem.

Like, I have a document, I cannot access the information in that document, but I need it to move forward, so let me see what tools I have at my disposal. So the agent theoretically could be making that buying decision. Now, for enterprises, and even more so for regulated enterprises, today it's not so realistic that the agent is going to do this completely autonomously.

But what is thinkable is that this buying group that so far has been developing the requirement and doing the POCs, would just establish the guardrails and certain compliance rules that the agents should stick, and give some level of autonomy that the agent can select the tools that are best for solving a task, whether it is based on cost, on reliability, on token consumption, whether it's based on accuracy or on speed, really depending on what the use case is and what the ultimate goal of the agent is. So, we've been thinking a lot about how agents would consume our technology because we see ourselves as this perception layer, if you want, that gives agents the ability to understand information, especially the one that is in documents, and how this selection of the tools was going to happen in the future. Fair enough.

com is the website. For people who want to take a deeper dive here- Mm-hmm ... people who want to find out more, where within the website, or even maybe at other places, can they go to kind of really explore this, peel it back a little bit, and dive in?

So we have a great resources section on our website where we post a lot- Okay ... about really current articles of what we see is happening on the market, and there's a lot of shifts happening on the market right now of what customer expectations are, what automation trends are, how agents are affecting this. " We have really great insights there, both from the experts within ABBYY, but also from our reselling partners who also have guest contributions there.

So a little bit more on the practical side of the implementations of our technology. Excellent. Sylvana, I want to thank you for coming on Techstrong TV and talking about this today.

This is the real world we're living in right now. Our coworkers are not necessarily human. The people we're dealing with...

I use an AI EA who handles all my email response and scheduling, and people don't recognize it. It is reshaping how things are. Mm-hmm.

So, it's going to be interesting. And also, again, congratulations on, what is it, 22 years at ABBYY. Thank you.

That's phenomenal. Thanks for coming here on Techstrong TV, and have a great day. Thank you for having me, Alan.

Thank you. My pleasure. Hey, we're going to take a break here.

com. We'll take a break. We're going to be back in just a moment.