3191: Exploring Agentic AI: ImageSource's Take on Hyperautomation

[00:00:04] How is AI-powered hyperautomation reshaping the future of business operations? Because as industries race to harness AI's potential in this AI gold rush of sorts, a new era of intelligent automation is emerging. One where AI doesn't just assist but actively collaborates with humans. But the bigger question behind all this is what does it all mean for businesses?

[00:00:33] How can business leaders navigate the opportunities and equally some of those challenges that it presents? Well joining me on the podcast today is Randy Weakley, Chief AI Architect at a company called ImageSource. And they are at the forefront of intelligent automation and enterprise content solutions. And my guest has a fascinating background spanning aerospace engineering, NASA

[00:01:00] and over a decade leading AI initiatives over at ImageSource. But today he's going to be bringing his unique insights into the transformative power of AI. Explore how natural language processing or NLP combined with computer vision are revolutionizing document understanding. So I want to dig deep on some of those real world use cases across industries from finance, healthcare and government.

[00:01:29] And yes also talk about agentic AI, how that is dominating the business landscape this year. And ultimately how you or your business can balance autonomous decision making, but most importantly with human oversight. So what does human in the loop AI mean for operational efficiency and ethics? And where does my guest see the most exciting opportunities in the next three to five years?

[00:01:59] Well with that scene perfectly set, it's time to get today's guest onto the show. So a massive warm welcome to the show. Can I tell everyone listening a little about who you are and what you do? Absolutely. Thank you, Neil, for having me on. It's a great honor. My name is Randy Weakley. I have, I started my career with a degree in aerospace engineering. I worked for NASA for a few years and in the flight design area where I worked, it was really all just computer stuff.

[00:02:28] And I fell in love with computer programming and building something from nothing on the computer. So I kind of moved away from aerospace, went to work for GTE government systems for a while, MCI, went to spend seven years at Oracle running some product groups there, a couple of smaller enterprise content management companies. And then I have been at image source for the last decade. And image source has been around for about 30 plus years.

[00:02:54] We've got a fantastic suite of enterprise content, intelligent automation, customer experience products. And for the last six years, but certainly the last couple of years, we've had an increased focus on AI and integrating and leveraging these fantastic capabilities within our product suite and then getting that power out to our customers. And so I lead that effort as our chief AI architect. Well, thank you so much for joining me on the podcast today.

[00:03:24] And every day on this show, I try to demystify some of those big tech buzzwords we keep hearing about and try and understand the actual business value that it can offer. And image source, image source appeared on my radar when I was reading how you're leveraging AI for intelligent document processing and automation, which is a huge topic right now and a lot of interest from enterprises around the world.

[00:03:49] So just to set the scene, can you give me a little overview of how AI, particularly natural language processing or NLP and computer vision, can transform document understanding across so many different industries? Because there's a lot of value out here, isn't there? There's a ton of value. So in our industry, 25, maybe 30 years ago, we were all saying, oh, paper's going to go away. We're moving to the electronic office. And we were so wrong. Paper has grown.

[00:04:17] And so the need to understand content that's coming into an organization in paper form or maybe email, and maybe it's just not very computer accessible is more than ever. And so we use technologies like computer vision. And specifically here, I'm talking about optical character recognition, OCR or ICR for handwriting and content classification. Now we've been, the industry has been doing this for decades.

[00:04:46] We've been doing this for over 20 years at image source. And so this isn't this, the concepts are not necessarily new, but back in, I think it was around 2019, there was an inflection point in this, in these capabilities. AWS announced and released their text track product roughly the same time, Azure released their form recognizer, which is now document intelligence.

[00:05:10] And these were really raising the bar for these OCR, these computer vision types of capabilities. So that was very exciting. Certainly changing the way we are making sense of content. Now you mentioned NLP. Now we've also been doing that for a long time as well. And of course, the huge inflection point for that was a couple of years ago with generative AI.

[00:05:35] And what we're finding is that these types of capabilities, generative AI are fantastic at extracting information from particularly unstructured documents. And also as well as, as generating content from, from that, from that document.

[00:05:54] So huge leaps in, in the, the, the computational linguistics capabilities like speech recognition and translation and named entity recognition, sentiment analysis, all of these types of fantastic capabilities are really are, and will continue to transform a document understanding. And you mentioned removing printed paper from the office. And I think every year we see self-proclaimed futurists making predictions.

[00:06:21] Very few of them predicted that big gen AI moment three years ago. And if we go, what, five, 10 years ago, everyone was predicting a paperless office and the, the death of email, but those things are still here, right? They are, they absolutely are. And of course, this year we're moving to a gentic AI. That's the big trend at the moment. I think Gartner predicted it would dominate 2025 late last year. And there's been a lot of hype around it.

[00:06:47] And ultimately a gentic AI and actively collaborates with humans rather than just assisting them. And it's gaining a lot of attention right now. So how do you see this shift impacting everyday business operations? And what kind of opportunities, what kind of opportunities does it create for organizations? Yeah, you're right, Neil. A gentic AI is definitely going to be the topic for 2025. And we're, I think we're really in the early stages here.

[00:07:14] This really just took form late last year or maybe mid, mid last year. And so we're, we're building software around these and leveraging these technologies and proof of concepts with some customers, but we're, we're still really early on.

[00:07:29] And I believe that much of the power here with these agentic AI capabilities is in the layer that the, that the foundational model vendors are building on top of the LLMs here, specifically these cognitive architectures, these reasoning capabilities that much more closely approximate how the human brain goes through the process of, of making decisions. So we, we take a question or a prompt and we, we look at that and we break it down into chunks.

[00:07:59] And then we try to solve each one of those. Now, the really interesting thing about what these capabilities can do today is they can re-execute if necessary. So a year ago, you, you could ask a question, chat GPT, whatever, and it wasn't uncommon for it to kind of paint itself into a corner because it was kind of a one shot deal. And then if it, if it got into a corner, sometimes it just hallucinate its way out.

[00:08:24] But with these, these cognitive architectures, they have the ability now to understand that they're at a stopping point and they, they back up and they re-execute down a different chain of thought or tree of thought type of, uh, of processing to increase the quality of the outcome.

[00:08:41] So within our enterprise, when we look at these capabilities and I'm going to get down in the weeds here just a little bit, there are two specific things that I'm excited about that I think are really going to be the fundamental change. And one of them is structured outputs. So this is the ability to instruct an LLM that you want a canonical form of output, a deterministic outcome.

[00:09:04] So this, this guarantees that when I ask it a question, it's going to give me back the same structure of let's say JSON every time. So this, this helps us on our side as we simplifies the prompt engineering. So we don't have to write all the, the prompt to make sure it spits it out correctly. The other area, and we've seen this before in some of the public cloud models where we have given it a document and said, Hey, can you get the social security number out of this? And it comes back and says, no, I can't do that for you. Now there's a ways around it.

[00:09:34] You can say, Oh, well, if you happen to find a value that conforms to this pattern and you give it the traditional social security, it'll, it'll get it. But with the structured outputs, it now has the ability to tell us if it's refusing to give us something based on some type of safety-based protocol. So that that's helpful to us. And then secondly, agentic functions they're called. So this allows an agent to access external data.

[00:10:00] So this isn't, this is a data that the model was trained on or pre-trained on or even embedded. This is real-time access to information, line of business system data. So, you know, these, these types of capabilities are going to be a game changer for organizations for sure in 2025.

[00:10:19] And I think it's worth highlighting that AI driven automation is already making ways in industries far and wide from finance and healthcare to government, for example. So just to bring to life the kind of impact this tech is having, are you able to share some real world use cases where image source with your solution? You've maybe delivered a real measurable impact because I think there's so much hype around the technology and there's a real focus now on the ROI of every tech project.

[00:10:49] So I think it'd be great to bring that to life a little. So you're absolutely right, Neil. And there's, there's a lot of honestly discouraging statistics about the percentage of companies that have tried POCs with AI and then a lower percentage that actually make it into a production. So yeah, one of the examples that I like to talk about, and the reason I like to talk about this is because it's, it's very familiar to everyone.

[00:11:14] We've all been on calls where say this call may be recorded for, you know, quality assurance or training or whatever. So, so this has happened to all of us. So we have a government agency in Alaska that we've been working with for several years and Alaska is kind of unique among the states in that their constituency is extremely spread out. And many of them don't even have internet access or smartphones or things like that. So things that we take for granted.

[00:11:43] So they've had to get creative about how they deliver services to constituents in these areas. And one of the ways is to allow them to call up on the phone and apply for state benefits. So we're not talking about like an IVR system. We're talking two people talking. And so what, what typically happens is the, the agency representative reads through a script and the person on the other end attests, this is me.

[00:12:11] And yes, I would like to sign up for this program or benefit or whatever it is. And by the time they finish talking about the weather or whatever's going on socially, and then they go through the script and get the information they need. These calls are 30 to 45 minutes long. And the organization that we were working with had thousands of these queued up because they are required by law to audit a certain percentage of these. And given their small staff, there was just no way they would ever catch up.

[00:12:38] So we've delivered a solution to them that automatically sees and retrieves these calls. So they do use a cloud-based service that records the call centers phone conversations. And we retrieve those for them. We use AI to perform speech-to-text, to create a transcript of the call. We use AI then to intelligently locate these areas of the call that are important to them, the attestation area being one of those.

[00:13:06] And there's some other information that we extract out as well. We then clip that audio at that point, and we store that along with the full recording and with the, the full transcripts, which is now full-text searchable, as well as having that metadata, which you can search on as well. So we've, we've taken something that may have taken 30 to 45 minutes and reduce that now to something that takes them one to two minutes to perform those audits.

[00:13:34] So they're going to have a challenge catching up, but I think we've, we've given them some real power to help, help, help them do that. Wow. That's incredibly cool. And of course, autonomous autonomous decision-making clearly from your examples, that can make significant benefits and measurable improvements. But of course, on the flip side, it can sometimes introduce risks too.

[00:13:58] So how should businesses approach things like governance, security, and ethical considerations when implementing AI-powered hyper-automation? And sorry to ask that question, but as an ex-IT guy, I feel drawn back into that world where we're introducing these new technologies and trying to be the guardian of the network, et cetera. You're not alone, Neil. So when we, when we talk at conferences or we attend conferences or speak, or we work with our customers,

[00:14:26] that is the most common question that comes up when we start talking to organizations about AI. How do I go about data governance and security? So what we have done is come up with a six step, what we call the AI empowerment blueprint. And there are key steps in this framework, this blueprint, where we work with an organization's, the appropriate people. So whoever that is. So that, that could be a CISO, IT security.

[00:14:56] They may have a AI ethics board, or maybe even an AI compliance officer. And if they don't already have an AI governance framework, we can help them to start building that. Now that's very, very much up to them and what that looks like. And the contents of that particular framework, if you're in a highly regulated industry, that of course gets much more complicated, but that is absolutely a key step in our blueprint. When we work with customers is, is to start establishing those.

[00:15:25] And Hey, this is not going away. Every organization is going to end up having to deal with this. So now is a great time to, to dig in, start thinking about a framework. Maybe a, if you don't have an AI compliance officer or a group within your organization that does that, now's the time to start thinking about this. And we're hearing copilot got dinged recently. There's people that were given access.

[00:15:50] We call it leaking to information that they shouldn't necessarily have because the copilot itself was given access to some, some documents or information that certain groups of users shouldn't have. So securing that the information that you're using for your fine tuning or your embedding is, is absolutely key. And then these, these functions, agentic functions that I mentioned earlier,

[00:16:15] obviously super, super powerful, but those are going to have to be secured as well. And when we talk about hyper automation or automation or business process automation, we still firmly believe that a human in the loop is absolutely necessary. So there's just things that the AI does that you, you can't necessarily predict every time we'd like to.

[00:16:39] But you have to have a human take a look at that certainly before you affect any, any type of transaction at all. And one of the other things that has come out recently that will affect us, we have an RPA or robotic process automation product. And these technologies are certainly going to disrupt that industry. Late last year, Anthropa came up with a product called computer use and opening.

[00:17:03] I just recently announced a similar product called, I think it's called operator where these tools actually use your computer. So they're opening browsers. They're logging in with a set of credentials that you've provided. They're searching for records, finding records, maybe grabbing stuff out of a spreadsheet and pushing that into a system and hitting the save button. So when these things start coming online, those are going to be absolutely imperative that are considered within any type of AI governance framework or security offering.

[00:17:34] And as a result of everything you just mentioned there, I think it really puts it into the, really strengthens the case for human in the loop AI. Why that remains such a critical part of ensuring effective and ethical automation. So again, with all that in mind, how do you image source balance automation with human oversight? And why is that balance so essential? Yeah, balance is very interesting.

[00:18:00] So a lot of people talk about this in terms of the jagged frontier of AI or technology where we see this incredible capability. So I think it was maybe GPT-4 early days, a couple of years ago. It scored like in the 90th percentile for the bar exam against human test takers. So just mind-boggling capabilities from a generally trained model.

[00:18:27] But then there's been over the last year or so, these memes that kind of come out like you ask chat GPT, how many letter R's are there in the word strawberry? And it comes back and says there are two. And it'll argue with you that there are only two. Now they've pitched fix that and some other little weird issues. But I think you could still trick it with possessions or some other words.

[00:18:48] But the point is, we've got this extremely powerful capability that just performs mind-boggling feats of processing. And then you ask it something that maybe a third grader would know, and it just falls on its face. So this balance is very important in how you create a system that you can trust.

[00:19:11] And so as we work with customers, we are finding and we recommend that you still employ codified workflows. So workflows that are defined, they're repeatable, they're auditable, and they can deliver these deterministic outcomes. So we provide tools and lots of other products do the same where you can leverage these AI capabilities at just the right point throughout your workflow.

[00:19:35] So it's not just handing over the keys to the kingdom to ChatGPT or Claude or whatever it is, and just letting them make decisions and update financial transactions and human resource records and things like that. And we have the motto, trust but verify, the old saying. But what that means is there has to be a human in the loop. And so you're absolutely right. Balancing this is critical.

[00:19:59] And the way that we see that, at least for the short term, is to put a human in every loop. And as we race through 2025 and AI-powered hyperautomation continues to evolve in front of our eyes, I'm going to ask you for a prediction of sorts now. Where do you believe this will be most transformative on business operations over the next few years? It feels like there's so many different opportunities. There's a lot of excitement in so many different areas.

[00:20:29] But where do you see it delivering the most transformation and delivering the most value? Yeah, so I think there's kind of two sides. Even just when you say few years, this stuff moves so fast. I think we've got a really short term here where we're seeing AI capabilities applied to automation, which is a bit evolutionary.

[00:20:51] Not that it's not significant, but it's like I mentioned earlier, where we are interjecting AI at just certain points in our automation pipeline or workflow. And our industry has really democratized these superpowers. But the need for this structured orchestration is going to continue. And using it in this way, what I think we'll see in the short term is we're going to see quality of outcomes go up.

[00:21:20] So we're getting better results. The time to value, the time to create these solutions within an organization is going to go down. So these tools are much more easy to use and deploy and monitor and evolve. And if you're thinking about this, this is a great time to re-engineer your processes anyways. And so just the fact that you're going back and taking a fresh look at old processes, that's going to create some value as well.

[00:21:47] So that's short term, but in the longer term, I think what we will see is companies starting to use AI to actually innovate. So creating completely new products and services. So this isn't transactional type stuff. This is very future looking ideations, helping them design that, of course, coding, coding software, software development. This is a huge focus for AI right now.

[00:22:17] As a developer, the tools are fantastic. I can tell you that they are really transforming the software development industry as we speak. And so, yeah, I think these are really going to have some transformative effects in the near term and long term. Incredibly cool. And, of course, implementing AI at scale presents numerous challenges.

[00:22:44] Especially when it comes to security, ethical frameworks, all the things that we mentioned earlier in our conversation. So I always try and give my guests, all my listeners and business leaders, et cetera, anybody listening, a valuable takeaway. So for those people listening, maybe inspired by some of those opportunities this technology can unlock for them. What would you say are some of the best practices that you'd recommend to organizations looking to integrate AI responsibly and effectively?

[00:23:14] Yeah. Yeah. So I think it kind of gets back to what we talked about before where you adopt a framework. And there's a lot of them out there. A lot of companies doing great work with these types of frameworks. But somebody that's been through this journey before and they've got the bumps and scrapes and scars to prove it are going to be invaluable to your organization.

[00:23:38] So bringing on maybe a consultant to help you build this competency inside or just relying on some external commercial software, build versus buy. But working very closely with your CISOs, whatever that appropriate group is within your organization, creating that AI governance framework or working within the bounds of that framework.

[00:23:58] And considering that these large language models are just tuned on generic information, they're not going to provide you real business value at scale unless you give them access to your enterprise data, which is something that needs to be thoughtfully done. And so these models and assistants and things that you're creating to work within your organization, you need to have a similar focus around security for them as you would just an employee.

[00:24:27] And then another thing to consider, as we've worked with some customers scaling from a proof of concept into production, longer term sustained operation is to build a competency around ML ops, so machine learning operations. So this is going to be important for larger organizations as it's unlikely that a large organization would standardize on a single model.

[00:24:52] So you may have, and then even if you did, you'd still have the need as those models evolve to make sure you're performing regression testing as you're upgrading your models, you're not breaking stuff and your outcomes are not, your quality is not going down. So there's a lot of reasons you would do this, but the organizations we work with frequently will use multiple models. And so establishing some processes around how you're going to handle those evolutions of these technologies is very important.

[00:25:20] And then lastly, monitoring and auditing these is going to be key as well, both from a security perspective and a performance and reliability perspective. And one of the most dominant themes of our conversation today has been the speed of technological change and how quickly AI is evolving. And I do think there's a real pressure on each and every one of us to be in a state of continuous learning.

[00:25:45] So to continue on that theme of giving the listener a valuable takeaway, where or how do you self-educate? How do you keep up to speed with this stuff? Any tips you could advise there? Oh, that's a great question. It can be fatiguing. And I don't know whether it's because of just the time in which we live and the availability of information and whatnot. But Gen. Gen. AI is talked about more than any other technical advancement in my career.

[00:26:14] And so it can be overwhelming. And so for me, I've had to really tighten up on what I read, what I listen to, and how I learn. So I think this comes down to an individual preference, but I think it's important to kind of self-evaluate and figure out how you best learn.

[00:26:35] So for me, I've gone to short audio, short podcasts, probably 30 minutes or less, or I go to do hands-on tutorials. So I either need that kind of short hit or I need to get my hands and roll up the sleeves, get my hands on a technology and start playing with it. Just throw a shout out. One of my absolute favorites is AI Daily Brief with Nathaniel Whitmore. If you're looking for just a quick hit on AI specifically, it's fantastic.

[00:27:04] And then, like I said, I think like a good innovation program, you need discipline. So you can't just expect to read what comes through your inbox. I would pick a time to do it and pick a few authors or gurus that you like to listen to. And things aren't resonating with you. Just move on. There's plenty of stuff out there, lots of material. Oh, and for me, it's easy to get distracted too. So I have to be careful.

[00:27:32] Sora comes out with their latest version or whatever it is. That's not necessarily something that we deal with on a daily basis in our industry, but it's super interesting. And so I kind of have to discipline myself to kind of stay focused on things that are really going to be material to helping my customers and developing really great solutions. Yeah, I completely agree with you. Podcasts that give you that quick hit of information while you're walking the dog or on the commute to work. And I would also say if you find any of this stuff intimidating, just have a play with it.

[00:28:02] There are so many YouTube guides that will just point you in the right direction and help you get to grips with some of that technology. I think we learn so much more by just simply playing sometimes. But thank you so much for sharing your insights today. But before I let you go, anyone listening, they want to find out more information about ImageSource or anything we talked about today. Where would you like to point everyone listening? You bet. So they can hit our website. That's imagesourceinc.com.

[00:28:30] We also put a lot of great stuff up on LinkedIn. I'm on LinkedIn. If you want to reach out to me, we do monthly webinars. And many of them are not product specific. We have industry experts that we bring on occasionally. And so they're very useful just in general. And then if you're a customer of ours, we have a customer community that uses our own customer experience platform. Well, I will add links to everything to make sure everyone listening can find you nice and easy.

[00:28:54] And as I said at the very beginning of the podcast, there's been a lot of hype around agentic AI this year being a big buzzword. But what I loved about our conversation today is how you demystified it, talked about how AI-powered hyper-automation can actually redefine business operations. And also how business leaders listening might be able to navigate some of the opportunities and the challenges of AI that doesn't just assist, but actively collaborates with that human workforce. So many big talking points.

[00:29:24] I'd love to hear what people listening are taking away from our conversation. But thank you for starting it today. Well, thank you, Neil. I really appreciate it. And shout out to you. I think you're a harbinger and wealth of great information in the industry and just really appreciate all you do for your community. So AI-powered hyper-automation is no longer a distant concept. It's here, redefining how businesses operate and innovate.

[00:29:52] And we've heard today from my guest about how agentic AI is capable of collaborating with humans and ultimately unlocking new opportunities for industries that are ready to embrace this change. From intelligent document processing to real-time decision-making, I think the future belongs to organizations that are somehow able to balance automation with human oversight and ensuring ethical, secure, and effective AI integration.

[00:30:23] Sounds simple, but it is a delicate balance. So how will you and your business tackle the risks of autonomous decision-making while seizing its benefits? And how do you see it reshaping software development, customer experiences, and operational models faster than we expect? Love to hear your thoughts on this. Remember, get me on LinkedIn at Neil C. Hughes. I'm just back from a trip in Texas.

[00:30:48] You probably didn't know that because I always make sure there's an episode going out every day during my holidays or conference visits. But if you pop over to my Instagram, at Neil C. Hughes, you'll see a few crazy pics there. And also on X, just at Neil C. Hughes, if you want to send me a DM over there. But that is it for today. So please, let me know your thoughts on today's episode. And I'll return again tomorrow with another guest. Hopefully, you'll join me again there. Bye for now.