Unlocking Business Value with AI Agents: Insights from Reddy Mallidi, Chief AI Officer & COO, J&R Consulting

Welcome to TBCY! In this insightful episode, we sit down with Reddy Mallidi, Chief AI Officer and COO at J&R Consulting, renowned for his practical expertise in implementing AI solutions across industries. Hosted by Stephen Ibaraki, this conversation explores the rise of agentic AI, practical AI adoption strategies, governance challenges, and how organizations can build trust in AI systems. Reddy Mallidi shares real-world lessons from deploying AI in enterprise environments, discusses the future of AI agents, and explains how leaders can move beyond the hype to create measurable business value. Key takeaways include:The future of agentic AI and autonomous workflowsOvercoming AI adoption challenges related to data, infrastructure, talent, and trustPractical frameworks for AI implementation and governanceReal-world use cases that demonstrate measurable business impactWhy curiosity, experimentation, and responsible AI practices matter

• 0:08Speaker 00

  Hey Reddy, thank you for coming in and doing this interview with me. Just checking my power, my power went all up on the prior interviews, so I just checking that that's not gonna happen. But you know, you've got this marvelous background, you know, really the top, one of the world's top fractional CTOs, CIOs, executive, and in fact, you have successfully implemented AI across so many different companies, always getting a measurable return. In fact, in February, I came from a YPO event where the whole conference was about AI and how to really get some practical value out of that. And I was, I was saying to people that, you know, you've got to read Reddy's books and the best selling books.

• 0:53Speaker 01

  Thank you.

• 0:54Speaker 00

  You're a top expert in this area. So thank you for coming in and sharing your experiences on AI and on technology.

• 1:03Speaker 01

  It's my pleasure, Stephen. Having both business and technology background, it's very important for me. I don't see technology for technology's sake. It's AI in my view is an enabler to make humans super humans. And how do we leverage AI to create business value? That's where my focus has been. And thank you for the time today.

• 1:34Speaker 00

  So Reddy, we're gonna get right into it. And we have about 40 minutes roughly remaining in terms of this interview. And you know, you see the rapid rise of agentic AI, you see Claude, Claude doing very, very well. And in fact, there's a little bit of a release, a leak that.

• 2:00Speaker 00

  People have been sort of talking about recently, but also you see OpenAI talking about their new model that they're working on finally.

• 2:08Speaker 01

  Yeah.

• 2:09Speaker 00

  I just came back from a week of a private conference under NDA with Microsoft. And so we can't talk about, but you know, there's a lot of amazing things and it's on everybody's mind. So where do you see Agentic AI going? And and as a follow up, we'll talk about the book you just finished releasing in December.

• 2:32Speaker 01

  Yeah, yeah. So Stephen, I am super excited about the AI agents in late 2024, there is the talk of AI agents, but I think what what Agentic AI takeoff was helped by one, obviously Anthropic released the MCP protocol in 2025 last year and no, I mean, 2024, late 2024. And then Google released A2A, one is focused on agents connecting to tools and databases and all the other things.

• 3:15Speaker 01

  And while the agent to agent, that is focused on agent interaction. Because any workflow effectively needs more than one AI agent. Because if you want to complete any workflow meaningfully, you need multiple tasks, just like, you know, we have in our daily lives, right? It's a, so if you're preparing an FPNA statement or you're preparing for earnings call or you're doing the testing, it all requires many, many people. And so the workflow requires multiple agents.

• 3:56Speaker 01

  So once those protocols are in place and people start to see, ah, okay, we have the, you know, very mature LLMs. Now they're very capable and then now agent, someone who draws. Drives these to solve business problems. That became very apparent. So agents start to take off, but I think where the challenges still remain is the infrastructure, the orchestration. And how do you build trust in AI agents? Because we had seen the incidents like a replet. Someone built a code and the agent deleted the entire database by mistake.

• 4:40Speaker 01

  So how do you make sure you trust them? Those are still the lingering major challenges, but AI agents are here to stay to augment the human capabilities. That's where I see AI moving in twenty twenty six, Stephen.

• 5:01Speaker 00

  So can you just briefly outline the books you've written, including, and then we'll focus on the last one?

• 5:07Speaker 01

  Yeah, so the first one I wrote was AI Unleashed. This came out in late twenty twenty four. This was at that time, because there was a lot of talk about what AI can do, and there was a lot of hype. And since in two thousand nineteen and twenty, I had an AI Center of Excellence when I was at Autodesk. I started a small team of five people to solve a gnarly problem to screen fifty million accounts for export compliance, very complex one, and then existing software didn't solve it.

• 5:49Speaker 01

  So I leveraged machine learning. We built our models and we automated it for ninety nine point seven percent. That saved a lot of time in screening. The experience for our salespeople was extremely good. And then, of course, we also saved thirty five million dollars a year from that production system. And when ChatGPT was released at the end of twenty twenty two, that people started to. Ask, oh, ready, you deployed a production AI system, twenty nineteen and twenty.

• 6:26Speaker 01

  So what is this about? And then, so, uh, based on the questions I was getting, I thought, okay, it is important to uh, show how to bring AI effectively from uh, security perspective, how do you build AI teams and how do you make sure that data is ready, because that's needs to be there for AI success. And then I dedicated a significant portion of the book, talking what AI can do for operations, for engineering, for sales and marketing and HR, uh, all the functions. So that was the focus of AI Unleashed.

• 7:09Speaker 01

  And then, uh, as the, as we just, uh, talked about, uh, the agents becoming reality uh, last year, when these, uh, protocols were released, then I thought, okay, you know, uh, it is important that, uh, uh, a book about AI agents, because it is beyond your use models, it is beyond integrating. So AI agents have their own bit of, uh, challenges. So I wrote, uh, this book, Leading with AI Agents, late last year, and it quickly became Amazon number one, uh, best seller, uh, and got a lot of praise.

• 7:53Speaker 01

  This is, uh, this book focuses on, uh, strategies, how do you implement as well as uh, governing the digital workforces as we are calling these agents. So those are the two books. And then, of course, I also have a small workbook. This is focused on uh, AI use cases. Uh, this is like, how do you choose the right AI use cases? Because every company is struggling with so many AI pilots and then they're not seeing the, um, return on an investment. And, uh, MIT study, uh, that came last year said, oh, most of them, they fail.

• 8:36Speaker 01

  They don't produce any value, business value. So that's why I, uh,

• 8:42Speaker 01

  I wrote this small workbook in how do you choose the right AI use cases. So that is the three books I have, Stephen.

• 8:52Speaker 00

  So you have one, AI Unleashed, the book really is centered on the practical applications of where you can put AI in general into your company, the questions you should be asking, the leadership questions, the team questions, the data questions, the kind of problems that you think AI can solve. So you don't artificially create a problem, you've got to have a real problem that you can look at to see if you can use AI. And because there's such a gap between people renting pilots and proof of concepts of not getting it into return on value, on their investment, you created a workbook so that he could can go out,

• 9:36Speaker 00

  think through the process strategically and also operationally of how to implement it so you'll get some measurable result. That's the workbook, companion workbook. And basically then you'll ensure that you'll get some return because you're going to do it in a structured, very thoughtful way. And then because of the rapid rise of agents, particularly last year and this year is sort of the proof point of using agents. You created a book on agents and like, again, how do you, how do you deploy them? Where do you put them in the workflow? Where does it make sense? How do you manage them, govern them and so on? Are you going to produce a workbook for that as well?

• 10:16Speaker 01

  Uh, not yet, but I think what I'm doing slightly differently. By the way, that's a great summary, Stephen. Thank you for succinctly putting it. Uh, the one thing I got requests from people that, oh, you know, it's, uh, it's hard to remember all the content. Can you do small nugget type videos? So I have been posting primarily on LinkedIn and then some on Substack. Uh, I'm starting to get into Substack as well. Uh, so what I do is a minute to two minute nuggets.

• 10:57Speaker 01

  I release, for example, uh, AI COE. Why do you need it? And what are the prime objectives? I try to.

• 11:07Speaker 01

  Put that entire thing under two minutes. So I'm producing those, but if there is a lot of demand for a workbook, I'm happy to do that. Nowadays, it's with AI's help. You can create these books. So once you have all the ideas, you dump the brain and you can quickly come up with a book. Yeah, happy to do that.

• 11:30Speaker 00

  Okay, so ready, let's do this. Um, because I do keynotes all the time, like continuously, right?

• 11:36Speaker 01

  Right.

• 11:36Speaker 00

  And and I always look at little video fragments I could put into this keynotes. Uh, I'm wondering, and this is now a challenge in real time. Can you give me three under two minute nuggets that I can insert into different keynotes? And then you could choose the topic area, but it's got to be on AI. So your first one.

• 11:57Speaker 01

  First, first one is I would say,

• 12:03Speaker 01

  what are the key challenges to overcome to become successful with AI? That's, that's one. You don't need to elaborate, but under two minutes you can cover uh, from trusted to from uh, integration to data and talent building examples, right? And another one is there's a lot of fight on AI. What is practical? Because at the end of the day, you know, we can philosophize and then talk about the philosophy and all that. Uh, but people need business leaders, technology leaders or individuals. They need things that they can take and do

• 12:48Speaker 01

  in an enterprise or in their small business. So give for key ways AI can help businesses and even individuals. Like, you know, writing content, uh, writing code, all the ways it can help. And then the third one is how do you build trust? Trust is very, very key. And now the, you know, between EU Act and NAST frameworks and GDPR and existing regulatory things.

• 13:26Speaker 01

  How do you make sure that your AI not only follows the regulations, but how do you build trust? You know, last year I read somewhere that the majority of the people, they don't trust AI. Over fifty plus percent. We need to overcome that because it's a great enhancer of human capability, but there's a lot of fear out there. So the third one is building trust with the AI. Those are the three topics I would say you can cover under two minutes, Stephen.

• 14:03Speaker 00

  Okay, now, now is the challenge. You want to pick one of those and actually do a two minute and then I'll cut it. I'll cut that fragment and insert it into my keynotes. Or it's not too much of a challenge.

• 14:17Speaker 01

  Let me start with the first one, not in any particular order. All are important, but let me start with the first one. The challenge is what in the, there's a lot of hype, especially I live in the Bay Area, Stephen, as you know. And everywhere I go, from the telephone company to even waste management, if you will. Everyone says they're AI enabled or they're AI. First, it's just mind numbing and then all the ads you see billboards when you're driving. So I would focus on what are the real challenges?

• 14:57Speaker 01

  And I would, I would say the things people need to be really, really focusing on, making sure that data is right. Because what happens, there are, you know, I talk to executives, I talk to startups. The big challenge that people don't necessarily address is they don't have data.

• 15:23Speaker 01

  Or they have data in lots of places. It's probably, I stated it differently, the data doesn't, it's not in quality and the right places. It is spread across everywhere and then getting access to the data, it's a huge challenge. And then, once you have the data, by the way, in many cases I have seen Stephen, majority of the costs in getting the AI ready or even deploying into production is from data. That's, that's a huge cost. And then

• 16:04Speaker 01

  uh, you, once you get the data cleaned, the relevant data in enough amount of data, then you need to have, well, the infrastructure readiness, because you know, the other day I heard

• 16:25Speaker 01

  healthcare, it's, it's, it's a huge part of our GDP in the US and they are using systems that are fifty plus years old. And how do you make sure that the AI can be integrated with all the CRMs, your ERPs and even the mainframes and various applications without redoing all of that, because that's going to be a huge undertaking, right? So the, uh, that part needs to be addressed. And then the third one is the talent, talent. It's a lot of companies, uh, struggling, they don't openly admit, is that

• 17:10Speaker 01

  it's a, they think that, oh, okay, I can hire a PhD in machine learning or deep learning or, or one of those areas, AI. And no, then I'm good. But it doesn't work like that. You need to have a cross-functional team that brings

• 17:29Speaker 01

  AI experts in machine learning and then people who train the models and who understand your niche area, the industry, what your IP you have and all the knowledge. You need to bring them, plus the ethicists who want to make sure that the AI is operating within the guardrails. And then that means security, that means governance, that means building the trust. So, once you address these aspects, then you are ready to bring AI or AI agents.

• 18:14Speaker 01

  Otherwise, your LLM or machine learning model, whatever model you bring, it's garbage in, garbage out. So that's a, that's the key that many of the people I consult with, the many enterprises, small and the big IT companies, they want an answer today and they want a success. They would say, oh, hey, I need AI because everyone is asking me if I have AI. And then I wanted the results tomorrow. That's not going to happen if you don't have these, the foundational elements in place. So you need to get the very strong foundation and then you can build the house.

• 19:00Speaker 00

  So that's a great set of nugget. And you want to try one more focus specific or maybe not. Why don't we get into agents itself?

• 19:10Speaker 01

  Yeah.

• 19:10Speaker 00

  And we'll get into a bit of a question and answer on agents. And, and what I'm going to do is so, so keep, keep in mind when I do my keynotes, I'm going to take little fragments of, of this, of this interview that I can sort of paste into my keynote and then I'll point to the full interview so they can see your books. And, and so make sure in your profile, I don't know if you said that, have in your profile, do you have all your books in there, the covers, it's done?

• 19:36Speaker 01

  Uh, in, uh, well, um, you mean on my website or?

• 19:41Speaker 00

  So when the interview gets published.

• 19:44Speaker 01

  Yeah.

• 19:44Speaker 00

  Uh, I always say, uh, provide a full profile.

• 19:47Speaker 01

  Oh, uh, yeah, yeah.

• 19:49Speaker 00

  And in that profile, make sure you said out loud.

• 19:51Speaker 01

  Yeah.

• 19:52Speaker 00

  Um, um, your books, so we can, you know, your books get profiled and links to your books and links to talks and, you know, any kind of additional, uh, material because it, when I approach you, my keynotes, then your profile will be like a business development tool for you.

• 20:09Speaker 01

  Yeah, absolutely. So I, I, the last time I gave you, but, uh, I need to update and I will send you, uh, later today, uh, Stephen. Absolutely.

• 20:19Speaker 00

  Okay. So now, uh, on agents, there's so much talk on agents and things like that. So, um, a, a question then, and, uh, did you, have you tried OpenClaw?

• 20:32Speaker 01

  Actually, I, I did and I didn't. Here's why I say, because I

• 20:41Speaker 01

  having seen many insta- instances of, uh, leaks, data leaks. I am very, uh, conscious. So OpenClaw is very powerful. And but I'm worried about, uh, the, um, the security aspects, the governance aspects are back there. That's my big concern, but now, uh, Stephen, Claude can almost get you to that capability today.

• 21:19Speaker 01

  So I am actually trying with a Claude, the same level of, uh, functionality I can do with the OpenClaw. Because it's, uh, you know, now I got a MacBook, which I don't use for my regular day in and day out, but, you know, I'm just, uh, it's a powerful one. I am trying, uh, I, I'm also, you know, it's, uh, traditionally long time.

• 21:46Speaker 01

  Back, I used to even write in a simpler code in the nineties. But we came long ways. So I actually allow to play even at the code level. And there is no code tools like allowable and the replit. But they go only that far, but OpenClaw. Man, it's it's extremely powerful. It can do so many things. The other thing that I'm my concern, Stephen, is

• 22:20Speaker 01

  the autonomous agents, it gets me a little bit concerned. In the sense like, you know, the example of the replit, right? It will completely wipe out the database. And then what it the agent did was it created dummy agents to show that, oh, no, no, I actually didn't delete. It was these are the ones I got the instructions from. Which was, oh, I don't want that situation to, you know, agents doing things to obfuscate me as the user or the human. So I want that guardrail, so I'm a bit careful.

• 23:02Speaker 01

  That's why if it is within the four walls of my my MacBook, then it is okay. But once I got get to outside, I don't want any leaks on the data, private information, all that. If it is just, you know, managing my schedule, managing my emails, then absolutely.

• 23:29Speaker 00

  OK, so what you've done then is it sounds like you played with OpenClaw, but in a safe sandbox to see what see what it can do. And then you and then you you looked at Anthropic Squad and they and they have an agentic mode that you can play with. And and it's got the guardrails and and the security features and things. And you'd be looking at that and you're saying it's actually pretty good. That's that's my interpretation of what you're saying. It it looks pretty good.

• 23:58Speaker 01

  Yep, it it looks pretty good with the guardrails because I don't want to compromise on the security and the trust aspects while the capabilities are are phenomenal in OpenClaw. Yeah.

• 24:12Speaker 00

  Okay, so and then Jensen, you know, GTC was just a few weeks ago and he talked about Nemo Claw. Have you tried Nemo Claw?

• 24:22Speaker 01

  I haven't. I haven't tried. Yeah, I actually watched. Obviously, you know, Jensen has a lot of interest in yeah, it being an evangelist or, you know, AI in general, but he's he's going a thousand miles an hour and he's putting resources everywhere and investing, but I haven't tried yet. I it's in in many ways, Stephen, what I see as there are many incarnations, if well, that's the right word, of the technology, L L M's, for example, right?

• 25:04Speaker 01

  It's a we had seen how they have evolved. I mean, you have been part of that through Microsoft engagement you have and

• 25:15Speaker 01

  they are leapfrogging one or the other, right? Oh, yesterday was OpenAI's ChatGPT, then Claude, and then Gemini. Their capabilities are getting near equal in many ways. You can't go wrong with any of these. Of course, Lama fell behind, but in general. So if if you, let's say, try OpenAI or even, I mean, I'm sorry, the Open Claw or Claude. Nemo probably its capabilities, I'm guessing, would be close enough. So if you try one, it's that's good enough for me.

• 25:58Speaker 00

  Have you have you tried the agent mode of OpenAI at all?

• 26:03Speaker 01

  Uh, actually, I did, but I

• 26:08Speaker 01

  I have been a big fan of both Gemini and Claude.

• 26:12Speaker 00

  That's it.

• 26:15Speaker 01

  It's it. I, I, well, initially I was all using OpenAI, but also in parallel, because of the security and more guardrails, I became a fan of Claude. And also, it was doing a way better job in, especially writing and creative side. Of course, not the images and videos, but I was using both of them simultaneously. And then Bard was replaced by Gemini. Bard used to be a piece of garbage at that time by Google. And then Gemini jumped quite ahead. And now I use both Bard and Gemini in parallel.

• 27:00Speaker 01

  And OpenAI, I don't use it as much, but I try. When new features are announced, I try and then say, oh, it is better, maybe slightly better. If it is, then I keep using it. Otherwise, I go back to Gemini or Claude.

• 27:16Speaker 00

  Yeah, I mean, I find that OpenAI's GPT-4 follows instructions a bit better, I find.

• 27:24Speaker 01

  Yeah.

• 27:25Speaker 00

  Or if I'm doing just pure research, something just pure research oriented. For some reason, I just find it maybe a little bit better, maybe more accurate, more current or something. But you're right. I mean, you're always, again, I guess it's 3.1 Pro right now with Gemini. And so you play with that and see what it looks like. And, and then I looked at Grok and I looked at DeepSeek as well. I've not played with Claude, so maybe I should poke around with Claude.

• 27:54Speaker 01

  You should try. You should try. You would be amazed. Oh, by the way, Stephen, I actually say in a nuanced way, what I didn't tell you is, I, from my consulting, I use Microsoft Office 365, the business version. So that gives me

• 28:16Speaker 01

  Microsoft's, the Copilot.

• 28:24Speaker 01

  So, so the through Copilot, I get ChatGPT. It's it's a, so I am, I am not wrong to say I don't use ChatGPT, but I, I, I quite frequently use Copilot. And then I actually built, this is before Claude got to this level or Gemini. I built agents with, with AI Foundry and with Copilot. You know, one thing I would say, Stephen, is they have done probably the best job, Microsoft, that they have integrated Copilot and actually the, the AI capabilities,

• 29:09Speaker 01

  whether it is in a through ChatGPT and different incarnations, into their products better than anyone else. Of course, now Gemini is integrated into Google products. But again, so I indirectly I use ChatGPT through Copilot.

• 29:28Speaker 00

  Do you, do you use Visual Studio then? And when you're doing development and things like that or?

• 29:34Speaker 01

  Well, it's, you know, what, what, you know, the Microsoft, the one way they impressed me is you can go to the same destination multiple ways. You can, you can do a hair foundry or you can get the Azure and then you through the studio, you can actually build your own agents. Or if that's too complicated, you can actually just go to Copilot and then say, hey, you know, this is actually my Copilot is open. It's, they can, through the chat, you can build the agents. You know, it just gives you the menu of create an agent. It's like, okay, well, I have five agents. I need to orchestrate them.

• 30:19Speaker 01

  You can do many ways, just like, you know, even Google, you can go through

• 30:29Speaker 01

  their

• 30:30Speaker 00

  AI Studio.

• 30:31Speaker 01

  BCP, right?

• 30:33Speaker 00

  Yeah.

• 30:33Speaker 01

  Uh, multiple ways, or you can go through their studio, or even indirectly you can go to their notebook alum, there are multiple ways, but I think uh, Microsoft gave more avenues, at least at the beginning, to build the things in AI. And you can do the, even you know, uh, Snowflake, for example, they came up with the uh, building agents, they're getting uh, better, it's like, that come from the data side and uh, uh, data cloud, but they're also providing integration to MCP and then uh, in the building AI agents through SQL or no code, uh, different ways.

• 31:17Speaker 00

  So, uh, so you're using uh, Foundry then, or have, have been using Foundry?

• 31:21Speaker 01

  Oh, yeah, yeah, yeah. Uh, Foundry is pretty good, yeah.

• 31:24Speaker 00

  Yeah, it's good.

• 31:26Speaker 01

  Yeah, it is, it is good, yeah. So, I actually just totally slipped, I was thinking, oh, uh, ChatGPT, no, but uh, well, I have a different incarnation of ChatGPT through Co-Pilot.

• 31:37Speaker 00

  So, uh, so Foundry's got a lot in it, what do you use in Foundry?

• 31:42Speaker 01

  Yeah, yeah, yeah, it's, it's, I, again, for comparison purposes, I use, because I have access to it, and I play uh, to see what capabilities are there, uh, but now it's like, you know, uh, between AI, uh, Azure, AI Foundry, uh, cloud is, of course, it doesn't have that same infrastructure, but with the cloud, then there are so many capabilities, connectors, and all these different modes that Gemini has it, again, you know, multiple ways you can get to Google's AI capabilities.

• 32:23Speaker 00

  So we got only maybe about five minutes left, and so in these five minutes, um. Do you think that these things are going to be fully autonomous? In other words, they're going to they're going to be to self-improve themselves on an autonomous cycle? You think agents are going to achieve that kind of capability where they can self-improve? And on this flywheel effect? And do you think they'll be fully autonomous this year? Or maybe next year where literally you just tell it and then task and it'll just carry it out? And you can reasonably trust it? Do you think that's going to happen this year or next year? Or do you think it'll never happen?

• 33:08Speaker 01

  I won't use the word never, Stephen. But here is my thought process, right? I am not going to give you yes, no answer to this. But it's a spectrum, right? It's zero autonomy to full autonomy, right? I would suppose that most of the agents will live somewhere in between.

• 33:38Speaker 01

  And that somewhere in between is going to be determined by your use case. Let me give you an example, Stephen. So, you know, when I ran operations at Autodesk, I had forty different services through my organization. And we had client support team. And when I look back,

• 34:04Speaker 01

  probably seventy percent of the things that the client, customer service experts were handling could be handled near autonomously by AI agents. Because, you know, it's the compendium of knowledge that's out there. Once you train these models, then they can give responses. They can do things automatically. For example, if someone wants to switch their licenses. They want to go from five licenses to eight licenses. Or their usage went up.

• 34:45Speaker 01

  So they want to buy some consumption instead of buying additional seats, right? So those things can be done. Really? By AI agents. So there, when I look back, oh, seventy percent, maybe max of eighty percent can be done. So in those cases, I would give, well, close to ninety percent or maybe a hundred percent of those tasks are handled. Uh, by AI. On the other hand, let's say, you know, in, uh, in, in the book I, I covered in the leading with AI agents. In, in Vilnius, in Lithuania, uh, a hospital system.

• 35:27Speaker 01

  They, uh, deployed AI to look at the x-rays and, uh. How did the radiologist? And initially, there was a lot of resistance because it's like, oh, is it going to replace or is it like, is it better than me? Right? There are all those aspects, but once they're trained and then the radiologist is the human in the loop. So what, uh, it does is AI handles now, there are actually seventy percent of the volume is handled by AI. But the expert radiologist looks at it and then, oh, does this make sense? Does it not make sense, right?

• 36:10Speaker 01

  So that is, it is an assistant and it's zero autonomous. But it is providing the intelligence and then the radiologist is making the decision. In that case, it's, uh, uh, you are making a very, very important, uh, decision. Uh, or diagnosis that affects human life. Or something like, you know, lone origination, for example, there are many, many tasks. I don't see agents making decisions, but humans. So it's on the spectrum depending on the use case. Uh, but will they get to a hundred percent autonomous, uh, status?

• 36:55Speaker 01

  In some situations, yes. In some situations, no. And you don't want them to be a hundred percent autonomous in situations like, uh, you know, looking at the, uh, x-rays. Yeah.

• 37:09Speaker 00

  Yeah, so you don't want to, so that's that's kind of a political business decision, but will they be capable of being fully autonomous?

• 37:20Speaker 01

  Uh, again.

• 37:23Speaker 00

  I'm not saying we're we're gonna let that happen as humans, but do you think it will have? So this year or maybe next year. You you see Sam Altman saying that you work, you'll have a fully autonomous researcher. Let's say by two thousand twenty eight, right? And they're just now doing the final training of this next model, which is supposed to be a fairly big leap. And Claude and others are sort of saying there's some really big things coming out this year.

• 37:52Speaker 01

  Yeah.

• 37:52Speaker 00

  So so then this becomes a sensitivity, a political sensitivity, business, you know, human survival, maybe question. Do you think it's capable, but we will not allow it?

• 38:07Speaker 01

  You know, the when, you know, one of the things I covered in the book, Stephen, is AI agent evaluation framework. I I don't know if you can, you can see it. So one of the aspects is its task completion accuracy. The other one is efficiency and performance, we can live with that, and then reliability. And then the fourth is the user experience. So what I would say is when anyone says autonomous, that means I don't need to look at, as a human being, I don't need to look at the decisions it made, no second guessing.

• 38:51Speaker 01

  I doubt we will get there. I doubt it, we will get there. That's why I still believe that in there, it's on the spectrum of somewhere zero autonomy to full autonomy. They may get to ninety percent accuracy, ninety percent of the time, but a hundred percent accuracy, a hundred percent of the time, I doubt it.

• 39:16Speaker 00

  So that's a kind of a philosophical question too, though, right? It's like, I have a company, let's say I have a company and I can, I have, I assign a duty to some people and then they just handle it. And they'll just, they'll report once a month what's happening.

• 39:31Speaker 01

  Yeah.

• 39:32Speaker 00

  You don't think that these agents will be at that level where they're like employees and you can assign them things and every week or every month they'll report back and you give them tasks with goals and it's like having employees working for you.

• 39:50Speaker 01

  Yeah, I mean, I, the one thing that Jan Lacon has been saying, which I conquer in a big way is the underlying LLM, the architecture, the transformer architecture has its limitations. And once something has its limitations, so the same where you go, it gives one result. And you give the same one at a later time or even five minutes later, it gives you slightly different. So is it within what is acceptable, right? Traditional software, we need, we used to have, it's one or zero.

• 40:34Speaker 01

  Now it is like a limit theorem, a cataclysm, right? As of, you know, A times to one. So it's like somewhere as we are getting closer, it's probabilistic, right? It is okay, you know, oh, I guess it is eighty percent. So that is the challenge with the current architecture and sometimes you get perfect results and the other times, oh, what is this? You scratch and then it's even with the, you know, Claude or Gemini, the memory, you can, you see all kinds of memory challenges. It remembers something from a different context and then it weaves that in. I see that with the chat GPT, I have seen that.

• 41:19Speaker 01

  A Claude, Gemini, then I had to explicitly instruct, forget about that, we are not talking about that discussion. Look at this one, then I had to give it the context again and then upload. So if An agent is behaving with that underlying LLM like that. Then there's going to be a challenge autonomy. No, it's that's why I believe. Uh, getting to that in in the, you know, next year or so with the limitations, the architectural limitations, I doubt we will get there. We may get close, but a fully autonomous, I am a skeptic.

• 42:00Speaker 00

  Okay. And okay, so we're down to our last two questions. What's your next book? And then I'll ask the last question. What's your next book?

• 42:09Speaker 01

  Well, the one thing I was just going to the more basics, Stephen, so. Uh, I keep getting again, we as technology, uh, very savvy people like ourselves, we tend to. Focus more on the, you know, uh, more sophisticated side. But actually recently I was traveling to India and then. Some of my friends, they're not technology people and it's like, how do you get started? And then, oh, I don't know anything about this then. Uh, I was thinking, hmm, they don't even know how to prompt. Uh, and they get scared when they hear prompt engineering because it's not engineering, it's how you effectively do that. So.

• 42:54Speaker 01

  I I I have been writing this, but it's a, you know, uh, uh. How do you use this, uh, chat G P T or. Pod, any of these prompt engineering for basics. I'm just going to write it and then, uh. Uh, give it, put it on, uh, uh, yeah, Amazon with no fanfare and. Whatever the minimum price is so everyone can get a copy from there. So that's on prompt engineering for all kinds of functions. Maybe, you know, I would put four hundred prompts in there.

• 43:30Speaker 00

  Okay, so show show the first book, A I Unleashed, and just so, and then I'll ask my final question. So. A I Unleashed, right? Best seller. Now show the next one.

• 43:46Speaker 01

  Okay, let me show the lead, how to choose the.

• 43:52Speaker 00

  The workbook.

• 43:53Speaker 01

  Workbook.

• 43:54Speaker 00

  Yeah.

• 43:54Speaker 01

  Uh, this is Compendium to the AI analyst. Uh, and then this is my, the second major book, Leading with AI Agents.

• 44:04Speaker 00

  Already a bestseller, it was released in December, the audience had half the flock to it because agents are everywhere and you're the master of this and now you're working on a book that you're gonna make it really accessible and basic prompting because people just wanna know how to use these things, right? And it's recommended. Okay, I have one final question. The final question is, what's your final recommendation to the audience and, and or questions you want to leave the audience?

• 44:34Speaker 01

  One thing I would tell anyone listening to this one, Stephen, is don't be afraid of AI, don't treat this as some magic wand that's going to replace all of us. It's use it to superpower yourselves because it's possible. It's possible to anything you're doing, you can do it more effectively, way faster with AI. It is something to embrace, everyone should embrace, don't believe in the hype, but embrace AI for your own good. And as far as the question is, uh, to the audience.

• 45:24Speaker 01

  I would, I would say.

• 45:28Speaker 01

  What would you do? If, if someone comes and tells you that, hey, I got this magic wand, I'm not saying AI is a magic wand, but it, it comes close. And.

• 45:47Speaker 01

  Would you say I'm going to just run away from it? Or let me at least see what it is, right? What would you do? That's my question. It is very related to what I said embrace. But this is for the people who are concerned about AI replacing them. What would you do if someone tells you I got a magic wand? At least you want to see an experiment, right?

• 46:14Speaker 00

  So, Reddy, thank you for coming in and I recommend the audience embrace your books. I mean, they're just fantastic. And of course, your work is just amazing. The entire universe out there should follow your work. So, thank you again and share your wisdom and expertise with our audience.

• 46:29Speaker 01

  Oh, absolutely. It was my pleasure, Stephen. Thank you again for your time.

• 46:36Speaker 02

  Thank you for listening to the brand called You, videocast and podcast. A platform that brings you knowledge, experience and wisdom of hundreds of successful individuals from around the world. Do visit our website, www.tbcy.in to watch and listen to the stories of many more individuals. You can also follow us on YouTube, Facebook, Instagram and Twitter. Just search for the brand called You.