Luke: [00:00:00] You’re listening to a new episode of The Brave Technologist, and this one features the authors of AI First Leader, a practical guide to organizational AI leadership structured around three progressive arcs, AI essentials, AI at work and AI advantage. The book offers a roadmap for leaders who want to build not just capable AI systems, but resilient, ethical, and high trust AI first organization.

The book was written by today’s guest, BH Meta, who is an AI first leader and engineer with 20 plus years of experience building scalable systems at Uber, Cisco, VMware, and high growth startups at Uber, he leads customer support technology, applying conversational and generative AI to reduce costs and improve user experience.

We’re also joined by his co-author and colleague. Mahesh Kumar who assist seasoned product executive with two decades of experience in AI cloud and enterprise software. At Uber. He helps define the AI ML strategy focused on [00:01:00] automation and intelligent customer experiences. In this episode, we discussed specific ways that small businesses and Fortune 500 companies can embrace AI from a place of empowerment versus fear.

How to align C-suite leaders and engineering teams around the AE Unified AI roadmap. And the most underestimated human factor that determines whether A and AI transformation succeeds. And now for this week’s episode of The Brave Technologist,

why don’t we start with a story that opens your book. There’s this really great bit about the Nova Bridge chatbot failure. What does that moment reveal kind of about the hidden risks of poorly governed and unoptimized AI systems? And maybe we can kind of like give a little bit of color.

Like high level around what happened with that?

Bhavesh: Yeah, so that opening story in the book, the Nova Bridge Health Chat bot runs into a critical failure. It is a fictional scene, but not really draws the readers in and captures a very real pattern that we have seen across industries. The bot gives a confident yet [00:02:00] unsafe answers to questions like, Hey, is ibuprofen okay with my heart me match?

And it exposes three layers of lack of controls lack. First, there was no problem governance. Second, there was no grounding to the real data and no ongoing checks for hallucination or compliance in the story itself. The team eventually fixes it by applying a framework that we advocate in the book called Craft framework.

C, clarify intent. R reinforce safety A, adjust with examples. F, focus on approved guidelines and T test in real world conditions. And also by introducing rag, which is retrieval augmented generation, so that every answer cites verified medical content. Those challenges. They turn fluent but risky model into a grounded, and that’s auditable.

But the point isn’t, point was of the story wasn’t a technical fix. It’s really about the leadership lesson that is in high stake [00:03:00] domains like healthcare whether a lot of regulations and trust is the key currency. Trust becomes an engineering problem and so does governance.

Mahesh: Yeah, I I, I tend to agree.

I mean, if you step back for a moment there, Luke. The deeper lesson that I take away and, you know, that we wanted to impart is around trust economics. And you’ll hear this a lot as we enter this transformative era where generative AI and AI in general is seen in every sphere of life and influencing everything and anything we do.

So AI can, on the one hand create value, but it can also erode it overnight. Specifically in the Nova Bridge case, a vague answer, you know, like ish pointed out, sent a patient to the ER and then triggered both, you know, reputational as well as potentially legal fallout for the healthcare company itself, right?

So the response here is not, let’s go tune the model. It’s about rebuilding that confidence. It’s about regaining the trust economics that you want to build with your patients, [00:04:00] with your end users. And in the case of Nova Bridge, you know, as you’ll see. The leadership decided to establish a dual product vision.

So on the one hand there’s the patient support chat bot for 24 7, compliance self-service that patients can avail of. And then there’s also the clinical co-pilot, and that’s for the staff. And both of these are framed explicitly as safeguards for trust, compliance, and quality of care. So the book makes clear the story is fictional.

But it’s built to mirror the real high stakes regulated environments that we see, you know, exactly where things like governance and product design become so enmeshed that they’re inseparable. And what applies to healthcare potentially can easily then apply to other verticals like financial services or retail or, you know, high tech.

And so.

Luke: Yeah, I mean it’s, it’s really interesting too because these details and kind of nuance tend to get overlooked. And I think we’re even seeing that when, you know, I think Sam Alman was talking about [00:05:00] how people are using chat, GBT as like a therapist or a lawyer or doctor and giving a lot of personal info away that they don’t necessarily, the end user doesn’t necessarily know like what the regulatory compliance should be. They just want help and some, and if they’re usually in a healthcare situation, they’re dealing with something health related, right? Like, so I love that you guys kind of touch on real world, even if it’s a fictional like, you know, case. There’s so many of these things happening and people can relate to them, right?

Like I. Seems like the, the book kind of takes a really good approach of practical ways that AI can kinda work with the real world and, and, and with people working in ai. Is that fair to say?

Guest 2: Very true. I would say, I mean, at the end of the day, if you’re looking at Geneva and how that’s getting enmeshed, I mean, we are still in innings, one or two at best of this transformative journey.

And this is going to affect all fields of our lives, right? Whether it’s personal, professional, every walk of life, you know, from medical care to research, to you know, manufacturing you’re gonna have physical AI and manufacturing, for [00:06:00] example. So all of these domains are going to get affected. And it’s one thing to rely on the efficiency in the autonomy that AI brings, but balancing that with governance, with compliance, with quality is gonna be super critical also, so that we are not just, you know, lemmings following AI off the cliff, so to speak.

Mm-hmm.

Luke: Makes sense. Makes sense. I think one thing you both argue too in the, in the book is uh, the AI transformation starts with you. What mindset shifts do executives, you know, need to, need to lead responsibly today from your, your guys’ point of view?

Bhavesh: Yeah. I think the biggest shift is realizing that AI is not a feature.

I can’t tell you how many times I have heard or how do I, you know, build AI into this this new model came out. How do we ship, you know, some application with this newest model? It’s not about. Feature integration. It’s a whole new operating paradigm you can bolt on and expect transformation. It has to start with leadership.

Sponsoring a real change blueprint. You know, create a sense [00:07:00] of urgency, build and guide a coalition, show early wins, and make that change stick. A lot of pilots that just fizzle out here’s the cashflow. I mean, leaders will need to become AI literate. It’s not just about. Learning about new AI terms and talking about it, it’s really about experimenting hands on lot of AI tools across the range of your business portfolio.

Trying them out and developing independent intuition about it. They have to understand terms like drift groundedness, evaluation buyers, because those are not technical details at this point. Those are real business risks. That is why we talk about pairing AI adoption with AI Academy so that leadership teams can reason about the technology, not just fund it.

It’s not about building fluency. It’s about building fluency not dependency.

Mahesh: Yeah, absolutely. I mean, like Bahi pointed out, right? I mean there’s the whole arc of the technology itself, the terminology, and the things that leaders need to [00:08:00] be aware of as they engage in this transformative journey.

I would say there’s also the aspect of leadership itself. You know, how do you lead with clarity and empathy in times of extreme and often, you know, what feels like a seismic transformation. So, you know, painting a concrete vision that is tied to the business outcomes, being able to communicate that relentlessly.

Designing early, you know, building visible wins that reduce fear about this new technology, but build momentum behind it, that becomes very important. And then how do you take that and institutionalize it with sort of an AI transformation roadmap itself? Right? So for example, components of such a roadmap may include things like a multi-phase, multi-year plan.

How do you institute governance? You know, where’s your center of excellence for ai and how are you developing that and spearheading that? What are the incentives you provide for employees for both this technological SLS cultural and, you [00:09:00] know, mind shift that they’re gonna endure? And then how do you monitor and constantly refine all of that?

So I feel when executives own the narrative and the cadence, the adoption follows, the alignment increases. And people operate or start operating from a position of empowerment rather than fear.

Luke: Yeah, I think that’s a really great point and, and kind of the whole approach too, because if you think about it, like, it’s a kind of a rare thing where you’ve got like Fortune 500 plus companies that are basically got mandates to put AI everywhere and now it’s like a national defense thing.

It, there’s, you know, actual like picks and shovels building out data centers and it’s in the news and, and all this stuff, but like. there isn’t also a ton of market fit with it yet. So it’s like, you’ve got this top down kind of, mandate and you’ve going down to engineers that may not have like, been interested in AI before this whole thing happened.

So I, I think it’s really cool that you guys are kind of helping to kind of shift the mindset [00:10:00] because it isn’t about fear, right? Like it’s really about just like, okay, here’s how you can, kind of like, get yourself in, in the right direction you know, for you and your team or your org or whatever.

How do you guys get C-suite leaders and engineering teams aligned around a unified AI roadmap?

Guest 1: Yeah. Now the one central theme of the book which we hope to communicate clearly is about ROI. So you really need to align the. Behavior of the model to the business outcomes. And we propose a simple loop, which is, you know, you pilot, validate, scale and irate and all by keeping a shared dashboards where all the evaluation matrix like groundness latency, hallucination rate, bias, risks, legal brand otherwise, but also business outcomes like revenue.

If you are in customer support industry, CSAT churn when everybody is looking at the same page. The conversations they shift from, hey, is it working or not? To what are they really optimizing for? [00:11:00] And AI is not one and done deal. It needs continuous observation. It’s quite nuanced. The responses are gonna be stochastic and hence not repeatable.

Borderline unpredictable. How do you tame that and what are the type of risks your business is willing to. Assume and what are absolute no-go for generative ai. We can go a step further. We can predefine how quality will be measured, so that means agreeing upfront what good looks like. It’s not just about right or wrong, it’s about relevance, safety, tone, truthfulness, and task completion.

You know, we, we also mix the human in the loop reviews with automated checks. For ground net scoring, hallucination detection and format validation. So it’s, it becomes less about chasing for accuracy, but more about aligning your customers and your business across trust and usefulness. So, and once the executors and engineers say share the same definition of quality alignment, [00:12:00] soft being an agenda item or a top down magnetic, it becomes a muscle memory.

Guest 2: Yeah. And it may be you know, as to your point earlier, Luke. There’s this top down narrative. There’s this top down mandate almost like, Hey, go do something with ai and go do it. Now, you know, there’s the FOMO factor. The competition just published this case study, or, you know, put out this PR blurb.

And it may be very tempting to do many things at once. Across lines of businesses, across stakeholders. You know, there’s bound to be varying interest levels for different initiatives. For example, should we be chasing efficiency or offering. More generative choice to our customers. Should we leverage AI for marketing campaigns or go full hog on automating complex workflows, and maybe it’s some or all of the above but without alignment behind a very strategic roadmap.

This strain is headed for what I call failure station. and here’s one way to tackle it. Run your portfolio for your business, for your company, like a product. So how North Star [00:13:00] outcomes. Two to three showcase use cases and you know, frequent business slash technology, you know, monthly or whatever that cadence is for your company, for your enterprise Reviews, very approve.

Scale up only when an evaluation gate turns green, for example. So in essence, the roadmap narrative, you know, what you’re planning to build and why you want to do that. Plus the operating rhythm of who meets when and decides on what and looks at what kind of operational metrics. That’s what keeps leaders in the product and what you deliver aligned as one cohesive story for the entire enterprise.

In our book, you know, the strategic payoff explicitly links optimization, continuous evaluation and infrastructure reliability to ensure that your board has the confidence that, you know, you really have a strategy that they can stand behind.

Luke: It makes sense. Just give people rolling together and then make sure you, you keep your eyes on where you’re going and that there’s no holes in the boat.

Right. Or something like

Guest 2: that. Like that. [00:14:00] Yeah. Or, or, you know, have the arrows pointing, you know, in the same direction.

Luke: Exactly. ‘cause you know, like you guys pointed out, there’s, there’s vocabulary to learn, there’s different capabilities that people might have different, you know, sets of might be different points in understanding, like, that’s super interesting.

You know, what are the most underestimated human factor that determines whether an AI transformation succeeds from what you guys have seen or people that you talk to?

Guest 1: Yeah, I think that there’s a lot of debate around this. I think a lot of people and. Roughly understands what in our terms, like human in the loop, et cetera, means.

But the no average health stories show this perfectly. The transformation is not because of smarter or better models, they will improve over time. It’s really about human judgment at scale, and I do want to differentiate between sort of treating human in the loop as a checkbox versus judgment. The team learns that automation can only go.

So far, what really drives the reliability is the human in the loop system that blends the speed of AI with expert insight and speed of [00:15:00] ai. Not just come from compute, but ability to sort of access a ton of information at fast speed. But expert oversight is what? Leverage the performance of AI systems, not models, AI systems end to end towards customer satisfaction and usefulness safety, compliance, et cetera.

So in the story, clinicians that continuously review AI’s recommendations, scoring them for em, empathy, compliance, shown, and clarity. And this is not just what. The customer expectations might be, but also how they wanna position that company’s brand about patient first, et cetera. And feeding that feedback continuously into your system to improve prompt retrieval updates.

And for that matter, you know, even data that AI model is accessing. So I mentioned this before, it’s not a one and done exercise. It’s a living. Feedback loop. And that is the real point. Human oversight is not safeguard that you need to add at the end. It’s a capability you need to [00:16:00] design in your development from the very start, the cautionary part here.

I feel a lot of organization that treat human in the loop like a checkbox. Hey, we got a bunch of data. We labeled it once. Evaluated it for a range of criteria and be done with it from, let’s say just compliance perspective. But the no averages journey shows the exact opposite.

When you invest in human feedback as part of learning cycle, you end up building systems that get smarter. Better and more effective over time.

Guest 2: Yeah, absolutely. I mean, I, I would add to what Bish just now said, and this applies I think regardless of any transformation, you know, we are at the cusp of beginning of the AI transformative journey, but this potentially applies anywhere.

And that is the, emphasis on building a culture and a desire for change readiness. You know, organizations have this inbuilt inertia to change. It is just who we are, even as humans, right? Try setting an alarm for 4:00 AM in the morning and waking up every day. It’s not gonna be easy [00:17:00] at all, right?

So, and the way this plays out in companies or in enterprises, you’ll see that leaders routinely or invest in tools and often under invest in transparency, listening forums and upskilling. You know, the thinking being that, Hey, you know, if I provide you the means, the job will get done. And it doesn’t work that way.

I mean, we all, you know, if you have kids, you know that there are incentives and there are, you know, you have to crack the whip Oh man. In terms of, you know, how to motivate and, you know, bring that out. So the culture section in the book that we’ve written shows how to diffuse fear. You know, things like job loss, compliance, risk with open communications, training, visible career pathways, you know, essentially a formula for turning anxiety into ambition.

And this is very important, and here’s why. You know, every day you wake up and the news bites, they hit you in the face. So many jobs are lost to ai. So many layoffs are happening attributed to ai. So whenever there’s [00:18:00] a technological transformation at the scale that we are seeing, there’s bound to be that fear and anxiety.

We saw that during the internet boom, the PC revolution and elsewhere, right? So the organizations that succeed tomorrow will actually harness that anxiety in a positive way. By ensuring their workforce is AI educated and not just as users of technology. You know, not the the person who writes the best prompt for chat GPD, but a workforce that understands how to harness AI across their daily roles affecting common daily themes of productivity, efficiency, quality in the work they do.

It is then I believe that, you know, the AI factor that magic X factor of AI truly kicks into high gear in any organization.

Luke: It makes sense for me, in my mind. Keeps going back. I mean, we’re, we’re kind of very privacy oriented at brave and a lot of that, whether it’s privacy or opsec, it, it is like a, a mindset that you apply to everything [00:19:00] you do.

And it sounds like what you guys are the, the message you guys Processes is the same with AI too. It’s like how you, you have to get in the mode of kind of like, how can I apply it to different things I’m doing? And, and not just about one feature that uses ai, but really like, how can AI be applied to like, you know, what I’m doing in, in, in my day to day.

Like, yeah, I, I think You both worked at large scale and of mission critical ai. What has that experience taught you about reliability and real world deployment for folks? I, I just think people might be really interested to hear your guys’ perspective on this, so, so many might or touching it for the first time.

Guest 1: Yeah. You know, when chat chip launched, you know, it seemed to do magical things. Write poems for you, you know, rephrase, summarize, and it seemed like a nice. To have add-on into your life. If you want to build AI into enterprise, it has it, it has to move beyond that, it’s not nice to have any more and no average health story that they do go through major reli reliability crisis.

Again, it’s a fictional story, but [00:20:00] very much drawn from real world patterns. The early system failure was not about a particular model misbehaving or going wrong. It was really about fragile system design. So you need to look at holistically, a great air engine that’s sitting on a shaky foundation, shaky, fra.

It’s like a high performance car with loose wheels. As soon as you take the first time, things will fall apart. And in, in this narrative, the team rebuilds from ground up. You know, model of a scaling fallback, routing, circuit breakers, caching, drift detection, anomaly alert safeguards against prompt injection and data poisoning.

So borrowing some of the reliability. I would say good reliability patterns from webscale, applications, cloud, et cetera, but also. Responding to the new type of reliability challenges that AI systems introduce. And after that overhaul, you know, the book shows that they achieve 99.9% uptime with significant latency reduction.

But it’s really not about the metrics. it’s a cultural learning, which is the team [00:21:00] learn the reliability. Is the ultimate form of trust. Look you mentioned earlier, now people are pouring their heart out to open ai, talking about mental health issues and whatnot. Imagine having you know, great therapist at your disposal of 24 7.

And when you need them the most, you’re not able to reach them. I mean, that’s, I would consider breach of trust. And so it’s not a technical outcome. You have to take it as a leadership discipline. In ai, you don’t earn confidence through marketing. You have to earn it every time the system stays up because predictably and for its users, they, they should feel that they have the AI systems have their back.

Guest 2: Yeah, a hundred percent. I would say, you know, I mean, if, if there’s a soundbite. Treat resilience as a product requirement and not as some back office upgrade that you’ll get to later. Right? There may not be a later especially when you’re, you know, dealing with healthcare. Like the example in the book that we use, patients [00:22:00] don’t care which large language model or provider you choose.

They care that, you know, reminders, fire drive, routes, and, you know, audits are possible. That they get the right prescriptions at the right time and the best treatment possible. So you are better off baking fault tolerance and auditability into the acceptance criteria from day one, and using that reliability story to earn both regulatory compliance okays, as well as, you know, winning over the customer trust for your product or, or your service.

I would say the analogy that Bish provided about about ai, you know, being a great engine, sitting on shaky infrastructure is like a high performance car with loose wheels. It’s apt AI can and actually tends to dazzle out the gate. I mean, we all remember our first experiences with chat GPD or you know, any of the other bots, and it was like a wow moment, you know, maybe during [00:23:00] tests had been.

You know, taken care of. And we were with this, you know, super powerful thing that could answer any question in the world. So you, you can be easily enamored by all of that. You know, you have a rag infuse chatbot, you are built or a voice bot that starts responding very empathetically, however, without, you know, the proper evaluations leveraging humans in the loop, leveraging AI itself as a judge.

Having, you know, golden data sets and edge cases that you can test and capture issues against, you cannot and will not be able to rely in terms of putting your, offering, your service out in the field, in production at scale. And those are things we have learned by doing working on large scale AI products.

To your question, Luke, making quality, reliability part of that cycle and not an afterthought. And ensuring that you have an evaluation based approach to go no go across, both [00:24:00] scaling and production as well as making it available is critical to success in real world scale deployments.

Luke: It makes sense.

And like there’s something too about, especially around like ai prompts that where I, I feel like it can be really useful for end users around I mean, I’ll touch on the medical example, right? you know, so many people are kind of make assumptions right around okay. If I hear this from a provider.

this is what it is. And you know, that’s that. Whereas, you know, in reality, a lot of times you kind of have to be your own advocate. So I’ve seen prompts do this too, where like they’ll suggest hey, and you also might wanna consider blah, blah, blah. And so that reliability piece is like super important if you’re gonna be suggesting have AI suggesting things.

But the power in that is like one of those things where. I feel like it can remove the fear element a lot if people realize, yeah, this thing is like helping to empower me to advocate for myself. I, I don’t know if you guys have any thoughts on that, but it seems like a, a low hanging [00:25:00] thing. As long as the reliability is and the quality of what you’re getting is good and it’s not telling you stuff that like is just way out there.

Guest 2: No, absolutely. I mean, ai, AI is very much a copilot for all of us as individuals today, and I mean, most of us don’t even realize. Much like we used to say, just Google it. You know, AI has become sort of that companion you reach out to, and, and more so in the generation that’s coming up. I mean, they are learning to be with ai.

They’re surrounded by AI either across applications or, you know, leveraging cho chat and voice box. So it’s very much integral to what they do, and that makes it all the more important that it better be reliable. And not be, you know, misleading or, you know, create issues or situations that can put people in harm’s way as well.

Luke: Yeah,

Guest 1: think. And if I might add, you know, similar to the cloud journey, cloud is supposed to make your infrastructure success reliable. Cloud themselves. Sometimes they’re not reliable, so a lot of. Enterprise companies, they [00:26:00] go with multi-cloud infrastructure. Similar thing, very similar thing is happening in AI world as well, where you will have to have reliability story when, let’s say open AI down, you need to maybe rely on Google or vice versa.

It’s not an option. You, you have to think about ai foundations also in very much similar terms of cloud level reliability.

Luke: That makes sense. No, I think those examples are like really great for people to do. you know, especially if you’re trying to change the mindset of something, relatability is like super important.

why are, why do you guys see are there like any other common pitfalls that people should avoid from, from your guys’ perspective?

Guest 1: You know, there are a lot of headlines about AI as a. Augment as a replacement versus augmentation like Mahesh was referring to from our experience, the most important truth about Gene e AI is the best systems are built for cowork, not replacement.

and you know, the, in the story itself, the patient support chatbot, which [00:27:00] is a more autonomous AI application, and the clinical copilot, which is a hybrid application, they evolve together. The AI handles high volume data, heavy work. Like triad, suggestions, reminders, document drafting, et cetera, by clinicians.

They focus on judgment. Empathy and context. The system explains every recommendation and leaves the final decisions to the humans. And this is critical, and let me be clear. There’s not to say that humans can keep working exactly the way they were working before. They’ll also need to adapt and evolve.

But you have to think sort of collaborative. A co copilot framework versus a complete replacement F framework. Sure. Also highlights real pitfalls. You know, when you know the no team no bridge team. Initially assumes trust will automatically come. It doesn’t even right now. A lot of people have, might even add borderline PTSD in talking to some of the older generation of chat bots sound very robotic.

They don’t understand, they don’t really [00:28:00] resolve. So you have to continuously collect the feedback, close the loop, and slowly, surely, incrementally, sometimes substantially increase the trust and engagement. It’s only when you completely redesign the workflows. Around collaboration, around transparency, that things will start clicking, and AI becomes a true assistant.

Somebody you can, you have it by your side and not a, black box. So, you know, again, to recap the deeper takeaways both humans and AI will, how to evolve together, workflows, training expectations, they have to adapt continuously, and when they do ai. Will stop being a threat. At least people will stop pursuing it as a threat.

It’ll become force multiplier for human expertise. Everybody and all of us here, they’re great at, you know, what they do and are now emerging. Doing it 10 times, a hundred times, thousand times. AI can enable you to do that, but you are at the center.

Guest 2: Yeah. I tend to agree. I mean, it, it’s, it’s about amplification and not replacement.

So it’s not about [00:29:00] replacing humans, it’s about augmenting them, extending them, making them more efficient, making you know, us as humans, more knowledgeable, more reliable with our own work, with our own responses. So in the book, you know, when we look at Nova Bridge, they go ahead and explicitly pair a patient support chatbot with a clinical copilot.

And this is, you know, for the staff to get leverage, leverage out of the copilot and not necessarily, you know, displacing them from what they do. They can actually now attend to higher order issues higher order problems than the routine, the mundane stuff, right. The alignment to the recommendations made by the clinical staff actually helps improve the system as it matures.

And it’s actually a better and a more tangible way to show overall benefit both to the staff and the patients. That’s what we are proposing in the book. [00:30:00] Now, in terms of your question around common pitfalls, I would call out, you know, things in addition to what Bish mentioned. Things suggest, you know, skipping consent, explainability, you know, drawing a curtain over it, over promising autonomy and, you know, not really delivering on that.

Then also, especially, you know, when you look at bias and hallucinatory responses, and especially the former ignoring equity for, you know, some of the more underserved or marginalized groups that becomes very important also to take note of. So the, the key takeaway is the common evolution that Bish mentioned.

AI enables humans. In turn, you know, humans make AI better, accurate, and more knowledgeable. And that sort of AI virtuous cycle is what you want to aim for.

Luke: Awesome. Yeah. And in the book too, you, you outlined prompt, engineering, rag and fine tuning as optimization levers. Which one of those is more most overlooked in enterprise use from what [00:31:00] you guys see?

Guest 1: From my perspective, I think a couple of things, they are sort of underrated all the time. First is prompt engineering as a discipline control interface. You know, you have to treat it like code. You got version a structure it, how safety scaffolds, they’re not to be treated as throwaway strings. And second, has to be the default pattern in regulated and fast moving domains. That’s how you’re gonna be able to keep your answers grounded in truth and not a faint memory. Fine tuning has its place, but it’s more about pre precis precision not a hammer. Lot of teams, they jump too early into fine tuning.

The smarter path we recommend is start with the cheapest lever first. Refined prompts, improve grounding. Drift, et cetera, and find you only when you know those lever stop moving the needle. And you might be surprised from ma ma many majority of the use cases prom engineering combined with rag, you know, gets you almost all the way there.

And this is how you, you know, go [00:32:00] fast also and improve on your ROI.

Guest 2: Yeah. You know, all of these techniques are important, right? Whether it’s just, you know, making sure that you’re prompting correctly. That, you know, you have infused rag knowledge into the system. You know that you have compliance checks checks for bias hallucination and including fine tuning.

Now, you know, many might be tempted to jump to fine tuning too early. Is our opinion the fastest path to value, however, and keeping the risk profile low. Is operate on your cheap levers first. You know those that’ll give you the most bang for the buck. The 80 20 rule still holds true for most cases, right?

And that’s around prompts and grounding. And then fine tune only, where, you know, those don’t close the gap for you. You know, maybe your domain is so specific. Let’s say you’re in the legal industry or you know, some of the domain where you just need that additional fine tuning. To make that model response so much more accurate and [00:33:00] relevant to your particular industry.

The sequence pays off in, you know, like ish mentioned in time to trust and time to ROI when you go with prompts and good grounding. I would also add that the real issue often and which folks overlook till it’s maybe a bit too late in the cycle is around data. You have in any enterprise, you have multiple conflicting sources of data.

Incomplete or incorrect or outdated information. You know, ggo, garbage in, garbage out. No AI can save you if you are tossing it garbage as feed, right? So the best of prompts and rack grounding, or even fine tuning can only do so much if the inputs are bad to begin with. So focus on your data, your data pipeline.

How you curate it, how you set it up how you cleanse it you know, the quality of the data that you have. All of those are extremely important.

Luke: That’s great. No, that’s great. Great. Great insight and yeah, I think so much of this applies to so many things outside of AI [00:34:00] too. I think it’s a, a really, really good good explainer.

Just kind of looking ahead what separates from your guys’ minds, organizations that will thrive in an AI first world, from those that will fall behind.

Guest 1: I believe the folks who will thrive are the ones who are looking at ai not as a tool, but as, as an inflection point. When you look at these capabilities of systems with that lens, they will be building systems that are observable, resilient, ethical, where every model behavior maps to a real business outcome.

They don’t, don’t just deploy models. They will design feedback loops, guardrails, and accountability into everything. I believe winners will understand that speed without stewardship is a mirage. You will take a step forward, maybe too fast, but then you’ll be forced to take two steps backwards when things fail.

You have a lawsuit, brand reputation risk, et cetera. And the real edge really comes from pairing velocity with trust from building teams. [00:35:00] And technology that earns confidence. And it’s not just about chasing headlines. The flip side is also too, the ones will fall behind. They’re still treating AI like an experiment too.

Scattered no focus at all. Or a side project. They’ll realize too late when that, that the game is not about adoption, it’s about alignment, aligning people, systems, values around. What responsible intelligence looks like in practice. At the end of the day, I strongly believe future belongs to the builders who are bold enough to move fast, but wise enough to move with care and, you know, you know, closing.

You know, we are big on frameworks. We have learned hardware through a lot of our experiences and experiences, and the book lays out. Quite clearly a lot of the frameworks you could adopt to your personal situation, your enterprise organization and, and, and move consistently and confidently towards your business goals.

Guest 2: Yeah, a hundred percent. I mean, the, the advantage goes to the AI [00:36:00] first leader or AI first leaders in an organization who pair, you know, alacrity and speed with stewardship. These are the people who are embedding AI into strategy, into product, into operations, into culture, into investing in their people.

Measuring the ROI measuring, you know, the responsible use of ai, right? So the full cycle of things that you would do, especially when you’re embarking on a. Rather transformational journey at an enterprise level and you know, also staying ahead of the ever changing and often undefined state of things, right?

So the leaders who are gonna be successful are those who can operate with, you know, hands full of ambiguity, if you will. Right? That’s all they have to latch onto. And it, it almost feels like things are slipping out of their hands. All of the time and trust becomes, in my opinion, a very competitive currency.

By that I mean the trust that [00:37:00] you gain by leveraging these technologies, definitely outside with your customers and your clients, but also inside in terms of how you execute on that vision and how do you lead from the front. That trust becomes an becomes a very competitive currency in terms of even attracting the right talent.

That comes in through your doors, right? Hesitation, like Bish pointed out, can prove to be costly. There’s no doubt in our mindset, you know, this is gonna be a big integral part of every enterprise. So the sooner you embrace it, the sooner you get going with it, the better off you are. And then finally, you know, the one point I’ve been thinking about is, you know, what applies to the rank and file employees.

In an organization also applies to the companies and its leaders, you know, so we all keep hearing this cliche. It is not AI that will replace you but a person leveraging ai that will take away your job, right? I mean, you wake up to that cliche almost [00:38:00] every other day. And I would, you can see that about that the leaders who help their companies and departments embrace and embed ai.

We’ll march ahead very quickly and the others will similarly be left behind in the dust.

Luke: Yeah, I think stewardship a very good way of framing it. I think that’s a, that’s a key takeaway too, from this, and you guys, I really appreciate both of you taking the time to make it here for the conversation today.

I’d love to really let you guys also one the book’s out. So like, maybe you guys can let people know where to get it, and then also for folks that want to find you guys online and and reach out or, or, or learn more from you guys on an ongoing basis. Maybe you guys can share too where people can find you.

Guest 1: Sure. So book is out. It’s on Amazon. You can look for AI first leader. Would really appreciate what resonated with you. What didn’t, any critical feedback that you have? I can tell you from early reviews and some of the leadership that we have talked to, they have found it to be quite insightful.

Help them double down on some of the things that. [00:39:00] Thinking to be true. It challenge their perspectives in some other ways. And they believe they’re marching on the business success more confidently with AI adoption. You can find me and Mahesh both on LinkedIn, happy to connect and jam about anything related to ai.

This is what we do. This is uh, who we are.

Guest 2: Yeah, absolutely. I mean, uh, feel free to reach out like Bish pointed out and you know, definitely get the book if you can. One thing I wanted to point out, all author proceeds of the book go towards pediatric neurological research. So it’s all written for a good cause.

Every penny, you know, there’s nothing in royalty that will retain It goes to charities that are focused on advancing neurological. Disease re research for uh, children. So it’s a good cause. There’s a lot of good stuff in the book to learn from. Go get it.

Luke: That’s fantastic. No, I, I love that.

and I just, you, you guys are a great duo too. I, it just, just having this conversation, you both kind of like, both compliment each other very well [00:40:00] and, and yeah, I think people go check out the book. There needs to be more, more, more content like this out there for people, because I think you guys covered it well too.

Like right now your options are to like tune into like, you know, a lot of fear-based things and people have real concerns about this, but people also really want to know like. What do I, how can I use these things? Like, and, and it’s a mindset like you guys said. So I really appreciate you guys taking the time.

Go check out their book. And I’d love to have you back sometime too, to kind of check back in and, you know, whenever you have anything else coming out let us know ‘cause we’d love to have you back on. Our pleasure. It’s been a

Guest 1: pleasure, Luke.

Luke: Yeah, thanks guys. Thanks for listening to the Brave Technologist Podcast.

To never miss an episode, make sure you hit follow in your podcast app. If you haven’t already made the switch to the Brave Browser, you can download it for free today@brave.com and start using Brave Search, which enables you to search the web privately. Brave also shields you from the ads trackers and other creepy stuff following you across the web.