Host: Luke: [00:00:00] You’re listening to a new episode of The Brave Technologist, and this is another one we recorded live at the AI Summit in London. This one features Joel McKelvey, who’s VP of Product Marketing at Glean, where he helps organizations turn AI initiatives into practical, metrics-driven business transformation.

His work focuses on how enterprises move from experimentation to execution using AI agents, enterprise search, and trusted data foundations to improve productivity, decision-making, and customer and employee experiences. In this episode, we discussed how context and a unified platform are key for enterprise, what metrics are important for AI across the enterprise, and how members of teams can change fear into confidence with AI.

And now for this week’s episode of The Brave Technologist.

​

Host: Luke: All right. Joel, welcome to The Brave Technologist. How are you?

Joel: Good. Thanks for having me. I appreciate it. It’s nice meeting you on this rainy day in London.

Host: Luke: Yeah, yeah. So we’re here in London at the, uh, AI Summit. and I think you spoke here yesterday, right?

Joel: I did, yep, yesterday morning.

Host: Luke: What was your talk about, and what was the point that you were really trying to get [00:01:00] across?

Joel: it was less my talk and more a talk from one of our clients- Okay … Ericsson.

Host: Luke: Oh, cool.

Joel: And Carl from Ericsson was talking about the journey that they took at Ericsson to deploy what will ultimately be tens of thousands of, AI licenses out to their clients.

Host: Luke: Cool.

Joel: And why that is the new norm, and what you need to think about when you’re doing that. So, I asked him lots of questions about, what pitfalls are like and what people should be thinking about and advice he should give. It was very, very, interesting and informative for me. and he surprised me with a few things, and maybe we’ll get to them.

Host: Luke: Yeah, yeah. Sure, sure.

Joel: Super interesting to hear his point of view on how to roll out AI in, at that scale.

Host: Luke: Yeah, I mean, Ericsson’s, you know, small, small name,well, and you’re Glean, and, and Glean kind of positions itself as trusted context and intelligence layer, right, for enterprise AI.

what does that mean in practice, and what is the context layer… Why does the context layer matter more than the model itself?

Joel: Yeah, so thank you. Excellent question. So one of the things that [00:02:00] we need to think about with AI is how suitable it is for the enterprise and for work. Mm-hmm. And so Glean is fundamentally work AI.

Mm-hmm. And it is focused on a corpus of data- Mm … which is really enterprise data.

Host: Luke: Okay.

Joel: And when we talk about it as being that trusted AI interface into the enterprise, what we mean is it’s foundationally based on a context layer, and that context layer contains the information that your business has.

Host: Luke: Okay.

Joel: But not just the information, but how people access it and use it, the identities of the people who are accessing and using it, the permissions models around that data. And that allows you to get, what we’ve been trying to do in a lot of industries for a while, which is a unified view-

Host: Luke: Yeah

Joel: into what’s going on in a business. And that is a non-trivial task.

Host: Luke: Right, right. No, absolutely.

Joel: And that’s why we, we are positioning ourselves as that trusted enterprise layer, because that’s what we’ve a- attempted to accomplish. And getting that context together- From, from an enterprise is, a task that we’ve worked [00:03:00] for the last, almost eight years, seven plus years on to do it clean.

Host: Luke: And that unified layer, right? so that’s kind of talking once you know all of the players and the game and kind of the input, is that where it- you can use that unified vi- like, where do you go from there with it?

Joel: Yeah, so once you have visibility into the systems of record in the enterprise- Right, right

once you have visibility into the systems of record and how they’re accessed, you can answer some very complicated questions. Mm-hmm. Uh, for example, if you have a project that a bunch of people are working on, you probably have a list of names of those people who are working on that project, but the true people working on that project are the ones accessing the project docs-

Host: Luke: Yeah

Joel: reading those project docs, updating those project documents, and those documents can be spread across, chat channels. It can be in, let’s say, software engineering, it might be in Jira. Mm-hmm. If you’re doing project management, it might be in Asana. Sure, sure. It might be in a, the Google Suite or the Microsoft Productivity Suite.

And so, really this understanding of how that project is [00:04:00] actually operating-

Host: Luke: Yeah …

Joel: is part of what is indexed and understood in Glean, and that’s a very powerful thing to be able to know about inside a company.

Host: Luke: Definitely, definitely.

and you’re not new to this, right? You’ve spent 30 years kind of watching enterprises adopt new technology, and, what’s different about the way that AI is landing in organizations compared to previous waves of tech?

Joel: And so another good question that I’ll, I’ll turn on its head.

Host: Luke: Okay.

Joel: Um, so the first thing is, in some ways it’s quite similar. Yeah. b- but what we’ve been thinking about is AI as experimentation. Mm-hmm. And that’s not the case anymore in the enterprise.

Host: Luke: Right.

Joel: Now AI is production organizational-wide, and so that means we have to think about it the same way we would think about deploying any new productivity tool to our entire workforce.

Host: Luke: Right.

Joel: And that means, the challenges there are huge. Yeah. And those challenges are ones of human beings: adoption, trust, uh, regular usage, understanding workflows, getting security clearance to- Sure … from all, from your security team to do [00:05:00] things, getting permissions to access data source, whatever that data source might happen to be.

Those things are time-consuming and hard. Mm-hmm. But they’re no different from other IT projects. Mm-hmm. And we know that we can kind of power through those.

Host: Luke: Yeah.

Joel: What’s different about AI is the rapid rate of change.

Host: Luke: Okay.

Joel: Right? It has come just as a tsunami to organizations that are looking at AI. We are really at that stage of, we were kicking the tires a year ago, and now it’s really in production, and, it’s because enterprises see that value, and they see the potential as well.

Host: Luke: Yeah.

Joel: But, when you’re an IT person with a historical IT project for rollout at scale of- months long.

Host: Luke: Yeah.

Joel: a rate of change that is measured in weeks or months is too fast for you to accommodate, some of those changes. And so what we’re seeing is some of those challenges are is, that arise are how quickly can an enterprise actually adopt not just the tool, but the [00:06:00] rate of change of that tool and be comfortable and confident with it that it remains secure, that it hasn’t changed so much that people won’t use it.

​

Host: Luke: Yeah. No, and I think that’s a really valid point, like, people kicking the tires about a year ago, and now I think, a lot of the costs come to bear around these things, and it’s like, okay, everybody was, like, talking about you gotta use all your tokens, whatever, but now it’s okay, these things have real cost, right?

And now it’s kinda naturally progressing to like, okay, measurement, right? Like measuring, like what metrics are our enterprises, looking at as like kinda transformative or the right m- metrics to be kinda weighing AI initiatives around? And, and are enterprises seeing a lot of [00:07:00] ROI at this point or is it still kind of…

I guess what I’m getting at is, is a balance between, experimenting, adopting, and then, ultimately the CFO saying, what are we doing here?" You know?

Joel: Yeah. So, so I think that’s a really valid point because if we’re deploying this into a, an organization, an enterprise, it has to pay for itself.

Host: Luke: Right.

Joel: But it doesn’t just have to pay for itself. It has to do more. And so, the organizations I talk to, and I talk to small ones and large ones all over the world, are really looking at it in three dimensions. they might look at it as, a tool for personal productivity that’s on everybody’s desktop.

Mm-hmm. You can think back to, like, the early days of ChatGPT.

Host: Luke: Sure.

Joel: If it’s on my desktop I can ask it questions, it can help me, I can become more productive. We see these co-pilots and productivity suites are very similar to that. that’s essentially that goal. And that can save hours in a day. days in a month, it can be a very powerful thing.

Essentially, though, when we look at those productivity gains, I think the way that the current [00:08:00] solutions are priced, those productivity gains essentially pay for the solution.

Host: Luke: Mm-hmm. Okay.

Joel: Right? And that’s great, so you’ve made back your money.

Host: Luke: Right, right, right, right. Which,

Joel: which, but AI is much more than that.

Right. And people aren’t investing in it so that it’s a, you know, a one-to-one ratio. Right. They’re investing it because they want to 10X something. Yeah. Right?

Host: Luke: Absolutely.

Joel: And the 10X comes when you move an order of magnitude past individual productivity into team productivity- Mm-hmm … and workflows- Mm-hmm

which is kind of the second way that people think about deploying, and those measurements are in, departmental level metrics. Yeah, yeah. Like, am I answering tickets faster? Am I shipping product more quickly? Are my sales reps all more productive? Yeah, yeah. Right. right, producing more revenue.

those types of, of business metrics at the departmental level are where, companies are now, like, really going past that desktop and thinking, “I need to start thinking about agents or other ways of automating workflows- Mm-hmm … rather than just simply making a person more productive.” And the third way we see people thinking about it is across the whole [00:09:00] company.

Host: Luke: Yeah.

Joel: And again, I think if you deploy across the whole company, you get another order of magnitude benefit, and that is tying department to department, where, an engineering department might typically be isolated from how a sales call goes out in the wild- Yeah … right, when you’re trying to sell a product.

but with AI, the feedback loop for that can be incredibly truncated in real time. Engineers can hear right away whether or not their product is being adopted and can themselves understand why it’s not being adopted or why people love it.

Host: Luke: Mm-hmm.

Joel: And that information sharing, that ability to query those types of pieces of information that spreads between departments knocks down a lot of barriers.

Host: Luke: Yeah.

Joel: And I think for some companies, that’s the Holy Grail.

Host: Luke: Yeah.

Joel: But you asked, like, where that value lies. when you’re trying to deploy in the enterprise, you need to have a value hypothesis- Yeah, sure … of like w- what I’m going to, what I’m going to get. And not only do those things sort of build in the amount of value you can receive, if you go global, you can just get more orders of magnitude.

But they also have very, you know, different corporate [00:10:00] metrics. Mm-hmm. When you’re talking about the whole company, we’re talking about whole company’s revenue, whole company’s cost, whole company’s risk.

Host: Luke: Mm-hmm.

Joel: Um, where you’re talking about an individual, we’re talking about saving a few days.

Host: Luke: Right.

Joel: Uh, people who are brought in, the new s- uh, CBAIOs or CAIOs, um, they often have a mandate.

They’ve been asked to do one of those three things.

Host: Luke: Right.

Joel: It turns out you can’t really separate them very well.

Host: Luke: Yeah.

Joel: But, uh, um, but it is valuable to think about what is the payback, whether you’re thinking about an individual, a team, or the whole org. what are those metrics I’m trying to hit, and being able to measure and make sure that you’re getting to

Host: Luke: hit.

Yeah, and the baselines are there, right? Like, and so it’s basically you know, looking at, okay, are, are we bringing this in for productivity or for revenue or whatever, but looking at it from like, kind of like a team or an org- a higher level, like, so you can see, ‘cause these things basically have like compounding returns or, or something like that.

Is that fair?

Joel: That’s right, yeah. Yeah, I mean, the, the more global you get in the organization, the more your returns compound.

Host: Luke: Yeah.

Joel: I mean, we talk a little bit facetiously about 10X-ing the person.

Host: Luke: Right.

Joel: Right. And so we can talk a little bit facetiously about 10X-ing the person, [00:11:00] 100X-ing the team, 1,000X-ing the company.

Host: Luke: Right.

Joel: Now, that’s hyperbole.

Host: Luke: Right, right, absolutely.

Joel: Right?

Host: Luke: Yeah, yeah,

Joel: yeah. Right? But, that concept I think holds true in the deployments I’ve seen.

Host: Luke: Yeah.

Joel: When you get to… And, uh, we’ve got customers with tens or hundreds of thousands of Glean users.

Host: Luke: Yeah.

Joel: Right? What you see is just this monumental increase in the value across the organization that’s very, very compelling from an ROI perspective.

Host: Luke: Well, and I wonder, too, like, I mean, given your point of view, you’re seeing these enterprise adoptions, are they following a specific pattern? Are you seeing interesting is it happening across orgs? Is it typically- Mm. Mm … happening kind of on the technical side and bleeding over, or more on the support side?

is there anything, like, say I’m an enterprise and I wanna do this. Are there any pointers based on what you’re seeing?

Joel: Yeah. So we do see things like that. Yeah. Um, um, we used to really see it started in software engineering armies- Uh-huh … because this is really, this is a highly technical product.

Yeah, it’s

Host: Luke: logical, yeah.

Joel: That used to be very, you know, lots of knobs and tuning and difficult to set up and the, you know, API driven and those sorts of things. that is [00:12:00] absolutely shifting.

Host: Luke: Yeah.

Joel: Right? AI today is for everyone.

Host: Luke: Uh-huh.

Joel: Um, you know, if my mom uses it, it’s probably for everybody.

Right, right. Um, she’s a smart lady, but she’s not in tech. And so, like, this ability to use AI has changed, and I think we saw a real shift in the industry when we got to the ability to ask questions naturally of AI.

Host: Luke: Yeah.

Joel: That was where, like, the GPT of the worlds took off. Oh, yeah.

Host: Luke: I… Yeah.

Joel: now what we’re seeing in enterprises is a real shift, and we see it particularly, among our clientele, of AI that understands how the business operates-

Host: Luke: Yeah …

Joel: gives you good business answers. Yeah. And that’s key to using it in the enterprise. Yeah. If it can’t understand what’s going on in the business, simply because it has too small a corpus to look at- Right

it doesn’t have enough context, or it’s isolated to individual desktops, then you end up with, okay answers.

Host: Luke: Yeah.

Joel: I think you can get moderate productivity gains- Mm-hmm … but, as a user, that doesn’t thrill me. Right. That I might save 15 minutes. I mean, I, I’d love to go get another cup of coffee- Sure, sure

[00:13:00] and not be so rushed.

Host: Luke: Yeah.

Joel: but I think what really happens is when I can come in the morning, see a list of probable tasks that are mined-

Host: Luke: Yeah …

Joel: from my information that day, and half of them have a do this task for me button-

Host: Luke: Yeah …

Joel: and I can just pound on them right now- … and, like, get half my tasks off my plate-

Host: Luke: Yeah

Joel: that is incredibly powerful.

Host: Luke: Yeah.

Joel: And, I used that this morning when I got up with Glean because I’m time shifted from California- Right … ‘cause we’re in London.

Host: Luke: Right.

Joel: And I was like, “Oh, these are all the things I missed.”

Host: Luke: Yeah.

Joel: And, roughly half of them in Glean had a button that said, “Let’s do that right now,” and I just said, “Great.

I’m gonna just keep pressing these buttons-

Host: Luke: Nice,

Joel: nice … until that stuff gets taken care of.”

Host: Luke: so the, the end user gets a, kind of a, a weighted, a priority list too, but also kind of across the teams and across the org too.

Joel: Yeah, and so it’s that understanding of the organization-

Host: Luke: Right

Joel: combined with the understanding of the individual-

Host: Luke: Mm-hmm …

Joel: it gives you a knowledge graph, the context-

Host: Luke: Yeah …

Joel: to actually do a task.

Host: Luke: Yeah.

Joel: but when you have a rich context layer, like we have in Glean, you can take it to that next level, [00:14:00] and what I’m describing is proactivity.

Host: Luke: Right.

Joel: It’s like find it- Uh, surface it to the user

Host: Luke: Yeah

Joel: ask them if you want to take action on it, take that action, have them confirm the action before anything big is done.

Host: Luke: Right, right,

Joel: right. But, but save somebody real meaningful time, that solves the adoption problem as well.

Host: Luke: Yeah,

Joel: sounds like it. Is because it only takes five minutes this morning for me to be sold, right?

Host: Luke: Right, right.

Joel: It’s the best thing ever.

Host: Luke: Yeah. Awesome, awesome. Yeah, I mean, so when an agent’s kind of making decisions and taking actions on behalf of the employee, trust is, like, really important in that front. How do enterprises build that trust and what have you guys seen that kind of can break that trust?

Joel: Yeah, there are many ways it can be a problem. Right. Um,I think originally we had a lot of pushback on the tools themselves, because they can, in some cases, be a little bit opaque.

Host: Luke: Sure.

Joel: So it’s a little bit hard to track them. So you see these, the rise of some of these, chattier models that say, “I’m thinking about this.

Now I’m gonna think about this- [00:15:00] Right, right … now I’m gonna think about this.” That has solved a lot of the trust problem because people are concerned with what is it up to?

Host: Luke: Yeah.

Joel: but the second, and I think the most important thing is, making sure that the answer’s right- Mm-hmm … is really a big issue.

Host: Luke: Yeah.

Joel: And without access to that context layer, without understanding the business, and also without guardrails, you need some- Sure, sure … some of the answers are just not as good.

Host: Luke: Right.

Joel: Right? We see, those with access to rich context in the enterprise, you see the answers are preferred two and a half, three times more than answers that don’t have access to that rich context.

Yeah, yeah. And that means that a user will trust it, come back to it, use it again. Mm-hmm. And that means, for the business owner, that means they’re actually gonna get those productivity gains. They’re not gonna be ephemeral.

Host: Luke: Yeah, absolutely. That makes a lot of sense. I think, what does a well-designed kind of human-in-the-loop look like f- for a genic AI at the enterprise level?

Joel: Sure. So good question. because there’s lots of ways this can go wrong. But the w- [00:16:00] the way it goes right, if y- if we think again about, like, that context, that context comes from access to the systems of record. Yeah. The corpus of information that a company runs on. Really, the DNA of the business, right?

Yeah. Where things go wrong is not only when you have wrong answers, but where you then take action against those systems of record-

Host: Luke: Mm-hmm … and

Joel: instantiate bad information- Right … back into where it might be used by others.

Host: Luke: Right.

Joel: That is a huge concern.

Host: Luke: Yeah.

Joel: And so human in the loop is there to make sure that does not happen.

Before you take action, before you overwrite, or before you write something into a system of record, someone should probably take a look at it. Yeah. Even for relatively minor things. and until we get to a point where that trust is built with the employee population, that human in the loop is absolutely the right place to be.

Now, we could also talk philosophically-

Host: Luke: Yeah, sure …

Joel: based on, um, some of the Anthropic stuff that’s come out about, like, what things AI should never do.

Host: Luke: Right, [00:17:00] right, right, right. And where

Joel: we would like, like, an, an AI, uh, constitution to make sure that people aren’t violating societal and ethical norms.

Host: Luke: Yeah.

Joel: But I think in the enterprise context, what we’re really talking about is- Um, use real data-

Host: Luke: Right …

Joel: to give real answers, and don’t put wrong data out to others.

Host: Luke: Right. Right. Right. Make sure… Yeah, exactly, and, and, and where you put the data out and all that too. Like, when, when you guys are looking at this, is it typically, do you have, like, a representative from a team that kind of, helps w- with that human in loop part?

Is it just part of the process that’s built into the software itself? Where does that warm start kind of like, hand off to being more automated?

Joel: so another good question, because there’s sort of two models that you would look at for doing this, and one is, uh, I’m a person who wants to do something.

Host: Luke: Right.

Joel: And, what I would like to be is prompted when I’m about to write something so that I can be sure that it’s writing properly. And, so I just wanna review before it goes somewhere. Think of me as the orchestrator of my day- Yeah … and I would like to be able to make sure that that [00:18:00] orchestration is going on and guide it to make sure that it does the right thing.

But there’s a second, thing that, AI is unlocking, and that is the building of things for others.

Host: Luke: Mm-hmm.

Joel: So, in a low-code, no-code, agent building environments, where you can generate effectively a tool for others-

Host: Luke: Mm-hmm …

Joel: those tools need to be able to put in a human in the loop moments-

Host: Luke: Right

Joel: so that you end up prompting the user.

Host: Luke: Mm-hmm.

Joel: And, so it, I mean, it, it sort of depends on how you think about it.

Host: Luke: Yeah. Yeah.

Joel: But, um, when you’re building for others, this is one of the things that you need to keep in mind-

Host: Luke: Yeah …

Joel: is at what point do we pause and make sure that user is aware?

And so in Glean, if you’re using our Glean assistance-

Host: Luke: Yeah …

Joel: or, um, and you’re using that for individual productivity, we try and write a file, write a ticket, overwrite a ticket, something like that, then we’re absolutely gonna prompt.

Host: Luke: Yeah, yeah, yeah.

Joel: Right? Makes sense. And, and we’re gonna make sure that that’s, that’s the appropriate piece of work.

but then you can also build an agent in Glean, and then we’ll suggest, particularly if you’re doing write, that you add in a prompt the user component- Yeah, yeah … there so that the, so that they get that same type of [00:19:00] maintenance.

Host: Luke: That’s awesome. That’s helpful. Yeah. I know we’ve been talking about productivity and the wins there.

sometimes employees kind of can feel threatened by that too. you know, how should leaders kind of position AI internally with their own staff to kind of mitigate that?

Joel: I think it’s a valid concern. I empathize with any employee that’s feeling that, that concern, ‘cause it is, it is absolutely valid.

I think the concern evaporates quickly-

Host: Luke: Yeah …

Joel: when the productivity is real.

Host: Luke: Mm-hmm.

Joel: And the things that the employee doesn’t like doing are taken off their plate-

Host: Luke: Right …

Joel: and they can focus on the things they do like doing. And maybe that’s an overtly optimistic way of looking at it, but the tasks that I automated this morning-

Host: Luke: Yeah

Joel: I was so glad to be rid of.

Host: Luke: Right.

Joel: And like-

Host: Luke: Right, right …

Joel: it did not make me at all sad, and I was not worried about my job- Oof … ‘cause I had plenty of other tasks.

Host: Luke: It feels like it’s where, where you’re feeling market fit across d- different parts of the enterprise or, or different functions or, or people are actually kinda getting the experience, uh, [00:20:00] beyond the talking about it, right?

‘Cause I feel like this is what most of the past year has been, has been talking about, “Oh, this is the revolution,” blah, blah, blah, blah, blah. And, and then, you know, there’s a lot of fear around that, too, and, uh, but actually getting in there and feeling the win is what it’s all about, right?

Joel: it is, and look, I think we hear a lot of hype about will there need to be fewer people in certain job sectors.

I call it hype, maybe the right word is philosophical arguments- Right. … because they haven’t, because they have not played out yet. Yeah,

Host: Luke: yeah.

Joel: I think every human being likes being augmented to do better.

Host: Luke: Yeah. I,

Joel: yeah. And if AI is doing that for people, the fear falls away, the joy comes into being-

Host: Luke: Yeah

Joel: people embrace it, and that’s where we see, you know, really wide adoption- Yeah … is when people look at it and they think, “Oh, this is for me.”

Host: Luke: Yeah,

Joel: yeah. And, and I think that as you roll out into the enterprise, rolling out something that has high adoption, that has high user trust, that sparks that joy in them when they use it, makes your life as an IT person [00:21:00] much more of a trying to meter out how quickly it’s deployed rather- Right

than to force people to deploy it. That

Host: Luke: makes sense. That makes sense. if you had to pick one thing enterprises are getting wrong about AI right now that they’ll look back and maybe regret, what would that be?

Joel: So let’s talk about enterprises in the market in general.

Host: Luke: Yeah, yeah.

Joel: Right? We are extremely over-rotated on the excitement around new models.

Host: Luke: Yeah, absolutely. And

Joel: new models are very important-

Host: Luke: Right …

Joel: don’t get me wrong. but an enterprise that’s sitting and trying to figure out how to adopt the latest model is wasting a lot of time-

Host: Luke: Yeah …

Joel: because it’s gonna only gonna be latest for a month- Right … before something else leapfrogs it. And so I think, instead of over-rotating on what’s that model, an enterprise needs to take a risk-adverse, future-proof, uh, way of looking at that- Yeah

and say, “What do I deploy that is model agnostic-

Host: Luke: Right …

Joel: that can accommodate new models when they come out but doesn’t lock me into any individual model or family of models?”

Host: Luke: Yeah.

Joel: [00:22:00] And this is where I think there’s some distrust with the LLM providers, because I think that there is a business disconnect-

Host: Luke: Mm-hmm

Joel: between sort of cost and efficiency- Yeah … and the fact that what they sell is tokens.

Host: Luke: Yeah,

Joel: yeah. Right? And so there’s a concern there that, if you are the customer, that goals around efficiency, goals around using the correct model- Right … are not necessarily in the LLM provider’s best interest. Right,

Host: Luke: right.

Joel: Right?

Host: Luke: No,

Joel: yeah. But they’re certainly in yours. And so, I think, um, being able to be model agnostic is one of the key things that an enterprise needs to think about. And so you asked me, what are people over-rotating on? There are like, “Well, which model’s the best?” And- Yeah … that’s anybody’s guess, honestly.

Host: Luke: Yeah. it’s… Yeah, I’m glad you mentioned that, ‘cause it’s, it’s just one of those things where it is, it’s more trendy than fashion, right? Like, it’s in and out, and, and you’re like, three months later, no one even uses the same thing or even cares about it anymore, you know? It’s just, it’s wild how fast that there’s so much hype around that, too, in, in the market and, and all that.

yeah, was there anything we didn’t cover that you want people to know about? [00:23:00]

Joel: Yeah, well, I mean, we could just w- f- close out this model discussion here, right? Sure, sure. Um, the next frontier is model efficiency- Yeah … without any question. or let’s call it token yield-

Host: Luke: Yeah …

Joel: might be a, or token efficiency might be the right way to look at it.

This is no longer, a space where we are using experimental budget. Mm-hmm. We’re talking about operational budget- Right … in an enterprise.

Host: Luke: Right.

Joel: And what that means is, we need to be focused as enterprises on value received for tokens burnt.

Host: Luke: Right.

Joel: Right?

Host: Luke: Right, right.

Joel: And, there’s a couple of ways you can be very smart about that, and the first one is, going back to context.

Host: Luke: Right.

Joel: Feed the right information into that LLM-

Host: Luke: Yeah …

Joel: and feed it in intelligently so it doesn’t burn unnecessary tokens on search and retrieval.

Host: Luke: Yeah.

Joel: Where it is actually reasoning against the most relevant data it can use to form a conclusion and to- Yeah … take action. That’s the first step. The second step is which model should you be using?

You should be able to intelligently route between models-

Host: Luke: Yeah …

Joel: so that you get the most bang for your buck- Yeah … and you’re [00:24:00] not burning tokens on the latest and greatest frontier model for no reason.

Host: Luke: Right.

Joel: Yeah. And so that, like if there’s one takeaway, it’s, you know, make sure your context is in a good spot-

Host: Luke: Yeah

Joel: so that you can feed the right model the right information.

Host: Luke: Right.

Joel: Make sure you can choose between models, and that that all is secure in a way that, that, that makes your corporation happy- Yeah … your enterprise happy. If you have those three things, the right context, the right intelligence layer, and the right protection- Yeah

around what you’re doing, then your deployment becomes a lot easier, and the odds of you getting a positive ROI and high adoption, all that just go way up.

Host: Luke: Awesome.

Joel: So it’s a definite improvement.

Host: Luke: it makes a lot of sense. Joel, thank you so much for, for making the time today.

Where can people find you online? Are you posting on LinkedIn or X or anywhere like that?

Joel: Yeah, you can find me on LinkedIn is probably the best place, and you can always head over to glean.com and see all our stuff.

Host: Luke: Awesome, awesome. Well, Joel, thank you again for making the time. It’s really great, talking about this with you, and, uh, love to have you back someday too to kinda check back in on how things

Joel: are- Yeah, for sure.

Well, perhaps when we’re back in California we can have-

Host: Luke: Yeah, I know, right? We can have more [00:25:00] chat. We’ll be back in our time zones. All right, thanks Joel. Appreciate it.

Joel: Yeah. Thank you so much.

Luke: 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 at 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.