WEBVTT
00:00:05.040 --> 00:00:07.839
Hello, everyone, and welcome to the SourceForge podcast.
00:00:07.919 --> 00:00:09.519
I'm your host, Bo Hamilton.
00:00:09.839 --> 00:00:11.919
Let's talk about enterprise AI.
00:00:12.000 --> 00:00:16.640
So, AI has created a little bit of a strange situation inside a lot of companies.
00:00:16.719 --> 00:00:27.839
Uh, the models themselves are getting more capable, but the answers people are getting back can still be rather inconsistent, perhaps a little bit outdated, or uh in some cases just flat out wrong.
00:00:28.000 --> 00:00:37.759
So there's sort of this weird uh disconnect where the technology feels pretty advanced, yet the output still isn't always something you know people can fully trust and get behind.
00:00:37.920 --> 00:00:41.759
And that's uh, I would say really the tension a lot of companies are dealing with right now.
00:00:41.920 --> 00:01:02.240
It's not necessarily that the models are failing, it's that the businesses are trying to layer AI on top of information that might be just scattered across teams, buried in old documents, trapped in various uh systems that might be hard to uh retrieve and access, um, or just you know, otherwise not structured in a way that AI can reliably use.
00:01:02.399 --> 00:01:08.480
So the big question is whether the real problem is the model itself or the mess content sitting underneath it.
00:01:08.640 --> 00:01:11.040
And that's really where our guest comes into play.
00:01:11.120 --> 00:01:18.480
Uh today's guest is Rahul Yadov, CEO of PaliGo, a company built around structured content and documentation workflows.
00:01:18.560 --> 00:01:20.799
Uh, we've got a lot to get into for today's conversation.
00:01:20.959 --> 00:01:30.959
We're going to dig into why enterprise AI keeps hallucinating, what a real source of truth actually looks like, and what companies need to fix before they expect uh AI to be trustworthy.
00:01:31.120 --> 00:01:32.719
So, Rahul, welcome to the podcast.
00:01:32.879 --> 00:01:34.000
Glad you could join us.
00:01:34.400 --> 00:01:35.040
Thank you, both.
00:01:35.280 --> 00:01:36.079
Really nice to meet you.
00:01:36.159 --> 00:01:37.920
And thanks for having me on the podcast.
00:01:38.239 --> 00:01:39.120
Of course, of course.
00:01:39.200 --> 00:01:41.120
So I just want to get right into it, Rahul.
00:01:41.280 --> 00:01:48.400
Um, you've uh, you know, you've argued that this really isn't an AI problem as much as it is a content problem.
00:01:48.560 --> 00:01:51.040
Um when you say that, what exactly do you mean?
00:01:51.120 --> 00:01:55.439
What uh where is like the real breakdown um happening inside most organizations?
00:01:56.319 --> 00:01:57.280
No, true, actually.
00:01:57.359 --> 00:02:03.760
So I think, you know, as we know that most of the enterprises now they're trying or they're racing to deploy AI, right?
00:02:04.000 --> 00:02:10.879
And when we also know that where the deployment fails, it's actually failing at what I call last mile, right?
00:02:11.039 --> 00:02:17.919
When they when they really try to put into production, uh, you know, these agents, super fancy agents, they fail, right?
00:02:18.159 --> 00:02:24.719
And I think most of the time people blame the model um, you know, when their AI gives the wrong answers.
00:02:24.960 --> 00:02:27.599
But but think about it, what actually happens, right?
00:02:27.759 --> 00:02:30.159
So essentially what as you started with, right?
00:02:30.319 --> 00:02:36.560
You take um you take a SharePoint folder, for example, which has hundreds of documents in different formats.
00:02:36.719 --> 00:02:38.639
It could be PDF, it could be Word.
00:02:39.039 --> 00:02:46.800
And we also know that how we even use our own documents on our on our own desktop, many are outdated, many are duplicated, right?
00:02:47.039 --> 00:02:56.240
And I think especially when you have these documents uh and PDFs, what I call flat file, they have no semantic structure.
00:02:56.560 --> 00:03:09.680
So what normally happens is that when you throw this to a more sophisticated or more sophisticated LLM, you know, what happens is that you chunk into vectors, you chunk it, you chunk the document using any kind of rack systems.
00:03:09.840 --> 00:03:17.360
So what happens is that you feed into one of the most sophisticated, whether it's a GPT-4 or a Claude, you know, with Opus 4.7.
00:03:17.599 --> 00:03:20.719
I mean, you know, these are brilliant reasoning engines.
00:03:21.280 --> 00:03:29.520
And when, and then you're surprised when you when it can't distinguish between, let's say, your product X400 versus a product X500.
00:03:29.759 --> 00:03:34.960
Actually, it's not, you know, the model isn't just hallucinating because it's stupid, right?
00:03:35.120 --> 00:03:39.520
It is hallucinating because it has no access to the truth, the structured truth, right?
00:03:39.680 --> 00:03:53.520
Because you give it a mess and then you ask your model to be more uh concrete and more confident, which by the way, it will be when you when you see it, but most of the time it actually hallucinates and give wrong answers.
00:03:53.840 --> 00:03:56.719
Yeah, when you when you put it like that, it makes it makes total sense.
00:03:56.879 --> 00:04:13.199
I mean, uh if it if some of these documentation structures um look anything like uh my my documents folder and my computer, what and and the various naming structures and um just where everything is is located, I I can see why there would be there would be issues, right?
00:04:13.360 --> 00:04:27.600
And so um when you deal with these these massive uh databases and internal systems, um I can see the complexity there trying to retrieve valuable information and then just having to lean heavily on AI trying to extrapolate and fill in the missing pieces.
00:04:27.759 --> 00:04:31.279
I can just see uh, you know, it's like it's a recipe for disaster, really.
00:04:31.439 --> 00:04:40.160
Um and you know, it kind of goes to the baseline of like AI uh or an LLM is really only as good as the data it has to work with, you know.
00:04:40.639 --> 00:04:41.040
Exactly.
00:04:41.199 --> 00:04:44.399
And I also feel actually people feel that the fix is a better model.
00:04:44.639 --> 00:04:52.240
But actually, on the contrary, the fix is, yes, a better model, but fix is also structured governed content, which is grounded in truth.
00:04:52.720 --> 00:04:53.279
Yeah, yeah.
00:04:53.439 --> 00:04:55.199
It's not always the better model.
00:04:55.279 --> 00:05:01.920
Because I think um we've kind of also plateaued in terms of you know some of these sophistications uh with the models themselves.
00:05:02.079 --> 00:05:07.199
And so a lot of it's just uh yeah, integrating the right data and and into the model.
00:05:07.439 --> 00:05:20.639
Um and also I think it's it's interesting to think about too, you know, with the enterprise market, you you know, you're working with perhaps AI systems and models that are a little bit more closed off.
00:05:20.800 --> 00:05:30.399
They're not able to go tap into some of these bigger parameters and resources that maybe some of the consumer-facing chatbots can do uh for you know for privacy and security reasons.
00:05:30.560 --> 00:05:32.079
So that kind of limits them further.
00:05:32.319 --> 00:05:46.000
But um, this this issue is uh the basis of one of the ideas I've heard you talk about in enterprise AI, and that's uh called the frontier gap, uh where the model itself is you know intelligent, but it has no access to reliable enterprise truth.
00:05:46.160 --> 00:05:50.399
Um when you talk about that gap, how does it actually show up in practice?
00:05:50.800 --> 00:05:51.360
Yeah, sure.
00:05:51.439 --> 00:05:55.439
So I think the frontier gap, I mean, you know, just for your listeners, right?
00:05:55.519 --> 00:06:03.839
So the frontier gap, what I call is a distance between what a model can do and what enterprise can safely deploy.
00:06:04.160 --> 00:06:09.439
That is actually, so yes, the model, these models, as you said, are becoming very sophisticated.
00:06:09.600 --> 00:06:12.879
You know, they are really getting better at reasoning.
00:06:13.120 --> 00:06:20.240
But how much, especially in the enterprise setting, how much you can actually safely uh deploy is a key here, right?
00:06:20.399 --> 00:06:21.920
So models are brilliant.
00:06:22.160 --> 00:06:31.519
I mean, you know that cloud can reason through complex problems, so is uh so is GPT-4, for example, can synthesize information better than us as humans, right?
00:06:31.839 --> 00:06:39.680
But when you point this powerful intelligence at an enterprise content, there is this gap that comes in, right?
00:06:39.839 --> 00:06:43.120
So model doesn't know which content is current, right?
00:06:43.439 --> 00:06:49.839
Model doesn't know which content is approved, which applies to which which content applies to which product.
00:06:50.560 --> 00:06:50.879
Right.
00:06:50.959 --> 00:06:56.560
And then also like you know, which content is written for internal use versus customer-facing use.
00:06:56.800 --> 00:07:07.759
It doesn't know which paragraph was deprecated six months ago, or whether this paragraph requires a legal review before you put it to external use.
00:07:08.079 --> 00:07:08.480
Right?
00:07:08.800 --> 00:07:13.839
So what you get out of all of this is you really get confident hallucinations.
00:07:13.920 --> 00:07:14.879
That's what I say, right?
00:07:15.040 --> 00:07:20.639
A customer asks about safety procedure, AI will cite a manual from 2019.
00:07:20.800 --> 00:07:30.000
Maybe that manual was deprecated before, but it sounds certain because the model is certain and it just didn't have the access to the truth that you have, right?
00:07:30.160 --> 00:07:31.759
And this is where people actually miss.
00:07:31.839 --> 00:07:37.519
This is where people miss that that you know the models are really doing a lot on model safety, right?
00:07:37.680 --> 00:07:41.519
They're making sure the model doesn't generate harmful content, right?
00:07:41.839 --> 00:07:44.879
Or or it doesn't help with illegal activities.
00:07:45.120 --> 00:07:52.959
But what's missing, especially from an enterprise perspective, is enterprise governance that is left to the enterprises, right?
00:07:53.199 --> 00:08:01.040
So so this is where this frontier gap exists because the model providers build the intelligence layer, but nobody built that truth layer.
00:08:01.199 --> 00:08:03.759
And that's the infrastructure that is missing in my mind.
00:08:04.079 --> 00:08:05.040
Yeah, that makes sense.
00:08:05.120 --> 00:08:14.720
Um I'm curious, like if can I go back going back to what you were saying earlier, is if if the real issue is not the model itself, but the content underneath it, right?
00:08:14.879 --> 00:08:20.959
That raises the next sort of question that comes to my mind is what does a strong foundation actually look like?
00:08:21.040 --> 00:08:29.439
You know, I know this is this is really how how your company has kind of made its bread and butter, its its name for itself, building around this idea of structured content.
00:08:29.600 --> 00:08:38.720
Um, but for people who may not live and and breathe documentation systems, what would you say makes for an effective documentation system in the first place?
00:08:38.879 --> 00:08:43.200
And then why does that structure just matter so much when AI enters the space?
00:08:43.600 --> 00:08:44.399
No, absolutely.
00:08:44.480 --> 00:08:49.360
So so I think you know, it all came from so Poligo was formed um, you know, 10 years ago, right?
00:08:49.519 --> 00:08:58.080
Ten years ago, we started with uh with creating structured authoring uh tool, which is what they call CCMS, content component management system, right?
00:08:58.240 --> 00:09:00.879
Which was based on DogBook XML as a standard.
00:09:01.120 --> 00:09:20.320
Now, the reason we created this is because we know that when you when you achieve a certain level of complexity in terms of your document needs into an organization, you need a tool like Poligo to not only author and create content, but also create a lot of reuse and publication and translation and so on and so forth, the entire life cycle of content.
00:09:20.879 --> 00:09:24.639
Now, and this is why we started with structure in the first place.
00:09:25.039 --> 00:09:39.120
Now, what happens is when you talk about what actually is structured content, structure content for us is content that has semantic understanding, that means all the metadata, version controlling, right?
00:09:39.279 --> 00:09:41.200
And all of that combined together.
00:09:41.360 --> 00:09:54.080
Um, you know, if you can offer this to your not only to humans, but also to AI, what you can do is you can actually uh get much better and accurate answers on your other side, right?
00:09:54.240 --> 00:09:58.720
So what Poligo and in Poligo, what we are trying to do now, we come from a structured content.
00:09:58.879 --> 00:10:17.519
And we are moving to what we call structured truth for enterprise AI, meaning that if you really write content and create content in Poligo, we will provide a solution, a mission critical infrastructure from content ingestion, authoring, publication, delivery to consumption and grounding on the other side.
00:10:17.759 --> 00:10:32.000
And this is what sound structured content infrastructure should look like, where you offer semantic, um, uh semantically rich content, which is perfect fuel for agent techs on the other side.
00:10:32.480 --> 00:10:42.320
Yeah, it's interesting to hear that you've you've you know been working on um structured data for the past decade, you know, before this this AI boom that we've we all found ourselves in.
00:10:42.399 --> 00:10:53.919
I imagine back in back then it was really um just around cloud-based systems and getting that all up and running and and making that easier to to access and and um you know utilize.
00:10:54.159 --> 00:11:09.840
Um, but you're kind of ahead of your time there in terms of what you guys are tackling because uh having a really thoughtful, well-structured um system is is just gonna reward yourself uh leaps and bounds with um some of these AI tools you're able to throw in on top.
00:11:10.080 --> 00:11:23.440
Um but uh yeah, I wish I I wish I could like interview you and um or the team like 10 years ago to see kind of your headspace and where where things were going back then and compare it where you're at now, because um it's so interesting how things have changed.
00:11:23.600 --> 00:11:31.360
But you know, a lot of a lot of the market right now, I would say it feels like there's sort of this race to add AI features as quickly as possible.
00:11:31.519 --> 00:11:35.279
And and I think from the outside that that looks like progress, right?
00:11:35.360 --> 00:11:40.799
Even though it's it's definitely a different story from some of the the insiders actually using the tools.
00:11:41.039 --> 00:11:43.840
Um, you've you've obviously taken a pretty different view.
00:11:43.919 --> 00:11:52.879
What what's what's the risk when companies try to layer AI onto a content environment that was already sort of fragmented or poorly organized to begin with?
00:11:53.279 --> 00:11:55.600
Yeah, actually internally I joke this around.
00:11:55.679 --> 00:11:56.399
Uh Bo.
00:11:56.720 --> 00:12:00.960
I actually call it uh that model of bolting features on top of the existing product.
00:12:01.120 --> 00:12:03.200
I call it a lipstick pick model, right?
00:12:03.360 --> 00:12:11.120
So you're putting a lipstick on a pick because everything else is ugly, and you just try to make it beautiful by just putting a lipstick, which is an AI lipstick, right?
00:12:11.279 --> 00:12:23.200
Um so I think what I really feel uh is that what we have taken in the approach that we have in Poligo is we will we have said that we are not gonna bolt just AI feature for the sake of AI feature, right?
00:12:23.440 --> 00:12:26.799
And and the reason is because we really believe that model is broken.
00:12:26.960 --> 00:12:39.200
Uh and if a foundation is not there, you just can't you can really throw in a lot of features, but but those features will not be very um you know, very helpful for our users, right?
00:12:39.360 --> 00:12:53.360
So, first of all, we said that even if we bolt some features, it has to be context-aware, meaning that we will not just have an assistant on the side, which no one uses because you have an AI assistant, because apparently everyone loves co-pilot over the last couple of years, right?
00:12:53.519 --> 00:13:00.240
But if the co-pilot is not in the context where the where the user is in their workflow, I think it's not useful enough.
00:13:00.320 --> 00:13:09.600
So we our idea has always been how can we augment a particular workflow or a step in a workflow for a user and make a user more efficient.
00:13:09.759 --> 00:13:11.600
So that's what one approach we are looking into.
00:13:11.759 --> 00:13:19.279
But we see that if we really have this bolt-on features on top of this, I actually feel that you end up into these three things, right?
00:13:19.519 --> 00:13:23.120
Number one, I would actually say that liability, right?
00:13:23.360 --> 00:13:32.159
Simply because if you don't have the right foundation, the content foundation, a clean structure, truth infrastructure, you will have hallucination.
00:13:32.320 --> 00:13:46.480
And especially in the regulated industries, Bo, we will end up actually having a lot of liability issues simply because just imagine if your agent hallucinates a wrong procedure, a safety procedure to a customer.
00:13:46.720 --> 00:13:52.080
Imagine the kind of liability, especially if you are processing an insurance claim, for example, right?
00:13:52.240 --> 00:13:56.480
It can probably refund a crazy amount by just simply because it's hallucinating.
00:13:56.639 --> 00:13:58.159
So there's a liability issue, right?
00:13:58.399 --> 00:14:09.120
Second thing I actually believe, which not that many people talk about, the benefit of structure content and structure content to feed AI, especially in enterprise, is your runway cost.
00:14:09.679 --> 00:14:21.440
If you don't have structured content underneath and structure uh structured infrastructure for content proper, proper infrastructure in place, you will have a lot of token cost consumption, right?
00:14:21.679 --> 00:14:28.559
Consumption that you use for tokens, the cost will be very high simply because you need to do multiple retries to get right answers.
00:14:28.720 --> 00:14:34.399
Your RA needs to create a lot of chunks in terms of and create vector database and then do all of that.
00:14:34.559 --> 00:14:38.639
It requires a lot of consumption from an AI, intelligence, and token cost perspective.
00:14:38.799 --> 00:14:47.440
But on the contrary, if you have structured content, your cost for consuming that will be much lower and you can get better answers for a lower price point.
00:14:47.600 --> 00:15:04.559
And third, if you don't have a good infrastructure, content infrastructure in place, I think you will promote that silo that we also used to talk about in the world of data and analytics in the past, where every team was creating their own tool and their own platform, and then the and the data layer was not connected together.
00:15:04.720 --> 00:15:13.840
You're gonna amplify this with Agentix simply because every single team will create their own agents and with on their own top on their own top of their own data.
00:15:14.000 --> 00:15:16.240
Like for example, marketing will create their own.
00:15:16.480 --> 00:15:23.039
So is product, so is uh sales, and and all of those will have their own data silo.
00:15:23.120 --> 00:15:28.960
So just imagine the same question asked through different three different agents will give three different answers.
00:15:29.200 --> 00:15:30.960
Simply because they are on this there, right?
00:15:31.279 --> 00:15:44.399
So I actually believe that the if you take a shortcut, you be you know, you you come into the trap where you will end up having these three risks I talk about the liability risk, your runway cost risk, and then your silo.
00:15:44.960 --> 00:15:46.000
Those are really good points.
00:15:46.080 --> 00:15:47.360
Yeah, all really good points.
00:15:47.519 --> 00:16:06.320
I I think the the tokens, um, uh the second point in particular really, really um speaks to the the current sort of um you know craze right now and with AI and and just how companies, people are burning through tokens, trying to utilize AI um because of all the the promise and what it's able to do with with fib coding and and whatnot.
00:16:06.399 --> 00:16:08.399
And the cost gets out of control.
00:16:08.480 --> 00:16:20.000
So if you're able to just have have it become more efficient, which is the whole goal of AI in the first place, is just kind of the pushing forward this efficiency era, um, then you're gonna you're gonna save some money.
00:16:20.080 --> 00:16:26.320
But also the legal standpoints and um all the other points you mentioned too are really, really do make a lot of sense.
00:16:26.480 --> 00:16:32.480
And um I again, you're really just by taking a shortcut, you're really just setting yourself up for a headache in the long run.
00:16:32.639 --> 00:16:38.159
Um so yeah, it's it's um these are really good points to to keep in keep in mind.
00:16:38.240 --> 00:16:54.159
And um I I want to continue to to drill down and um, you know, I guess just also open this up to some more tangible examples that you might think of than uh, you know, where you've implemented a strategy um that's actually really had a lot of reward for for clients.
00:16:54.320 --> 00:17:05.359
I know um one that stood out to me is how you've talked about taking a translation process that used to take weeks uh and shrinking that down to something that can happen in literally hours inside the platform.
00:17:05.519 --> 00:17:07.599
Can you walk us through that that example?
00:17:07.759 --> 00:17:14.799
What had to be true behind the scenes from an architectural standpoint for that kind of lead to even be possible?
00:17:15.279 --> 00:17:17.359
No, I think that's uh that's a great example.
00:17:17.519 --> 00:17:21.680
And I think, by the way, we released that AI translation just a month ago, right?
00:17:21.759 --> 00:17:23.680
So so take an example, right?
00:17:23.759 --> 00:17:26.640
I think um traditional translation, we all know that, right?
00:17:26.720 --> 00:17:28.079
It's painful, right?
00:17:28.240 --> 00:17:33.519
The reason it's painful is because you create content one in one place, then you need to export that content.
00:17:33.839 --> 00:17:42.079
You need to send it to this uh this uh exported content that you have uh to a translation management system, right?
00:17:42.319 --> 00:17:48.079
Or you send it uh or you send it manually basically to a translator translation agency.
00:17:48.480 --> 00:17:50.160
Then you wait for two, three weeks.
00:17:50.720 --> 00:17:55.279
Then once the translation comes in, you re-import that content inside your system.
00:17:55.680 --> 00:18:01.440
And then basically you end up into this, what I call uh reconciliation nightmare, right?
00:18:01.680 --> 00:18:09.200
Because maybe in this two, three weeks, your original version, which was written in English, maybe that has evolved, right?
00:18:09.440 --> 00:18:16.400
Uh because now the version that you sent for translation compared to what you have actually now have in your system, it is different, right?
00:18:16.720 --> 00:18:19.920
So how do you now with that, how do we keep this?
00:18:20.000 --> 00:18:26.960
Maybe you have 47 languages that that needs to be in sync when when the English version keeps evolving, right?
00:18:27.119 --> 00:18:29.839
And this is actually a very tangible example, right?
00:18:30.000 --> 00:18:33.200
Where we have seen a lot of our customers struggling with that.
00:18:33.440 --> 00:18:40.000
And now when you replace this with AI translation, which doesn't happen in weeks, it happens in minutes, right?
00:18:40.160 --> 00:18:43.920
Simply because you can actually use uh AI to do that, right?
00:18:44.079 --> 00:18:56.400
Within and inside the platform, the authoring tool that you're using, like Poligo CCMS, you can use AI translation where you just click a few buttons and then you translate, and you really get a very accurate translated content back.
00:18:56.640 --> 00:19:15.519
Because what we do is that we we are very since it's a component content management, we can go all the way to granular level of a component while also getting all the advantage of uh component content management in terms of governance, in terms of versioning, in terms of uh semantic and metadata and all of that, right?
00:19:15.759 --> 00:19:24.079
So so this is what actually you know, you can see a very concrete example where where where users are already getting benefit out of that, right?
00:19:24.319 --> 00:19:37.920
And and architecturally, I mean, I will be I think that basically what you need to do is that you will only be able to do it if you have structured content, which is which is written in a component which is which are reusable, right?
00:19:38.000 --> 00:19:39.599
So you can reuse it, right?
00:19:40.000 --> 00:19:42.799
Each component is linked to a version.
00:19:42.960 --> 00:19:52.559
So you actually know that that because of versioning, even though your original content continues to evolve, you can still reconcile, can still uh still do that, right?
00:19:52.640 --> 00:19:53.359
So that's there.
00:19:53.519 --> 00:20:02.319
And then also make sure that the translation governance is to be in few, you know, is is uh is and the versioning governance is enforced as well, right?
00:20:02.480 --> 00:20:05.200
So for example, Gross V style guides, right?
00:20:05.279 --> 00:20:13.200
For example, brand terminology, taxonomy, they are all uh not deployed manually by human, but actually it's already taken care of by translation.
00:20:13.359 --> 00:20:15.359
This is enforced automatically by system.
00:20:15.680 --> 00:20:19.440
And I can tell you, this is the byproduct of all of this is speed.
00:20:19.839 --> 00:20:29.359
And especially when teams are deploying code or or releasing product based on continuous delivery and continuous deployment, you don't have weeks to wait for.
00:20:29.440 --> 00:20:37.599
And now with AI translation, you you know, translation is no longer, thanks to AI, is no longer a bottleneck in terms of your throughput type.
00:20:37.839 --> 00:20:42.799
Yeah, I mean, it literally has a direct impact on the operational impact, right?
00:20:42.880 --> 00:20:46.000
Um, which it has a direct impact on speed and so many other things.
00:20:46.079 --> 00:21:03.839
Um not only is it hallucinating less, but it's also it's it's retrieving information faster, it's delivering, you know, more accurate, truthful answers uh faster, which is kind of the whole, again, the whole point of AI in the first place is doing things uh more efficiently and quicker and helping you do things more efficiently and quicker than before.
00:21:04.000 --> 00:21:10.559
Um and I think the this this conversation also changes how we think about people behind the scenes, right?
00:21:10.640 --> 00:21:20.480
Um and I uh one sort of uh a job and role that comes to mind is is are the technical writers that have uh been seen as more of a maybe supporting role um over the years.
00:21:20.640 --> 00:21:25.759
You know, they're important, but they're not always brought into the conversation as like a strategic part of the business.
00:21:25.920 --> 00:21:29.519
Um, but things are changing in this AI first uh world.
00:21:29.680 --> 00:21:40.000
Um in this world we live in with AI kind of uh just being incorporated in every way, shape, and form, um what does that role for technical writers look like now?
00:21:40.160 --> 00:21:44.240
I'm curious, like, where do technical writers sit in the bigger picture?
00:21:44.559 --> 00:21:46.640
Yeah, I think I think this is very interesting.
00:21:46.720 --> 00:21:49.839
And and I think this is also one of the missions that we have at Poligo.
00:21:50.160 --> 00:21:58.480
We actually believe that the right the role of tech writers is gonna become even more important in this AI era, it's very, very clear.
00:21:58.720 --> 00:21:59.839
I actually believe that the tech writers are not.
00:22:00.160 --> 00:22:13.440
Writers are going to become the superhero in this in this AR race, especially if you're coming from a very sound background of technical documentation, where structure is really what you really bring to the table.
00:22:13.599 --> 00:22:15.279
I think your role will become very, very important.
00:22:15.440 --> 00:22:28.400
I actually believe that the role of technical writer will evolve into, I mean, you know, internally we call it content curator or content strategist or maybe content architect or content directors, right?
00:22:28.640 --> 00:22:30.079
This is the role you're going to take.
00:22:30.319 --> 00:22:40.480
Because now the content that you used to write for human consumption, right, is it's of course, there will still be need actually in the future in certain industries, in certain verticals.
00:22:40.640 --> 00:22:51.680
But I actually believe that the content that you will orchestrate, the content that you will direct, will also be relevant for, will be more relevant, I would say, for agentics and AI.
00:22:51.839 --> 00:22:53.279
So I think that all will evolve.
00:22:53.440 --> 00:23:02.799
So meaning that yes, I think if you come from purely writing content from scratch, I can tell you AI can take you from zero to 80% in no time, right?
00:23:02.960 --> 00:23:14.079
So content generation will, of course, if you only do content generation and if you are really not architecting content and not structuring and creating structure and creating semantics, then of course you need to think about it.
00:23:14.160 --> 00:23:20.000
But I actually believe that content generation from zero to 80% will become very easy, which is a solved problem.
00:23:20.160 --> 00:23:30.160
But then still governing that content, making sure that it is rich in metadata, in context, in versioning and all of that, really curating that, that will be very important.
00:23:30.319 --> 00:23:36.079
And I actually believe, which is even more important, which I don't think that there are that many many people talk about it.
00:23:36.319 --> 00:23:43.119
Remember, most of those tech writers are also having a very strong domain expertise, a certain domain they represent.
00:23:43.359 --> 00:23:49.200
And as we know very clearly, what is more important in this world of AI is domain expertise.
00:23:49.359 --> 00:24:03.359
So if you are really coming from, let's say, a pharmaceutical industry or a manufacturing industry, where you have been tech writing those content pieces for safety regulation, uh right, use, right, manuals, all of that will become even more important.
00:24:03.519 --> 00:24:23.359
And how it is written, how it is stored, you know, we are really powered by metadata and semantic, all of that will become even more relevant for agent tech simply because now you need to make sure that you know when you have rich content at the source, you can actually have better answers or hallucination-free or fault-free answers on the other side.
00:24:23.519 --> 00:24:34.559
And then I think to take a further down the line, when the content is being consumed by AI, these tech writers could also become human in the loop to enforce reinforcement learning.
00:24:34.720 --> 00:24:43.599
Meaning that, meaning that if they model uh hallucinate, this, you know, when someone needs to correct it, we need humans to to really correct it.
00:24:43.759 --> 00:24:44.799
And who are those humans?
00:24:44.960 --> 00:24:50.240
There's gonna be, again, our content curators or the tech writers, because they know what is good and what is bad.
00:24:50.559 --> 00:24:52.319
I I love I I love this take.
00:24:52.480 --> 00:24:52.640
Yeah.
00:24:53.839 --> 00:24:54.960
I love that take that you have.
00:24:55.039 --> 00:25:19.680
And um, I mean, in full transparency, this is what um you know, what we rely on and and lean heavily into with SourceForge and um using our technical writers to you know um work directly with with companies on that have uh listings on our site, uh, having these conversations, you and you and I having this conversation, um, and using our our domain authority to to kind of bring it to the masses, right?
00:25:19.759 --> 00:25:22.400
Um and get that valuable, you know, truthful information out there.
00:25:22.480 --> 00:25:26.720
So you don't have to the uh AI can use our content as a source.
00:25:26.880 --> 00:25:32.079
Um and again, it's all about coming back to this, this creating this knowledge layer that AI depends on.
00:25:32.160 --> 00:25:36.559
Um and I I think that's that's really the the value um that tactical writers offer.
00:25:36.640 --> 00:25:41.680
And I think that uh, you know, writers of all kinds have been put on a pedestal for me, I would say.
00:25:41.759 --> 00:25:52.000
Um I'm a little bit biased, of course, but um this way of thinking about the role really does uh, you know, it feels a lot more strategic than the way it's traditionally been viewed.
00:25:52.160 --> 00:25:54.880
And I I think that's um that's that's great.
00:25:55.039 --> 00:26:04.559
I think that's that I think a lot of uh writers who are uncertain about their future um are feeling a lot better based off your answer here.
00:26:05.039 --> 00:26:06.160
No, I I hope so.
00:26:06.240 --> 00:26:14.160
And I generally believe actually that both this is this is so crucial now because I think you know AI is really good in a lot of generic things.
00:26:14.400 --> 00:26:21.039
The moment you come become very specific into an enterprise, into a certain workflow with certain business logic, business rule.
00:26:21.200 --> 00:26:27.200
I mean, you know, tech writers are the champions for that business, and I think they need to come in the center of it.
00:26:27.279 --> 00:26:33.920
So I also believe that the companies that treat documentation as a cost center, I actually believe they will have a tough time.
00:26:34.000 --> 00:26:40.400
They will lose to the companies that treat the documentation and the documentation team as a strategic asset, right?
00:26:40.559 --> 00:26:42.720
And that transition is happening, I would say, now.
00:26:42.880 --> 00:26:49.920
And I actually am having so many conversations with our customers, and I can really see that that conversations are changing.
00:26:50.000 --> 00:27:07.440
But in some places, also conversations are still about cost center, where where you know you can see that hey, that team of documentation uh managers or documentation or tech writers have been shrinked without decision makers being aware actually that it's just such a strategic function in the world of AI.
00:27:07.839 --> 00:27:19.680
Yeah, I mean, they're becoming more um integrated and a part of the their company with business leaders trying to piece together all these different um sources of information and um structure their the data accordingly.
00:27:19.839 --> 00:27:24.319
So I think it's um yeah, I think it's a it's it's really uh I guess a win-win.
00:27:24.400 --> 00:27:42.160
And also um it just it it's again, it's great to hear because I think there's a lot of uh writers and all um you know capacities that are feeling a little bit discouraged, I guess, with um some of these these AI content generation, you know, capabilities out there.
00:27:42.319 --> 00:27:49.039
Um but it's like you're saying, it's not just the content generation, it's the structuring of the data that's increasingly more important.
00:27:49.519 --> 00:28:05.680
So um, but yeah, I think getting back to the the content layer part of the conversation and how important that is, um I know there's obviously a lot of pressure on CTOs right now to move to move quickly on AI, but um I know speed can also hide a lot of the sort of bad assumptions.
00:28:05.920 --> 00:28:14.640
If you could get every CTO to really understand one big thing about their content before they they launch an AI project, um what would you want that lesson to be?
00:28:15.039 --> 00:28:17.599
I think I think the less the lesson is very simple.
00:28:17.759 --> 00:28:20.799
I think you know, garbage in, garbage out, right?
00:28:20.960 --> 00:28:22.079
Is very, very clear.
00:28:22.319 --> 00:28:29.119
I know, you know, in my one of my uh companies that I worked um a couple of years ago, it was a media streaming company.
00:28:29.279 --> 00:28:32.319
I used to say that content we used to say content is king in media.
00:28:32.480 --> 00:28:33.759
We used to say content is king.
00:28:33.920 --> 00:28:36.480
I actually believe the structured content is king comm.
00:28:36.720 --> 00:28:42.079
Right and that's what that's what the CTOs, CIOs, CAIOs need to understand.
00:28:42.240 --> 00:28:53.279
I think your your AI project that you're you are you're deploying for millions and billions now will succeed or fail, will depend on the foundation of your structured content.
00:28:53.440 --> 00:28:58.319
If you have really good quality structured content, I think your project will succeed.
00:28:58.480 --> 00:29:02.079
If not, you're gonna spend the next couple of years building that foundation.
00:29:02.240 --> 00:29:08.160
Just like what happened when the data uh revolution came happens, right, for analytics, right?
00:29:08.319 --> 00:29:16.079
Every single company, every single CTO, CIOs, what they have done is they've been chasing to really bring data out of silos, right?
00:29:16.240 --> 00:29:21.279
Which was enterprise data, HRM data, CRM data, you name all kinds of business data, right?
00:29:21.359 --> 00:29:27.839
You brought all of that, and then you wanted to really host it somewhere in a data lake where you can actually do a lot of analytics on top of it.
00:29:28.079 --> 00:29:29.920
Content has been overlooked.
00:29:30.160 --> 00:29:37.680
And now we know that these LLMs, they are fed on human language, normal language, uh data, right?
00:29:37.839 --> 00:29:39.759
Written docs documentation, right?
00:29:39.920 --> 00:29:40.720
Content, right?
00:29:41.039 --> 00:29:48.400
Which is residing in um silo and retire and residing in Word doc, Google Doc, PDFs, and all of that, right?
00:29:48.559 --> 00:29:51.440
Duplicated version, multiple copies, all of that.
00:29:51.599 --> 00:29:54.319
All of that needs to be needs to be structured.
00:29:54.640 --> 00:30:06.559
And and if you are a CTO, CIO listening, this, if you feel you can fix it retrospectively, what I call fixing it downstream, the cost is very high and the technology is not there.
00:30:07.039 --> 00:30:10.799
You need to start thinking about uh you know, creating purity at source.
00:30:10.960 --> 00:30:23.759
And purity of source is that if you create structured, semantically rich, componentized structure using a tool, I think that will be that will be what and then you need to start thinking content as an infrastructure layer.
00:30:24.480 --> 00:30:25.119
Great advice.
00:30:25.279 --> 00:30:26.400
Structure is king.
00:30:26.480 --> 00:30:29.680
That's that's uh that's a great um that's a great thing to remember.
00:30:29.839 --> 00:30:37.200
And again, it feels like AI, you know, success, it starts with chasing the model, but um in reality, it's it starts much earlier than that.
00:30:37.279 --> 00:30:45.440
It starts with the the understanding your content, your systems, um, and and your internal source of truth, make sure the structure is all there.
00:30:45.599 --> 00:30:49.359
Um so so very good, very good things to underline and underscore.
00:30:49.519 --> 00:30:56.079
Um now looking ahead at where things are going, I'm curious to see where things are uh continue to evolve from here.
00:30:56.319 --> 00:31:06.319
What emerging trends do you think will have the biggest impact on enterprise documentation in the next few years, let alone five years from now, if we can even um plan that far in advance?
00:31:06.640 --> 00:31:12.079
Yeah, I think maybe I will say something which probably is not very popular, but I will say it anyway, right?
00:31:12.319 --> 00:31:20.400
I think if we believe that the documentation is gonna read will be read only by humans, even I would even say five years is too a long time.
00:31:20.559 --> 00:31:26.079
I think even for next three years, probably after three years, I actually we are we are grossly wrong.
00:31:26.319 --> 00:31:32.400
I think most of the most most of the consumption will be will will happen by agents through agentics, right?
00:31:32.640 --> 00:31:35.599
So I think um I think that is one big trend that I see.
00:31:35.759 --> 00:31:47.279
That means if you are a company writing documents or creating documentation only for human consumption, uh for humans to act and read and act, I think we are neglecting the reality that we are living in.
00:31:47.440 --> 00:31:52.799
So that also means that agentic AI, and when I say agentic AI, I'm not talking about chatbots.
00:31:52.960 --> 00:31:57.920
I'm not talking about uh uh your your fancy chat interface that you have.
00:31:58.240 --> 00:32:06.000
No, actually agent that can reason based on your data, and agent with agency, meaning that they can actually act on your behalf.
00:32:06.160 --> 00:32:19.200
So that means agent, I actually believe in five years from now, agent will read a documentation that is grounded in structured truth and not only reason, but actually act on your behalf, right?
00:32:19.359 --> 00:32:32.640
Meaning that agent, because agent needs agency, agency needs control, and control comes from trust, and trust only comes from quality, and and the quality for content comes from structure content, right?
00:32:32.799 --> 00:32:36.559
So that means if you have structured content, you can give your agent agency.
00:32:36.640 --> 00:32:41.920
And when agent have agency, meaning that they can actually not only reason, they can act on your behalf.
00:32:42.079 --> 00:32:43.519
And that's the big trend I see.
00:32:43.680 --> 00:32:47.519
And by the way, this trend is not only for documentation or for content industry.
00:32:47.680 --> 00:32:54.240
Actually, this is basically broadly for whether you're in manufacturing, whether you are in any other industry, this is actually happening, right?
00:32:54.400 --> 00:32:56.400
So I think this is one of the biggest trends we see.
00:32:56.559 --> 00:33:00.480
And and of course, the frontier models are really leading this.
00:33:00.720 --> 00:33:15.039
But but yes, you can do a lot, but the moment you go to enterprise where precision, accuracy, governance, quality is super important, you really need to reside uh, you know, on top of content infrastructure.
00:33:15.440 --> 00:33:15.839
Absolutely.
00:33:16.000 --> 00:33:25.200
Yeah, I picture it like the uh a food pyramid or or you know, where it's like the base, the base model, the base foundation is the structured data, and then everything else flows up from the top.
00:33:25.279 --> 00:33:36.559
Um and if you don't have that good foundation, I mean you're just setting yourself up for uh um, you know, some some problems, some some health issues if it's a food issue, or um just strategical issues if it's a building.
00:33:36.640 --> 00:33:38.480
You know, you don't have that structure.
00:33:38.640 --> 00:33:40.240
Um so very well said.
00:33:40.400 --> 00:34:00.000
Um I I'm also curious about you know your your view on the market and how it's evolved as you've worked in this space, you know, um uh zooming out a little bit from Poligo uh itself and and looking at the broader, broader industry, I know there tends to be you know um habits and assumptions that that hold industries back.
00:34:00.160 --> 00:34:04.160
I'm curious if you could change maybe one thing about how this industry operates today.
00:34:04.240 --> 00:34:06.799
Um, what would that one be one thing be?
00:34:07.440 --> 00:34:11.280
If that is the case, I think I would kill the term documentation.
00:34:11.519 --> 00:34:14.400
And I like and I'll I'll I'll tell why actually.
00:34:14.559 --> 00:34:21.440
I think because documentation, especially coming from the world that we come from and now where we are, it sounds like an overhead, right?
00:34:21.599 --> 00:34:22.480
It's an overhead.
00:34:22.719 --> 00:34:25.679
Something that you do after the product is built, right?
00:34:25.920 --> 00:34:31.920
A cost center that is very easily to minimize simply because you have no understanding of it.
00:34:32.079 --> 00:34:43.920
So, for example, if you're a CTO sitting on top of a large budget where your documentation team is such a small teen teeny winy team sitting in the corner, you can easily be a cost center that can be minimized.
00:34:44.000 --> 00:34:46.880
And by the way, we see a lot of that actually in the news today, right?
00:34:46.960 --> 00:34:51.679
And that where we say that team or that team has been shrunk simply because of cost, right?
00:34:52.079 --> 00:34:56.239
And uh, because nobody gets promoted for a great documentation in an organization.
00:34:56.400 --> 00:34:57.440
We know that, right?
00:34:57.840 --> 00:35:13.440
And uh that's why I think what I would like, rather than calling it documentation, that's why I want to kill the term, I think we need to talk about, as you mentioned, about enterprise knowledge or content infrastructure, which becomes so much mission critical in this age of AI, right?
00:35:13.679 --> 00:35:27.280
So I really, I would really like uh this industry to move away from thinking documentation as a point tool to a mission critical content infrastructure for you to have success with your agent text.
00:35:27.599 --> 00:35:29.760
And that's me is super important.
00:35:29.920 --> 00:35:36.400
The moment you change that conversation, you also change conversation with who you sell, your economic buyer changes, right?
00:35:36.559 --> 00:35:41.840
Because now you're not no longer selling a tool to a document, a tech writer, or a documentation manager.
00:35:42.000 --> 00:35:53.519
You're selling to a CIO, CTO, uh CXOs simply because it becomes a mission critical infrastructure for you to have success with all your fancy uh goals you have with AI.
00:35:54.000 --> 00:35:54.400
Well said.
00:35:54.480 --> 00:35:55.199
Yeah, great answer.
00:35:55.280 --> 00:36:19.360
I wasn't sure um exactly what you you'd say with that question and throwing you on the spot there, but that's uh that's a good uh it's a good thing to to reiterate is just uh the emphasis, the high emphasis on documentation and and and the that that role and and um you know valuing it because again, it's so important to be able to structure all the information that flows into these uh important AI tools.
00:36:19.599 --> 00:36:22.880
So um very, yeah, very very interesting insight there.
00:36:22.960 --> 00:36:33.280
And um I I want to continue to um ask you about what's something surprising you learned about your customers or the market that really changed how you think or operate.
00:36:33.760 --> 00:36:49.920
What I've learned over the over over the you know my career, I would say, I think if you really deeply, deeply empathize with customers, right, and try to understand their pains in really true sense, not superficially, sit with them, listen to them, right?
00:36:50.159 --> 00:37:06.000
And and then and then really see how they go about using your product daily and what kind of pains uh that they have, and then come back and then again empathize with your team to sit and solve their problems and those problems in a new way.
00:37:06.159 --> 00:37:08.559
I always say this actually internally as well.
00:37:08.719 --> 00:37:12.480
The best line of code that you write is the line of code that you don't write.
00:37:12.639 --> 00:37:29.360
So what I mean by that is that because most of the time, what I've seen, and which is quite surprising, if you deeply empathize with the customer about the problem or about uh something that they ask you to build, most of the time what I've seen is that actually it's not about building things, right?
00:37:29.599 --> 00:37:34.320
It's mainly about probably you solve the same problem in a very innovative way.
00:37:34.480 --> 00:37:42.639
If you empathize enough with the customer, and then you come back and do the discovery of that problem space in a deep, meaningful way.
00:37:42.800 --> 00:37:46.079
So you solve the problem without even line writing a single line of code.
00:37:46.320 --> 00:38:12.960
And this is quite a surprise because most of the time what we do is that we try to talk to a customer on very superficially, and then you come back and then you try to fix it in all different ways, and then you really got, you know, you really get caught into that trap of uh just keep on thinking uh, you know, a bit very shallow, based on very limited information, because you haven't enough empathize with the customer and their product, right?
00:38:13.199 --> 00:38:14.239
So that's one side.
00:38:14.400 --> 00:38:23.519
Other thing which I generally believe as well is that especially, so I think just for for your for your audience here, I have been uh part of several industries.
00:38:23.679 --> 00:38:28.239
So so this uh content or documentation industry, this is my seventh industry.
00:38:28.400 --> 00:38:39.840
Before that, I have I happened to work in different uh industries across uh um you know across the globe, where what I've seen is that I have seen several shifts, technological shift that has happened, right?
00:38:40.000 --> 00:38:52.800
Whether whether it's the whether is the uh you know the the shift from you know uh wired to wireless, you know, to you know to to e-com, to app economy, to cloud, right?
00:38:53.360 --> 00:38:55.440
And now to actually agent text, right?
00:38:55.519 --> 00:38:57.679
And agent, first AI and now agentecs.
00:38:57.920 --> 00:39:09.280
What I've seen is every time that paradigm shifts happen, your old problems are probably either they don't exist, or if they exist, you solve it in a very different way.
00:39:09.440 --> 00:39:10.880
So I think so.
00:39:10.960 --> 00:39:27.760
It's so important that whenever these shifts happen, and the one is happening right now with AI, I think we need to start thinking about in this new world, you as an organization and your product that you offer today, what how do you can actually amplify your mode, your DNA in this new world?
00:39:28.000 --> 00:39:33.679
And that's one thing that has always kept me on my toes to always think about whenever this shifts happen.
00:39:33.840 --> 00:39:40.159
We need to make sure that we dress our team, our product, our company to be ready in this new paradigm, right?
00:39:40.320 --> 00:39:41.920
And that's what I'm trying to do with Poligo.
00:39:42.239 --> 00:39:44.079
We come from a documentation for humans.
00:39:44.239 --> 00:39:53.760
Now we are actually creating a mission critical content infrastructure that will become the structured truth layer for enterprise AI, simply because that is what is needed, and that is possible.
00:39:53.840 --> 00:39:56.960
It wasn't possible three years ago, and it was also not needed.
00:39:57.119 --> 00:40:06.719
So, and and in this in this phase, even our users, which is tech writers, as as we discussed earlier, they are also grappling with what is their role.
00:40:06.960 --> 00:40:15.679
But if you come as a thought leader and help them, together you can actually create new opportunities, both for the company, but also for your user in this new one.
00:40:16.079 --> 00:40:16.960
I I think that's great.
00:40:17.119 --> 00:40:18.800
I think empathize with your customers.
00:40:18.880 --> 00:40:23.280
You mean you you start to understand their problems, their their perspective.
00:40:23.360 --> 00:40:35.360
Um, and then yeah, like you were saying, you just you're able to create more um solutions, unique sort of answers that you weren't able to come up with otherwise because you didn't have a maybe better understanding, you didn't know where they're coming from.
00:40:35.519 --> 00:40:44.800
But my mind is going to, you mentioned you, you've you've worked in across seven different industries and whether all these different trans uh transformative technologies.
00:40:44.960 --> 00:40:46.159
I think that's really fascinating.
00:40:46.239 --> 00:40:52.639
And I'd love to hear a little bit more about that and also how it impacts your your leadership style over the over the years.
00:40:52.719 --> 00:41:05.119
You know, I don't want to put you too much on the spot or get you know too crazy personal, but um I'm curious if there's maybe anyone or anything in particular that has really inspired you the most with how you lead a company.
00:41:05.519 --> 00:41:11.760
No, I think that's a I think that's a that's a probably a whole uh podcast in itself of going to their ball, right?
00:41:11.840 --> 00:41:19.199
But I can just tell you, I think, I think one thing which I've learned very early in my career is I think the only constant is change, right?
00:41:19.360 --> 00:41:20.639
I think it's super important, right?
00:41:20.800 --> 00:41:22.400
And I think it's becoming even more, right?
00:41:22.480 --> 00:41:24.079
I mean, we look at what's happening.
00:41:24.159 --> 00:41:36.320
I mean, even though I try to keep myself up to speed on what's happening, uh, right, with technology, but the technology is moving so fast every single day now that I really feel that I have a big time FOMO, right?
00:41:36.400 --> 00:41:38.159
Where I feel like I can't follow it, right?
00:41:38.320 --> 00:41:44.960
So I think what embrace the uncertainty, I think is something that that I've learned in the early part of my career.
00:41:45.119 --> 00:41:53.599
And this is something that I've also tried to bring, uh bring in the teams, in the organizations that that that I have pleasure of leading, right?
00:41:53.840 --> 00:42:09.199
Where I think, you know, if you can create an organization that is grounded in growth mindset, right, with the organization that really values curiosity, high agency, and a lot of experimentation involved, right?
00:42:09.360 --> 00:42:35.760
If you manage to do this and if you create an organization, a system in such a way where your teams, your uh your product teams, your engineering teams, your go-to-market teams, if they thrive in experimentation of learning fast, failing fast, trying new stuff, I think that's the only thing because the technology around us is changing so fast that you can't champion a technology because by the time we champion a technology, something else is coming.
00:42:35.840 --> 00:42:47.039
And now, even the technology is becoming uh so democratized by AI that you don't even need to be a developer actually to write uh to create a new new application, new solutions, right?
00:42:47.199 --> 00:43:05.519
So I think really what you need to do is you need to become, I think I've learned this uh from uh one of the interviews that I have heard from uh CEO from Microsoft, Satya Nadela, that he's he actually, when he took up uh as a CEO of Microsoft, he said that we need to move from a culture of know it all to learn it all.
00:43:05.679 --> 00:43:08.320
And that's the growth mindset, what he talks about all the time.
00:43:08.480 --> 00:43:31.519
And actually, it really resonated very well with me in the in the early part of my own career, and which I really try to bring back, bring with me into the new teams, new organization where I where I come to, where I really want people to learn and and be curious and show that growth mindset simply because everything else is changing so fast that what you think that you knew so well in the past, maybe it's not relevant anymore.
00:43:31.760 --> 00:43:32.320
Right.
00:43:32.719 --> 00:43:40.800
And that's sort of what I uh what I really uh what I really got inspired from back in the days change, growth mindset, right?
00:43:41.039 --> 00:43:42.239
Curiosity, right?
00:43:42.480 --> 00:43:54.480
And then and make sure that you create a system, uh you create a team, a you know, a way of working where you people thrive on change because that's the only constant we have now.
00:43:54.880 --> 00:43:55.920
Very, very well said.
00:43:56.000 --> 00:43:59.760
Yeah, uh the only constant is change, you gotta embrace the uh uncertainty.
00:44:00.559 --> 00:44:02.480
And you know, you can't succeed unless you try.
00:44:02.639 --> 00:44:05.519
I think these are all just great things to keep in mind.
00:44:05.679 --> 00:44:15.679
And yeah, I would love to have you back and just kind of continue to pick your brain about, you know, where the current industry is, the current challenges.
00:44:15.920 --> 00:44:24.159
Rahul, thank you so much for just everything that you've shared with us and everything that you've you've talked about in real regards to PaliGo.
00:44:24.239 --> 00:44:34.800
Um for those who are listening and that would like to get in touch with you or your team and just learn more about this structured data approach that you you've mentioned with PaliGo, um, where should they go?
00:44:35.199 --> 00:44:39.599
Yeah, so I think that of course they can go to Polygo.net, which is our homepage.
00:44:39.760 --> 00:44:42.639
They can also look me up um at LinkedIn.
00:44:42.880 --> 00:44:46.159
I actually share, uh I generally believe in sharing.
00:44:46.320 --> 00:44:53.920
So what we whatever we're building, we're building it an open, meaning that I actually write a lot uh a lot of posts on LinkedIn and a lot of articles on LinkedIn.
00:44:54.079 --> 00:44:56.800
So of course you can look me up with my first name, last name.
00:44:56.880 --> 00:45:00.559
And if you uh search it by Poligo, you will definitely be able to find me.
00:45:00.719 --> 00:45:08.000
Um and also, I mean, you know, feel free to reach out to me through any social media, and I'd be more than happy to engage with that dialogue.
00:45:08.320 --> 00:45:08.639
Perfect.
00:45:08.800 --> 00:45:09.519
All right, thank you so much.
00:45:09.599 --> 00:45:12.400
That's Rahul Yadav, CEO of Paligo.
00:45:12.559 --> 00:45:15.199
Uh, thanks again for all the insights and recommendations you shared with us.
00:45:15.360 --> 00:45:16.159
I really appreciate it.
00:45:16.320 --> 00:45:17.679
Uh, thank you for having me, Bill.
00:45:17.920 --> 00:45:18.880
It was really great.
00:45:19.119 --> 00:45:21.280
And thank you all for listening to the SourceForge podcast.
00:45:21.360 --> 00:45:22.559
I'm your host, Bo Hamilton.
00:45:22.639 --> 00:45:26.800
Make sure to subscribe to stay up to date with all of our upcoming B2B software related podcasts.
00:45:26.960 --> 00:45:28.639
I will talk to you in the next one.
00:00:05.040 --> 00:00:07.839
Hello, everyone, and welcome to the SourceForge podcast.
00:00:07.919 --> 00:00:09.519
I'm your host, Bo Hamilton.
00:00:09.839 --> 00:00:11.919
Let's talk about enterprise AI.
00:00:12.000 --> 00:00:16.640
So, AI has created a little bit of a strange situation inside a lot of companies.
00:00:16.719 --> 00:00:27.839
Uh, the models themselves are getting more capable, but the answers people are getting back can still be rather inconsistent, perhaps a little bit outdated, or uh in some cases just flat out wrong.
00:00:28.000 --> 00:00:37.759
So there's sort of this weird uh disconnect where the technology feels pretty advanced, yet the output still isn't always something you know people can fully trust and get behind.
00:00:37.920 --> 00:00:41.759
And that's uh, I would say really the tension a lot of companies are dealing with right now.
00:00:41.920 --> 00:01:02.240
It's not necessarily that the models are failing, it's that the businesses are trying to layer AI on top of information that might be just scattered across teams, buried in old documents, trapped in various uh systems that might be hard to uh retrieve and access, um, or just you know, otherwise not structured in a way that AI can reliably use.
00:01:02.399 --> 00:01:08.480
So the big question is whether the real problem is the model itself or the mess content sitting underneath it.
00:01:08.640 --> 00:01:11.040
And that's really where our guest comes into play.
00:01:11.120 --> 00:01:18.480
Uh today's guest is Rahul Yadov, CEO of PaliGo, a company built around structured content and documentation workflows.
00:01:18.560 --> 00:01:20.799
Uh, we've got a lot to get into for today's conversation.
00:01:20.959 --> 00:01:30.959
We're going to dig into why enterprise AI keeps hallucinating, what a real source of truth actually looks like, and what companies need to fix before they expect uh AI to be trustworthy.
00:01:31.120 --> 00:01:32.719
So, Rahul, welcome to the podcast.
00:01:32.879 --> 00:01:34.000
Glad you could join us.
00:01:34.400 --> 00:01:35.040
Thank you, both.
00:01:35.280 --> 00:01:36.079
Really nice to meet you.
00:01:36.159 --> 00:01:37.920
And thanks for having me on the podcast.
00:01:38.239 --> 00:01:39.120
Of course, of course.
00:01:39.200 --> 00:01:41.120
So I just want to get right into it, Rahul.
00:01:41.280 --> 00:01:48.400
Um, you've uh, you know, you've argued that this really isn't an AI problem as much as it is a content problem.
00:01:48.560 --> 00:01:51.040
Um when you say that, what exactly do you mean?
00:01:51.120 --> 00:01:55.439
What uh where is like the real breakdown um happening inside most organizations?
00:01:56.319 --> 00:01:57.280
No, true, actually.
00:01:57.359 --> 00:02:03.760
So I think, you know, as we know that most of the enterprises now they're trying or they're racing to deploy AI, right?
00:02:04.000 --> 00:02:10.879
And when we also know that where the deployment fails, it's actually failing at what I call last mile, right?
00:02:11.039 --> 00:02:17.919
When they when they really try to put into production, uh, you know, these agents, super fancy agents, they fail, right?
00:02:18.159 --> 00:02:24.719
And I think most of the time people blame the model um, you know, when their AI gives the wrong answers.
00:02:24.960 --> 00:02:27.599
But but think about it, what actually happens, right?
00:02:27.759 --> 00:02:30.159
So essentially what as you started with, right?
00:02:30.319 --> 00:02:36.560
You take um you take a SharePoint folder, for example, which has hundreds of documents in different formats.
00:02:36.719 --> 00:02:38.639
It could be PDF, it could be Word.
00:02:39.039 --> 00:02:46.800
And we also know that how we even use our own documents on our on our own desktop, many are outdated, many are duplicated, right?
00:02:47.039 --> 00:02:56.240
And I think especially when you have these documents uh and PDFs, what I call flat file, they have no semantic structure.
00:02:56.560 --> 00:03:09.680
So what normally happens is that when you throw this to a more sophisticated or more sophisticated LLM, you know, what happens is that you chunk into vectors, you chunk it, you chunk the document using any kind of rack systems.
00:03:09.840 --> 00:03:17.360
So what happens is that you feed into one of the most sophisticated, whether it's a GPT-4 or a Claude, you know, with Opus 4.7.
00:03:17.599 --> 00:03:20.719
I mean, you know, these are brilliant reasoning engines.
00:03:21.280 --> 00:03:29.520
And when, and then you're surprised when you when it can't distinguish between, let's say, your product X400 versus a product X500.
00:03:29.759 --> 00:03:34.960
Actually, it's not, you know, the model isn't just hallucinating because it's stupid, right?
00:03:35.120 --> 00:03:39.520
It is hallucinating because it has no access to the truth, the structured truth, right?
00:03:39.680 --> 00:03:53.520
Because you give it a mess and then you ask your model to be more uh concrete and more confident, which by the way, it will be when you when you see it, but most of the time it actually hallucinates and give wrong answers.
00:03:53.840 --> 00:03:56.719
Yeah, when you when you put it like that, it makes it makes total sense.
00:03:56.879 --> 00:04:13.199
I mean, uh if it if some of these documentation structures um look anything like uh my my documents folder and my computer, what and and the various naming structures and um just where everything is is located, I I can see why there would be there would be issues, right?
00:04:13.360 --> 00:04:27.600
And so um when you deal with these these massive uh databases and internal systems, um I can see the complexity there trying to retrieve valuable information and then just having to lean heavily on AI trying to extrapolate and fill in the missing pieces.
00:04:27.759 --> 00:04:31.279
I can just see uh, you know, it's like it's a recipe for disaster, really.
00:04:31.439 --> 00:04:40.160
Um and you know, it kind of goes to the baseline of like AI uh or an LLM is really only as good as the data it has to work with, you know.
00:04:40.639 --> 00:04:41.040
Exactly.
00:04:41.199 --> 00:04:44.399
And I also feel actually people feel that the fix is a better model.
00:04:44.639 --> 00:04:52.240
But actually, on the contrary, the fix is, yes, a better model, but fix is also structured governed content, which is grounded in truth.
00:04:52.720 --> 00:04:53.279
Yeah, yeah.
00:04:53.439 --> 00:04:55.199
It's not always the better model.
00:04:55.279 --> 00:05:01.920
Because I think um we've kind of also plateaued in terms of you know some of these sophistications uh with the models themselves.
00:05:02.079 --> 00:05:07.199
And so a lot of it's just uh yeah, integrating the right data and and into the model.
00:05:07.439 --> 00:05:20.639
Um and also I think it's it's interesting to think about too, you know, with the enterprise market, you you know, you're working with perhaps AI systems and models that are a little bit more closed off.
00:05:20.800 --> 00:05:30.399
They're not able to go tap into some of these bigger parameters and resources that maybe some of the consumer-facing chatbots can do uh for you know for privacy and security reasons.
00:05:30.560 --> 00:05:32.079
So that kind of limits them further.
00:05:32.319 --> 00:05:46.000
But um, this this issue is uh the basis of one of the ideas I've heard you talk about in enterprise AI, and that's uh called the frontier gap, uh where the model itself is you know intelligent, but it has no access to reliable enterprise truth.
00:05:46.160 --> 00:05:50.399
Um when you talk about that gap, how does it actually show up in practice?
00:05:50.800 --> 00:05:51.360
Yeah, sure.
00:05:51.439 --> 00:05:55.439
So I think the frontier gap, I mean, you know, just for your listeners, right?
00:05:55.519 --> 00:06:03.839
So the frontier gap, what I call is a distance between what a model can do and what enterprise can safely deploy.
00:06:04.160 --> 00:06:09.439
That is actually, so yes, the model, these models, as you said, are becoming very sophisticated.
00:06:09.600 --> 00:06:12.879
You know, they are really getting better at reasoning.
00:06:13.120 --> 00:06:20.240
But how much, especially in the enterprise setting, how much you can actually safely uh deploy is a key here, right?
00:06:20.399 --> 00:06:21.920
So models are brilliant.
00:06:22.160 --> 00:06:31.519
I mean, you know that cloud can reason through complex problems, so is uh so is GPT-4, for example, can synthesize information better than us as humans, right?
00:06:31.839 --> 00:06:39.680
But when you point this powerful intelligence at an enterprise content, there is this gap that comes in, right?
00:06:39.839 --> 00:06:43.120
So model doesn't know which content is current, right?
00:06:43.439 --> 00:06:49.839
Model doesn't know which content is approved, which applies to which which content applies to which product.
00:06:50.560 --> 00:06:50.879
Right.
00:06:50.959 --> 00:06:56.560
And then also like you know, which content is written for internal use versus customer-facing use.
00:06:56.800 --> 00:07:07.759
It doesn't know which paragraph was deprecated six months ago, or whether this paragraph requires a legal review before you put it to external use.
00:07:08.079 --> 00:07:08.480
Right?
00:07:08.800 --> 00:07:13.839
So what you get out of all of this is you really get confident hallucinations.
00:07:13.920 --> 00:07:14.879
That's what I say, right?
00:07:15.040 --> 00:07:20.639
A customer asks about safety procedure, AI will cite a manual from 2019.
00:07:20.800 --> 00:07:30.000
Maybe that manual was deprecated before, but it sounds certain because the model is certain and it just didn't have the access to the truth that you have, right?
00:07:30.160 --> 00:07:31.759
And this is where people actually miss.
00:07:31.839 --> 00:07:37.519
This is where people miss that that you know the models are really doing a lot on model safety, right?
00:07:37.680 --> 00:07:41.519
They're making sure the model doesn't generate harmful content, right?
00:07:41.839 --> 00:07:44.879
Or or it doesn't help with illegal activities.
00:07:45.120 --> 00:07:52.959
But what's missing, especially from an enterprise perspective, is enterprise governance that is left to the enterprises, right?
00:07:53.199 --> 00:08:01.040
So so this is where this frontier gap exists because the model providers build the intelligence layer, but nobody built that truth layer.
00:08:01.199 --> 00:08:03.759
And that's the infrastructure that is missing in my mind.
00:08:04.079 --> 00:08:05.040
Yeah, that makes sense.
00:08:05.120 --> 00:08:14.720
Um I'm curious, like if can I go back going back to what you were saying earlier, is if if the real issue is not the model itself, but the content underneath it, right?
00:08:14.879 --> 00:08:20.959
That raises the next sort of question that comes to my mind is what does a strong foundation actually look like?
00:08:21.040 --> 00:08:29.439
You know, I know this is this is really how how your company has kind of made its bread and butter, its its name for itself, building around this idea of structured content.
00:08:29.600 --> 00:08:38.720
Um, but for people who may not live and and breathe documentation systems, what would you say makes for an effective documentation system in the first place?
00:08:38.879 --> 00:08:43.200
And then why does that structure just matter so much when AI enters the space?
00:08:43.600 --> 00:08:44.399
No, absolutely.
00:08:44.480 --> 00:08:49.360
So so I think you know, it all came from so Poligo was formed um, you know, 10 years ago, right?
00:08:49.519 --> 00:08:58.080
Ten years ago, we started with uh with creating structured authoring uh tool, which is what they call CCMS, content component management system, right?
00:08:58.240 --> 00:09:00.879
Which was based on DogBook XML as a standard.
00:09:01.120 --> 00:09:20.320
Now, the reason we created this is because we know that when you when you achieve a certain level of complexity in terms of your document needs into an organization, you need a tool like Poligo to not only author and create content, but also create a lot of reuse and publication and translation and so on and so forth, the entire life cycle of content.
00:09:20.879 --> 00:09:24.639
Now, and this is why we started with structure in the first place.
00:09:25.039 --> 00:09:39.120
Now, what happens is when you talk about what actually is structured content, structure content for us is content that has semantic understanding, that means all the metadata, version controlling, right?
00:09:39.279 --> 00:09:41.200
And all of that combined together.
00:09:41.360 --> 00:09:54.080
Um, you know, if you can offer this to your not only to humans, but also to AI, what you can do is you can actually uh get much better and accurate answers on your other side, right?
00:09:54.240 --> 00:09:58.720
So what Poligo and in Poligo, what we are trying to do now, we come from a structured content.
00:09:58.879 --> 00:10:17.519
And we are moving to what we call structured truth for enterprise AI, meaning that if you really write content and create content in Poligo, we will provide a solution, a mission critical infrastructure from content ingestion, authoring, publication, delivery to consumption and grounding on the other side.
00:10:17.759 --> 00:10:32.000
And this is what sound structured content infrastructure should look like, where you offer semantic, um, uh semantically rich content, which is perfect fuel for agent techs on the other side.
00:10:32.480 --> 00:10:42.320
Yeah, it's interesting to hear that you've you've you know been working on um structured data for the past decade, you know, before this this AI boom that we've we all found ourselves in.
00:10:42.399 --> 00:10:53.919
I imagine back in back then it was really um just around cloud-based systems and getting that all up and running and and making that easier to to access and and um you know utilize.
00:10:54.159 --> 00:11:09.840
Um, but you're kind of ahead of your time there in terms of what you guys are tackling because uh having a really thoughtful, well-structured um system is is just gonna reward yourself uh leaps and bounds with um some of these AI tools you're able to throw in on top.
00:11:10.080 --> 00:11:23.440
Um but uh yeah, I wish I I wish I could like interview you and um or the team like 10 years ago to see kind of your headspace and where where things were going back then and compare it where you're at now, because um it's so interesting how things have changed.
00:11:23.600 --> 00:11:31.360
But you know, a lot of a lot of the market right now, I would say it feels like there's sort of this race to add AI features as quickly as possible.
00:11:31.519 --> 00:11:35.279
And and I think from the outside that that looks like progress, right?
00:11:35.360 --> 00:11:40.799
Even though it's it's definitely a different story from some of the the insiders actually using the tools.
00:11:41.039 --> 00:11:43.840
Um, you've you've obviously taken a pretty different view.
00:11:43.919 --> 00:11:52.879
What what's what's the risk when companies try to layer AI onto a content environment that was already sort of fragmented or poorly organized to begin with?
00:11:53.279 --> 00:11:55.600
Yeah, actually internally I joke this around.
00:11:55.679 --> 00:11:56.399
Uh Bo.
00:11:56.720 --> 00:12:00.960
I actually call it uh that model of bolting features on top of the existing product.
00:12:01.120 --> 00:12:03.200
I call it a lipstick pick model, right?
00:12:03.360 --> 00:12:11.120
So you're putting a lipstick on a pick because everything else is ugly, and you just try to make it beautiful by just putting a lipstick, which is an AI lipstick, right?
00:12:11.279 --> 00:12:23.200
Um so I think what I really feel uh is that what we have taken in the approach that we have in Poligo is we will we have said that we are not gonna bolt just AI feature for the sake of AI feature, right?
00:12:23.440 --> 00:12:26.799
And and the reason is because we really believe that model is broken.
00:12:26.960 --> 00:12:39.200
Uh and if a foundation is not there, you just can't you can really throw in a lot of features, but but those features will not be very um you know, very helpful for our users, right?
00:12:39.360 --> 00:12:53.360
So, first of all, we said that even if we bolt some features, it has to be context-aware, meaning that we will not just have an assistant on the side, which no one uses because you have an AI assistant, because apparently everyone loves co-pilot over the last couple of years, right?
00:12:53.519 --> 00:13:00.240
But if the co-pilot is not in the context where the where the user is in their workflow, I think it's not useful enough.
00:13:00.320 --> 00:13:09.600
So we our idea has always been how can we augment a particular workflow or a step in a workflow for a user and make a user more efficient.
00:13:09.759 --> 00:13:11.600
So that's what one approach we are looking into.
00:13:11.759 --> 00:13:19.279
But we see that if we really have this bolt-on features on top of this, I actually feel that you end up into these three things, right?
00:13:19.519 --> 00:13:23.120
Number one, I would actually say that liability, right?
00:13:23.360 --> 00:13:32.159
Simply because if you don't have the right foundation, the content foundation, a clean structure, truth infrastructure, you will have hallucination.
00:13:32.320 --> 00:13:46.480
And especially in the regulated industries, Bo, we will end up actually having a lot of liability issues simply because just imagine if your agent hallucinates a wrong procedure, a safety procedure to a customer.
00:13:46.720 --> 00:13:52.080
Imagine the kind of liability, especially if you are processing an insurance claim, for example, right?
00:13:52.240 --> 00:13:56.480
It can probably refund a crazy amount by just simply because it's hallucinating.
00:13:56.639 --> 00:13:58.159
So there's a liability issue, right?
00:13:58.399 --> 00:14:09.120
Second thing I actually believe, which not that many people talk about, the benefit of structure content and structure content to feed AI, especially in enterprise, is your runway cost.
00:14:09.679 --> 00:14:21.440
If you don't have structured content underneath and structure uh structured infrastructure for content proper, proper infrastructure in place, you will have a lot of token cost consumption, right?
00:14:21.679 --> 00:14:28.559
Consumption that you use for tokens, the cost will be very high simply because you need to do multiple retries to get right answers.
00:14:28.720 --> 00:14:34.399
Your RA needs to create a lot of chunks in terms of and create vector database and then do all of that.
00:14:34.559 --> 00:14:38.639
It requires a lot of consumption from an AI, intelligence, and token cost perspective.
00:14:38.799 --> 00:14:47.440
But on the contrary, if you have structured content, your cost for consuming that will be much lower and you can get better answers for a lower price point.
00:14:47.600 --> 00:15:04.559
And third, if you don't have a good infrastructure, content infrastructure in place, I think you will promote that silo that we also used to talk about in the world of data and analytics in the past, where every team was creating their own tool and their own platform, and then the and the data layer was not connected together.
00:15:04.720 --> 00:15:13.840
You're gonna amplify this with Agentix simply because every single team will create their own agents and with on their own top on their own top of their own data.
00:15:14.000 --> 00:15:16.240
Like for example, marketing will create their own.
00:15:16.480 --> 00:15:23.039
So is product, so is uh sales, and and all of those will have their own data silo.
00:15:23.120 --> 00:15:28.960
So just imagine the same question asked through different three different agents will give three different answers.
00:15:29.200 --> 00:15:30.960
Simply because they are on this there, right?
00:15:31.279 --> 00:15:44.399
So I actually believe that the if you take a shortcut, you be you know, you you come into the trap where you will end up having these three risks I talk about the liability risk, your runway cost risk, and then your silo.
00:15:44.960 --> 00:15:46.000
Those are really good points.
00:15:46.080 --> 00:15:47.360
Yeah, all really good points.
00:15:47.519 --> 00:16:06.320
I I think the the tokens, um, uh the second point in particular really, really um speaks to the the current sort of um you know craze right now and with AI and and just how companies, people are burning through tokens, trying to utilize AI um because of all the the promise and what it's able to do with with fib coding and and whatnot.
00:16:06.399 --> 00:16:08.399
And the cost gets out of control.
00:16:08.480 --> 00:16:20.000
So if you're able to just have have it become more efficient, which is the whole goal of AI in the first place, is just kind of the pushing forward this efficiency era, um, then you're gonna you're gonna save some money.
00:16:20.080 --> 00:16:26.320
But also the legal standpoints and um all the other points you mentioned too are really, really do make a lot of sense.
00:16:26.480 --> 00:16:32.480
And um I again, you're really just by taking a shortcut, you're really just setting yourself up for a headache in the long run.
00:16:32.639 --> 00:16:38.159
Um so yeah, it's it's um these are really good points to to keep in keep in mind.
00:16:38.240 --> 00:16:54.159
And um I I want to continue to to drill down and um, you know, I guess just also open this up to some more tangible examples that you might think of than uh, you know, where you've implemented a strategy um that's actually really had a lot of reward for for clients.
00:16:54.320 --> 00:17:05.359
I know um one that stood out to me is how you've talked about taking a translation process that used to take weeks uh and shrinking that down to something that can happen in literally hours inside the platform.
00:17:05.519 --> 00:17:07.599
Can you walk us through that that example?
00:17:07.759 --> 00:17:14.799
What had to be true behind the scenes from an architectural standpoint for that kind of lead to even be possible?
00:17:15.279 --> 00:17:17.359
No, I think that's uh that's a great example.
00:17:17.519 --> 00:17:21.680
And I think, by the way, we released that AI translation just a month ago, right?
00:17:21.759 --> 00:17:23.680
So so take an example, right?
00:17:23.759 --> 00:17:26.640
I think um traditional translation, we all know that, right?
00:17:26.720 --> 00:17:28.079
It's painful, right?
00:17:28.240 --> 00:17:33.519
The reason it's painful is because you create content one in one place, then you need to export that content.
00:17:33.839 --> 00:17:42.079
You need to send it to this uh this uh exported content that you have uh to a translation management system, right?
00:17:42.319 --> 00:17:48.079
Or you send it uh or you send it manually basically to a translator translation agency.
00:17:48.480 --> 00:17:50.160
Then you wait for two, three weeks.
00:17:50.720 --> 00:17:55.279
Then once the translation comes in, you re-import that content inside your system.
00:17:55.680 --> 00:18:01.440
And then basically you end up into this, what I call uh reconciliation nightmare, right?
00:18:01.680 --> 00:18:09.200
Because maybe in this two, three weeks, your original version, which was written in English, maybe that has evolved, right?
00:18:09.440 --> 00:18:16.400
Uh because now the version that you sent for translation compared to what you have actually now have in your system, it is different, right?
00:18:16.720 --> 00:18:19.920
So how do you now with that, how do we keep this?
00:18:20.000 --> 00:18:26.960
Maybe you have 47 languages that that needs to be in sync when when the English version keeps evolving, right?
00:18:27.119 --> 00:18:29.839
And this is actually a very tangible example, right?
00:18:30.000 --> 00:18:33.200
Where we have seen a lot of our customers struggling with that.
00:18:33.440 --> 00:18:40.000
And now when you replace this with AI translation, which doesn't happen in weeks, it happens in minutes, right?
00:18:40.160 --> 00:18:43.920
Simply because you can actually use uh AI to do that, right?
00:18:44.079 --> 00:18:56.400
Within and inside the platform, the authoring tool that you're using, like Poligo CCMS, you can use AI translation where you just click a few buttons and then you translate, and you really get a very accurate translated content back.
00:18:56.640 --> 00:19:15.519
Because what we do is that we we are very since it's a component content management, we can go all the way to granular level of a component while also getting all the advantage of uh component content management in terms of governance, in terms of versioning, in terms of uh semantic and metadata and all of that, right?
00:19:15.759 --> 00:19:24.079
So so this is what actually you know, you can see a very concrete example where where where users are already getting benefit out of that, right?
00:19:24.319 --> 00:19:37.920
And and architecturally, I mean, I will be I think that basically what you need to do is that you will only be able to do it if you have structured content, which is which is written in a component which is which are reusable, right?
00:19:38.000 --> 00:19:39.599
So you can reuse it, right?
00:19:40.000 --> 00:19:42.799
Each component is linked to a version.
00:19:42.960 --> 00:19:52.559
So you actually know that that because of versioning, even though your original content continues to evolve, you can still reconcile, can still uh still do that, right?
00:19:52.640 --> 00:19:53.359
So that's there.
00:19:53.519 --> 00:20:02.319
And then also make sure that the translation governance is to be in few, you know, is is uh is and the versioning governance is enforced as well, right?
00:20:02.480 --> 00:20:05.200
So for example, Gross V style guides, right?
00:20:05.279 --> 00:20:13.200
For example, brand terminology, taxonomy, they are all uh not deployed manually by human, but actually it's already taken care of by translation.
00:20:13.359 --> 00:20:15.359
This is enforced automatically by system.
00:20:15.680 --> 00:20:19.440
And I can tell you, this is the byproduct of all of this is speed.
00:20:19.839 --> 00:20:29.359
And especially when teams are deploying code or or releasing product based on continuous delivery and continuous deployment, you don't have weeks to wait for.
00:20:29.440 --> 00:20:37.599
And now with AI translation, you you know, translation is no longer, thanks to AI, is no longer a bottleneck in terms of your throughput type.
00:20:37.839 --> 00:20:42.799
Yeah, I mean, it literally has a direct impact on the operational impact, right?
00:20:42.880 --> 00:20:46.000
Um, which it has a direct impact on speed and so many other things.
00:20:46.079 --> 00:21:03.839
Um not only is it hallucinating less, but it's also it's it's retrieving information faster, it's delivering, you know, more accurate, truthful answers uh faster, which is kind of the whole, again, the whole point of AI in the first place is doing things uh more efficiently and quicker and helping you do things more efficiently and quicker than before.
00:21:04.000 --> 00:21:10.559
Um and I think the this this conversation also changes how we think about people behind the scenes, right?
00:21:10.640 --> 00:21:20.480
Um and I uh one sort of uh a job and role that comes to mind is is are the technical writers that have uh been seen as more of a maybe supporting role um over the years.
00:21:20.640 --> 00:21:25.759
You know, they're important, but they're not always brought into the conversation as like a strategic part of the business.
00:21:25.920 --> 00:21:29.519
Um, but things are changing in this AI first uh world.
00:21:29.680 --> 00:21:40.000
Um in this world we live in with AI kind of uh just being incorporated in every way, shape, and form, um what does that role for technical writers look like now?
00:21:40.160 --> 00:21:44.240
I'm curious, like, where do technical writers sit in the bigger picture?
00:21:44.559 --> 00:21:46.640
Yeah, I think I think this is very interesting.
00:21:46.720 --> 00:21:49.839
And and I think this is also one of the missions that we have at Poligo.
00:21:50.160 --> 00:21:58.480
We actually believe that the right the role of tech writers is gonna become even more important in this AI era, it's very, very clear.
00:21:58.720 --> 00:21:59.839
I actually believe that the tech writers are not.
00:22:00.160 --> 00:22:13.440
Writers are going to become the superhero in this in this AR race, especially if you're coming from a very sound background of technical documentation, where structure is really what you really bring to the table.
00:22:13.599 --> 00:22:15.279
I think your role will become very, very important.
00:22:15.440 --> 00:22:28.400
I actually believe that the role of technical writer will evolve into, I mean, you know, internally we call it content curator or content strategist or maybe content architect or content directors, right?
00:22:28.640 --> 00:22:30.079
This is the role you're going to take.
00:22:30.319 --> 00:22:40.480
Because now the content that you used to write for human consumption, right, is it's of course, there will still be need actually in the future in certain industries, in certain verticals.
00:22:40.640 --> 00:22:51.680
But I actually believe that the content that you will orchestrate, the content that you will direct, will also be relevant for, will be more relevant, I would say, for agentics and AI.
00:22:51.839 --> 00:22:53.279
So I think that all will evolve.
00:22:53.440 --> 00:23:02.799
So meaning that yes, I think if you come from purely writing content from scratch, I can tell you AI can take you from zero to 80% in no time, right?
00:23:02.960 --> 00:23:14.079
So content generation will, of course, if you only do content generation and if you are really not architecting content and not structuring and creating structure and creating semantics, then of course you need to think about it.
00:23:14.160 --> 00:23:20.000
But I actually believe that content generation from zero to 80% will become very easy, which is a solved problem.
00:23:20.160 --> 00:23:30.160
But then still governing that content, making sure that it is rich in metadata, in context, in versioning and all of that, really curating that, that will be very important.
00:23:30.319 --> 00:23:36.079
And I actually believe, which is even more important, which I don't think that there are that many many people talk about it.
00:23:36.319 --> 00:23:43.119
Remember, most of those tech writers are also having a very strong domain expertise, a certain domain they represent.
00:23:43.359 --> 00:23:49.200
And as we know very clearly, what is more important in this world of AI is domain expertise.
00:23:49.359 --> 00:24:03.359
So if you are really coming from, let's say, a pharmaceutical industry or a manufacturing industry, where you have been tech writing those content pieces for safety regulation, uh right, use, right, manuals, all of that will become even more important.
00:24:03.519 --> 00:24:23.359
And how it is written, how it is stored, you know, we are really powered by metadata and semantic, all of that will become even more relevant for agent tech simply because now you need to make sure that you know when you have rich content at the source, you can actually have better answers or hallucination-free or fault-free answers on the other side.
00:24:23.519 --> 00:24:34.559
And then I think to take a further down the line, when the content is being consumed by AI, these tech writers could also become human in the loop to enforce reinforcement learning.
00:24:34.720 --> 00:24:43.599
Meaning that, meaning that if they model uh hallucinate, this, you know, when someone needs to correct it, we need humans to to really correct it.
00:24:43.759 --> 00:24:44.799
And who are those humans?
00:24:44.960 --> 00:24:50.240
There's gonna be, again, our content curators or the tech writers, because they know what is good and what is bad.
00:24:50.559 --> 00:24:52.319
I I love I I love this take.
00:24:52.480 --> 00:24:52.640
Yeah.
00:24:53.839 --> 00:24:54.960
I love that take that you have.
00:24:55.039 --> 00:25:19.680
And um, I mean, in full transparency, this is what um you know, what we rely on and and lean heavily into with SourceForge and um using our technical writers to you know um work directly with with companies on that have uh listings on our site, uh, having these conversations, you and you and I having this conversation, um, and using our our domain authority to to kind of bring it to the masses, right?
00:25:19.759 --> 00:25:22.400
Um and get that valuable, you know, truthful information out there.
00:25:22.480 --> 00:25:26.720
So you don't have to the uh AI can use our content as a source.
00:25:26.880 --> 00:25:32.079
Um and again, it's all about coming back to this, this creating this knowledge layer that AI depends on.
00:25:32.160 --> 00:25:36.559
Um and I I think that's that's really the the value um that tactical writers offer.
00:25:36.640 --> 00:25:41.680
And I think that uh, you know, writers of all kinds have been put on a pedestal for me, I would say.
00:25:41.759 --> 00:25:52.000
Um I'm a little bit biased, of course, but um this way of thinking about the role really does uh, you know, it feels a lot more strategic than the way it's traditionally been viewed.
00:25:52.160 --> 00:25:54.880
And I I think that's um that's that's great.
00:25:55.039 --> 00:26:04.559
I think that's that I think a lot of uh writers who are uncertain about their future um are feeling a lot better based off your answer here.
00:26:05.039 --> 00:26:06.160
No, I I hope so.
00:26:06.240 --> 00:26:14.160
And I generally believe actually that both this is this is so crucial now because I think you know AI is really good in a lot of generic things.
00:26:14.400 --> 00:26:21.039
The moment you come become very specific into an enterprise, into a certain workflow with certain business logic, business rule.
00:26:21.200 --> 00:26:27.200
I mean, you know, tech writers are the champions for that business, and I think they need to come in the center of it.
00:26:27.279 --> 00:26:33.920
So I also believe that the companies that treat documentation as a cost center, I actually believe they will have a tough time.
00:26:34.000 --> 00:26:40.400
They will lose to the companies that treat the documentation and the documentation team as a strategic asset, right?
00:26:40.559 --> 00:26:42.720
And that transition is happening, I would say, now.
00:26:42.880 --> 00:26:49.920
And I actually am having so many conversations with our customers, and I can really see that that conversations are changing.
00:26:50.000 --> 00:27:07.440
But in some places, also conversations are still about cost center, where where you know you can see that hey, that team of documentation uh managers or documentation or tech writers have been shrinked without decision makers being aware actually that it's just such a strategic function in the world of AI.
00:27:07.839 --> 00:27:19.680
Yeah, I mean, they're becoming more um integrated and a part of the their company with business leaders trying to piece together all these different um sources of information and um structure their the data accordingly.
00:27:19.839 --> 00:27:24.319
So I think it's um yeah, I think it's a it's it's really uh I guess a win-win.
00:27:24.400 --> 00:27:42.160
And also um it just it it's again, it's great to hear because I think there's a lot of uh writers and all um you know capacities that are feeling a little bit discouraged, I guess, with um some of these these AI content generation, you know, capabilities out there.
00:27:42.319 --> 00:27:49.039
Um but it's like you're saying, it's not just the content generation, it's the structuring of the data that's increasingly more important.
00:27:49.519 --> 00:28:05.680
So um, but yeah, I think getting back to the the content layer part of the conversation and how important that is, um I know there's obviously a lot of pressure on CTOs right now to move to move quickly on AI, but um I know speed can also hide a lot of the sort of bad assumptions.
00:28:05.920 --> 00:28:14.640
If you could get every CTO to really understand one big thing about their content before they they launch an AI project, um what would you want that lesson to be?
00:28:15.039 --> 00:28:17.599
I think I think the less the lesson is very simple.
00:28:17.759 --> 00:28:20.799
I think you know, garbage in, garbage out, right?
00:28:20.960 --> 00:28:22.079
Is very, very clear.
00:28:22.319 --> 00:28:29.119
I know, you know, in my one of my uh companies that I worked um a couple of years ago, it was a media streaming company.
00:28:29.279 --> 00:28:32.319
I used to say that content we used to say content is king in media.
00:28:32.480 --> 00:28:33.759
We used to say content is king.
00:28:33.920 --> 00:28:36.480
I actually believe the structured content is king comm.
00:28:36.720 --> 00:28:42.079
Right and that's what that's what the CTOs, CIOs, CAIOs need to understand.
00:28:42.240 --> 00:28:53.279
I think your your AI project that you're you are you're deploying for millions and billions now will succeed or fail, will depend on the foundation of your structured content.
00:28:53.440 --> 00:28:58.319
If you have really good quality structured content, I think your project will succeed.
00:28:58.480 --> 00:29:02.079
If not, you're gonna spend the next couple of years building that foundation.
00:29:02.240 --> 00:29:08.160
Just like what happened when the data uh revolution came happens, right, for analytics, right?
00:29:08.319 --> 00:29:16.079
Every single company, every single CTO, CIOs, what they have done is they've been chasing to really bring data out of silos, right?
00:29:16.240 --> 00:29:21.279
Which was enterprise data, HRM data, CRM data, you name all kinds of business data, right?
00:29:21.359 --> 00:29:27.839
You brought all of that, and then you wanted to really host it somewhere in a data lake where you can actually do a lot of analytics on top of it.
00:29:28.079 --> 00:29:29.920
Content has been overlooked.
00:29:30.160 --> 00:29:37.680
And now we know that these LLMs, they are fed on human language, normal language, uh data, right?
00:29:37.839 --> 00:29:39.759
Written docs documentation, right?
00:29:39.920 --> 00:29:40.720
Content, right?
00:29:41.039 --> 00:29:48.400
Which is residing in um silo and retire and residing in Word doc, Google Doc, PDFs, and all of that, right?
00:29:48.559 --> 00:29:51.440
Duplicated version, multiple copies, all of that.
00:29:51.599 --> 00:29:54.319
All of that needs to be needs to be structured.
00:29:54.640 --> 00:30:06.559
And and if you are a CTO, CIO listening, this, if you feel you can fix it retrospectively, what I call fixing it downstream, the cost is very high and the technology is not there.
00:30:07.039 --> 00:30:10.799
You need to start thinking about uh you know, creating purity at source.
00:30:10.960 --> 00:30:23.759
And purity of source is that if you create structured, semantically rich, componentized structure using a tool, I think that will be that will be what and then you need to start thinking content as an infrastructure layer.
00:30:24.480 --> 00:30:25.119
Great advice.
00:30:25.279 --> 00:30:26.400
Structure is king.
00:30:26.480 --> 00:30:29.680
That's that's uh that's a great um that's a great thing to remember.
00:30:29.839 --> 00:30:37.200
And again, it feels like AI, you know, success, it starts with chasing the model, but um in reality, it's it starts much earlier than that.
00:30:37.279 --> 00:30:45.440
It starts with the the understanding your content, your systems, um, and and your internal source of truth, make sure the structure is all there.
00:30:45.599 --> 00:30:49.359
Um so so very good, very good things to underline and underscore.
00:30:49.519 --> 00:30:56.079
Um now looking ahead at where things are going, I'm curious to see where things are uh continue to evolve from here.
00:30:56.319 --> 00:31:06.319
What emerging trends do you think will have the biggest impact on enterprise documentation in the next few years, let alone five years from now, if we can even um plan that far in advance?
00:31:06.640 --> 00:31:12.079
Yeah, I think maybe I will say something which probably is not very popular, but I will say it anyway, right?
00:31:12.319 --> 00:31:20.400
I think if we believe that the documentation is gonna read will be read only by humans, even I would even say five years is too a long time.
00:31:20.559 --> 00:31:26.079
I think even for next three years, probably after three years, I actually we are we are grossly wrong.
00:31:26.319 --> 00:31:32.400
I think most of the most most of the consumption will be will will happen by agents through agentics, right?
00:31:32.640 --> 00:31:35.599
So I think um I think that is one big trend that I see.
00:31:35.759 --> 00:31:47.279
That means if you are a company writing documents or creating documentation only for human consumption, uh for humans to act and read and act, I think we are neglecting the reality that we are living in.
00:31:47.440 --> 00:31:52.799
So that also means that agentic AI, and when I say agentic AI, I'm not talking about chatbots.
00:31:52.960 --> 00:31:57.920
I'm not talking about uh uh your your fancy chat interface that you have.
00:31:58.240 --> 00:32:06.000
No, actually agent that can reason based on your data, and agent with agency, meaning that they can actually act on your behalf.
00:32:06.160 --> 00:32:19.200
So that means agent, I actually believe in five years from now, agent will read a documentation that is grounded in structured truth and not only reason, but actually act on your behalf, right?
00:32:19.359 --> 00:32:32.640
Meaning that agent, because agent needs agency, agency needs control, and control comes from trust, and trust only comes from quality, and and the quality for content comes from structure content, right?
00:32:32.799 --> 00:32:36.559
So that means if you have structured content, you can give your agent agency.
00:32:36.640 --> 00:32:41.920
And when agent have agency, meaning that they can actually not only reason, they can act on your behalf.
00:32:42.079 --> 00:32:43.519
And that's the big trend I see.
00:32:43.680 --> 00:32:47.519
And by the way, this trend is not only for documentation or for content industry.
00:32:47.680 --> 00:32:54.240
Actually, this is basically broadly for whether you're in manufacturing, whether you are in any other industry, this is actually happening, right?
00:32:54.400 --> 00:32:56.400
So I think this is one of the biggest trends we see.
00:32:56.559 --> 00:33:00.480
And and of course, the frontier models are really leading this.
00:33:00.720 --> 00:33:15.039
But but yes, you can do a lot, but the moment you go to enterprise where precision, accuracy, governance, quality is super important, you really need to reside uh, you know, on top of content infrastructure.
00:33:15.440 --> 00:33:15.839
Absolutely.
00:33:16.000 --> 00:33:25.200
Yeah, I picture it like the uh a food pyramid or or you know, where it's like the base, the base model, the base foundation is the structured data, and then everything else flows up from the top.
00:33:25.279 --> 00:33:36.559
Um and if you don't have that good foundation, I mean you're just setting yourself up for uh um, you know, some some problems, some some health issues if it's a food issue, or um just strategical issues if it's a building.
00:33:36.640 --> 00:33:38.480
You know, you don't have that structure.
00:33:38.640 --> 00:33:40.240
Um so very well said.
00:33:40.400 --> 00:34:00.000
Um I I'm also curious about you know your your view on the market and how it's evolved as you've worked in this space, you know, um uh zooming out a little bit from Poligo uh itself and and looking at the broader, broader industry, I know there tends to be you know um habits and assumptions that that hold industries back.
00:34:00.160 --> 00:34:04.160
I'm curious if you could change maybe one thing about how this industry operates today.
00:34:04.240 --> 00:34:06.799
Um, what would that one be one thing be?
00:34:07.440 --> 00:34:11.280
If that is the case, I think I would kill the term documentation.
00:34:11.519 --> 00:34:14.400
And I like and I'll I'll I'll tell why actually.
00:34:14.559 --> 00:34:21.440
I think because documentation, especially coming from the world that we come from and now where we are, it sounds like an overhead, right?
00:34:21.599 --> 00:34:22.480
It's an overhead.
00:34:22.719 --> 00:34:25.679
Something that you do after the product is built, right?
00:34:25.920 --> 00:34:31.920
A cost center that is very easily to minimize simply because you have no understanding of it.
00:34:32.079 --> 00:34:43.920
So, for example, if you're a CTO sitting on top of a large budget where your documentation team is such a small teen teeny winy team sitting in the corner, you can easily be a cost center that can be minimized.
00:34:44.000 --> 00:34:46.880
And by the way, we see a lot of that actually in the news today, right?
00:34:46.960 --> 00:34:51.679
And that where we say that team or that team has been shrunk simply because of cost, right?
00:34:52.079 --> 00:34:56.239
And uh, because nobody gets promoted for a great documentation in an organization.
00:34:56.400 --> 00:34:57.440
We know that, right?
00:34:57.840 --> 00:35:13.440
And uh that's why I think what I would like, rather than calling it documentation, that's why I want to kill the term, I think we need to talk about, as you mentioned, about enterprise knowledge or content infrastructure, which becomes so much mission critical in this age of AI, right?
00:35:13.679 --> 00:35:27.280
So I really, I would really like uh this industry to move away from thinking documentation as a point tool to a mission critical content infrastructure for you to have success with your agent text.
00:35:27.599 --> 00:35:29.760
And that's me is super important.
00:35:29.920 --> 00:35:36.400
The moment you change that conversation, you also change conversation with who you sell, your economic buyer changes, right?
00:35:36.559 --> 00:35:41.840
Because now you're not no longer selling a tool to a document, a tech writer, or a documentation manager.
00:35:42.000 --> 00:35:53.519
You're selling to a CIO, CTO, uh CXOs simply because it becomes a mission critical infrastructure for you to have success with all your fancy uh goals you have with AI.
00:35:54.000 --> 00:35:54.400
Well said.
00:35:54.480 --> 00:35:55.199
Yeah, great answer.
00:35:55.280 --> 00:36:19.360
I wasn't sure um exactly what you you'd say with that question and throwing you on the spot there, but that's uh that's a good uh it's a good thing to to reiterate is just uh the emphasis, the high emphasis on documentation and and and the that that role and and um you know valuing it because again, it's so important to be able to structure all the information that flows into these uh important AI tools.
00:36:19.599 --> 00:36:22.880
So um very, yeah, very very interesting insight there.
00:36:22.960 --> 00:36:33.280
And um I I want to continue to um ask you about what's something surprising you learned about your customers or the market that really changed how you think or operate.
00:36:33.760 --> 00:36:49.920
What I've learned over the over over the you know my career, I would say, I think if you really deeply, deeply empathize with customers, right, and try to understand their pains in really true sense, not superficially, sit with them, listen to them, right?
00:36:50.159 --> 00:37:06.000
And and then and then really see how they go about using your product daily and what kind of pains uh that they have, and then come back and then again empathize with your team to sit and solve their problems and those problems in a new way.
00:37:06.159 --> 00:37:08.559
I always say this actually internally as well.
00:37:08.719 --> 00:37:12.480
The best line of code that you write is the line of code that you don't write.
00:37:12.639 --> 00:37:29.360
So what I mean by that is that because most of the time, what I've seen, and which is quite surprising, if you deeply empathize with the customer about the problem or about uh something that they ask you to build, most of the time what I've seen is that actually it's not about building things, right?
00:37:29.599 --> 00:37:34.320
It's mainly about probably you solve the same problem in a very innovative way.
00:37:34.480 --> 00:37:42.639
If you empathize enough with the customer, and then you come back and do the discovery of that problem space in a deep, meaningful way.
00:37:42.800 --> 00:37:46.079
So you solve the problem without even line writing a single line of code.
00:37:46.320 --> 00:38:12.960
And this is quite a surprise because most of the time what we do is that we try to talk to a customer on very superficially, and then you come back and then you try to fix it in all different ways, and then you really got, you know, you really get caught into that trap of uh just keep on thinking uh, you know, a bit very shallow, based on very limited information, because you haven't enough empathize with the customer and their product, right?
00:38:13.199 --> 00:38:14.239
So that's one side.
00:38:14.400 --> 00:38:23.519
Other thing which I generally believe as well is that especially, so I think just for for your for your audience here, I have been uh part of several industries.
00:38:23.679 --> 00:38:28.239
So so this uh content or documentation industry, this is my seventh industry.
00:38:28.400 --> 00:38:39.840
Before that, I have I happened to work in different uh industries across uh um you know across the globe, where what I've seen is that I have seen several shifts, technological shift that has happened, right?
00:38:40.000 --> 00:38:52.800
Whether whether it's the whether is the uh you know the the shift from you know uh wired to wireless, you know, to you know to to e-com, to app economy, to cloud, right?
00:38:53.360 --> 00:38:55.440
And now to actually agent text, right?
00:38:55.519 --> 00:38:57.679
And agent, first AI and now agentecs.
00:38:57.920 --> 00:39:09.280
What I've seen is every time that paradigm shifts happen, your old problems are probably either they don't exist, or if they exist, you solve it in a very different way.
00:39:09.440 --> 00:39:10.880
So I think so.
00:39:10.960 --> 00:39:27.760
It's so important that whenever these shifts happen, and the one is happening right now with AI, I think we need to start thinking about in this new world, you as an organization and your product that you offer today, what how do you can actually amplify your mode, your DNA in this new world?
00:39:28.000 --> 00:39:33.679
And that's one thing that has always kept me on my toes to always think about whenever this shifts happen.
00:39:33.840 --> 00:39:40.159
We need to make sure that we dress our team, our product, our company to be ready in this new paradigm, right?
00:39:40.320 --> 00:39:41.920
And that's what I'm trying to do with Poligo.
00:39:42.239 --> 00:39:44.079
We come from a documentation for humans.
00:39:44.239 --> 00:39:53.760
Now we are actually creating a mission critical content infrastructure that will become the structured truth layer for enterprise AI, simply because that is what is needed, and that is possible.
00:39:53.840 --> 00:39:56.960
It wasn't possible three years ago, and it was also not needed.
00:39:57.119 --> 00:40:06.719
So, and and in this in this phase, even our users, which is tech writers, as as we discussed earlier, they are also grappling with what is their role.
00:40:06.960 --> 00:40:15.679
But if you come as a thought leader and help them, together you can actually create new opportunities, both for the company, but also for your user in this new one.
00:40:16.079 --> 00:40:16.960
I I think that's great.
00:40:17.119 --> 00:40:18.800
I think empathize with your customers.
00:40:18.880 --> 00:40:23.280
You mean you you start to understand their problems, their their perspective.
00:40:23.360 --> 00:40:35.360
Um, and then yeah, like you were saying, you just you're able to create more um solutions, unique sort of answers that you weren't able to come up with otherwise because you didn't have a maybe better understanding, you didn't know where they're coming from.
00:40:35.519 --> 00:40:44.800
But my mind is going to, you mentioned you, you've you've worked in across seven different industries and whether all these different trans uh transformative technologies.
00:40:44.960 --> 00:40:46.159
I think that's really fascinating.
00:40:46.239 --> 00:40:52.639
And I'd love to hear a little bit more about that and also how it impacts your your leadership style over the over the years.
00:40:52.719 --> 00:41:05.119
You know, I don't want to put you too much on the spot or get you know too crazy personal, but um I'm curious if there's maybe anyone or anything in particular that has really inspired you the most with how you lead a company.
00:41:05.519 --> 00:41:11.760
No, I think that's a I think that's a that's a probably a whole uh podcast in itself of going to their ball, right?
00:41:11.840 --> 00:41:19.199
But I can just tell you, I think, I think one thing which I've learned very early in my career is I think the only constant is change, right?
00:41:19.360 --> 00:41:20.639
I think it's super important, right?
00:41:20.800 --> 00:41:22.400
And I think it's becoming even more, right?
00:41:22.480 --> 00:41:24.079
I mean, we look at what's happening.
00:41:24.159 --> 00:41:36.320
I mean, even though I try to keep myself up to speed on what's happening, uh, right, with technology, but the technology is moving so fast every single day now that I really feel that I have a big time FOMO, right?
00:41:36.400 --> 00:41:38.159
Where I feel like I can't follow it, right?
00:41:38.320 --> 00:41:44.960
So I think what embrace the uncertainty, I think is something that that I've learned in the early part of my career.
00:41:45.119 --> 00:41:53.599
And this is something that I've also tried to bring, uh bring in the teams, in the organizations that that that I have pleasure of leading, right?
00:41:53.840 --> 00:42:09.199
Where I think, you know, if you can create an organization that is grounded in growth mindset, right, with the organization that really values curiosity, high agency, and a lot of experimentation involved, right?
00:42:09.360 --> 00:42:35.760
If you manage to do this and if you create an organization, a system in such a way where your teams, your uh your product teams, your engineering teams, your go-to-market teams, if they thrive in experimentation of learning fast, failing fast, trying new stuff, I think that's the only thing because the technology around us is changing so fast that you can't champion a technology because by the time we champion a technology, something else is coming.
00:42:35.840 --> 00:42:47.039
And now, even the technology is becoming uh so democratized by AI that you don't even need to be a developer actually to write uh to create a new new application, new solutions, right?
00:42:47.199 --> 00:43:05.519
So I think really what you need to do is you need to become, I think I've learned this uh from uh one of the interviews that I have heard from uh CEO from Microsoft, Satya Nadela, that he's he actually, when he took up uh as a CEO of Microsoft, he said that we need to move from a culture of know it all to learn it all.
00:43:05.679 --> 00:43:08.320
And that's the growth mindset, what he talks about all the time.
00:43:08.480 --> 00:43:31.519
And actually, it really resonated very well with me in the in the early part of my own career, and which I really try to bring back, bring with me into the new teams, new organization where I where I come to, where I really want people to learn and and be curious and show that growth mindset simply because everything else is changing so fast that what you think that you knew so well in the past, maybe it's not relevant anymore.
00:43:31.760 --> 00:43:32.320
Right.
00:43:32.719 --> 00:43:40.800
And that's sort of what I uh what I really uh what I really got inspired from back in the days change, growth mindset, right?
00:43:41.039 --> 00:43:42.239
Curiosity, right?
00:43:42.480 --> 00:43:54.480
And then and make sure that you create a system, uh you create a team, a you know, a way of working where you people thrive on change because that's the only constant we have now.
00:43:54.880 --> 00:43:55.920
Very, very well said.
00:43:56.000 --> 00:43:59.760
Yeah, uh the only constant is change, you gotta embrace the uh uncertainty.
00:44:00.559 --> 00:44:02.480
And you know, you can't succeed unless you try.
00:44:02.639 --> 00:44:05.519
I think these are all just great things to keep in mind.
00:44:05.679 --> 00:44:15.679
And yeah, I would love to have you back and just kind of continue to pick your brain about, you know, where the current industry is, the current challenges.
00:44:15.920 --> 00:44:24.159
Rahul, thank you so much for just everything that you've shared with us and everything that you've you've talked about in real regards to PaliGo.
00:44:24.239 --> 00:44:34.800
Um for those who are listening and that would like to get in touch with you or your team and just learn more about this structured data approach that you you've mentioned with PaliGo, um, where should they go?
00:44:35.199 --> 00:44:39.599
Yeah, so I think that of course they can go to Polygo.net, which is our homepage.
00:44:39.760 --> 00:44:42.639
They can also look me up um at LinkedIn.
00:44:42.880 --> 00:44:46.159
I actually share, uh I generally believe in sharing.
00:44:46.320 --> 00:44:53.920
So what we whatever we're building, we're building it an open, meaning that I actually write a lot uh a lot of posts on LinkedIn and a lot of articles on LinkedIn.
00:44:54.079 --> 00:44:56.800
So of course you can look me up with my first name, last name.
00:44:56.880 --> 00:45:00.559
And if you uh search it by Poligo, you will definitely be able to find me.
00:45:00.719 --> 00:45:08.000
Um and also, I mean, you know, feel free to reach out to me through any social media, and I'd be more than happy to engage with that dialogue.
00:45:08.320 --> 00:45:08.639
Perfect.
00:45:08.800 --> 00:45:09.519
All right, thank you so much.
00:45:09.599 --> 00:45:12.400
That's Rahul Yadav, CEO of Paligo.
00:45:12.559 --> 00:45:15.199
Uh, thanks again for all the insights and recommendations you shared with us.
00:45:15.360 --> 00:45:16.159
I really appreciate it.
00:45:16.320 --> 00:45:17.679
Uh, thank you for having me, Bill.
00:45:17.920 --> 00:45:18.880
It was really great.
00:45:19.119 --> 00:45:21.280
And thank you all for listening to the SourceForge podcast.
00:45:21.360 --> 00:45:22.559
I'm your host, Bo Hamilton.
00:45:22.639 --> 00:45:26.800
Make sure to subscribe to stay up to date with all of our upcoming B2B software related podcasts.
00:45:26.960 --> 00:45:28.639
I will talk to you in the next one.
