The Real Reason Enterprise AI Projects Stall w/ Larissa Schneider Unframe | S2E12

Full Episode Transcript

This is the full recorded conversation from The Human Protocol Podcast, published by The Human Protocol Group. © 2026 The Human Protocol Group. Please link to and attribute the original episode when referencing this conversation.

THE HUMAN PROTOCOL (00:01.635)
Most enterprise projects don't fail. They really stall. The gap isn't in the models. For what we've seen, it's really in how we're managing these projects. If you can't deploy AI with trust, you really are not deploying it at all. And strategy decks are easy. Operational reality is where AI really earns its trust. Welcome to The Human Protocol, the podcast about staying human.

while building the future with AI. I am your host, Michael Salomon. Today's conversation is for anyone who has seen it over and over again. An AI strategy deck that looks perfect and a pilot that never ships. Because enterprise AI isn't hard because of the technology. It is hard because of the organization, the incentives, risk, governance, politics, and trust. My guest today,

is Larissa Schneider. Thank you very much for being with us today, Larissa.

Larissa (01:04.909)
Thanks for having me on today.

THE HUMAN PROTOCOL (01:07.457)
Awesome. It's a real pleasure to have you with us. And I'm very much looking forward to it for this conversation and to kick things off and warm ups a little bit. And I'm going to ask you something that I ask our customers and our guests every single time, which is tell us a little bit about who Larissa is for us.

Larissa (01:30.391)
Absolutely happy to do that. So Larissa, I am born and raised German, like really countryside German, like most people that were born here.

now I'm the co-founder and COO of a company, called no name. Sorry. Can we cut this?

THE HUMAN PROTOCOL (01:50.363)
We're going to cut it. Don't worry. You can start again. No worries. This is the good thing that we are not live. Okay. Pause the video and we're going to cut.

Larissa (01:54.766)
Okay. This is random. my God.

Larissa (02:04.462)
All right. Okay. So I'm now the co-founder and COO of a company called Unframe. We started the company about two years ago. I founded the company together with Shayan and Adi, who are my co-founders. And we previously met while we were working in the cyber security industry at No Name Security. Spent a couple of years there. My professional background is in...

go-to-market partnerships, marketing, and so on. Before that, I spent about six, seven years at Nutanix, so more on the cloud infrastructure operations side of things. And I kind of got out of German countryside as quickly as I could, spent like 10-ish years abroad between Australia and the UK and France. So really liked the international aspect and then lived in San Francisco where I my

masters for a while, joined a local startup there and they were actually looking for someone to open an office in Berlin and moved me back to my home country. yeah, somehow fast forward 10 years later, still here and kind of build out an office here in Berlin for Unframe Now, a very global footprint, but on the road a lot, always happy to come back here too.

THE HUMAN PROTOCOL (03:26.293)
Awesome. I love the global aspect of your story, right? I think when you have the opportunity in life to visit and to get to experience different cultures, especially on the business side, it adds a different flavor. So you bring a different perspective to the business. And so that's very interesting. Thanks for sharing with us. Appreciate that. Hey, Larissa, let's start with what we're here for, right? Let's get down to business.

Larissa (03:43.397)
you you

THE HUMAN PROTOCOL (03:55.667)
And which is what a lot of people want to know today from your point of view and experience when you remove the hype from AI and what we've seen today, where is AI project styling today? Where is it failing? What have you seen? What can you share with us on that?

Larissa (04:16.394)
Yeah, hype is actually a great word to use because I mean, I'm sure you've opened LinkedIn, Twitter, X, whatever. Every five minutes, it feels like there's a new high product that comes out. mean, vibe coding just came and took over the world and all of that good stuff.

But if you're really putting our like enterprise leadership type head on, coming with a product that looks cool on the internet and saying, Hey, let's find a home for it in our business and see if it can do anything for us. It's usually not the right order of attack. would say, because ultimately what you want to do with technology is not just buy technology for the fun of it, right? You want to solve a business problem that you have. And so I always argue like.

before we kick off anything and even give like a trial or POC to any of our customers, it's like, let's actually talk about what you're trying to achieve here. What is it that you want to move? it like measuring from the top line, from the bottom line, efficiencies, customer response times, replacing manual workflows? Like there's so many different aspects that you could look at, but let's define it let's put it on paper because AI is new to many. It moves extremely fast and

THE HUMAN PROTOCOL (05:10.533)
You

Larissa (05:30.849)
I mean, it's also scary for some people. So you really want to see that you tackle the problem and then you'll find the tech. Because as you said in your intro, it's not like a technology problem. It's not a model problem. There's great technology out there, but having that technology that fits your business processes, your tech stack, your people to use, that is the challenge right now. And that's where...

THE HUMAN PROTOCOL (05:53.701)
Hmm.

Larissa (05:54.443)
enterprises specifically really need to make sure that they find the right partners and the right approach to getting value to the users.

THE HUMAN PROTOCOL (06:02.553)
Yeah, thanks for that. And it seems that you're seeing what we are seeing as well, which is companies are just trying to implement AI without knowing or trying to think about and putting in the hard work, which is what is the business problem I'm trying to solve? So it seems that you're seeing that over and over again as well, right? So you have to spend time. And I'm curious to hear from you, when you get to talk to those customers, do they know what they're trying to solve or you need to help on that?

to find that what they need to do.

Larissa (06:34.514)
Yeah, that's a great question. And as I said, we've been around for two years pretty much on the spot now. And in year one,

It was very much the conversation. We know we need to do something, but can you help us figure out where is a good place to start? You know, because it was like people in their personal lives were starting to play around with chat GPT and like really saw a change in their personal life, but not so much in their business life yet. So they were like, what are others doing? They were tiptoeing around it a little bit. Fast forward to the next year. So 2025, we really started seeing the list of like

top priorities. Most enterprises have a person that keeps some kind of master list of the business requirements, the type of most burning AI use cases that come out. And it's a matter of chasing through that list and getting things out. So that's really the shift. And it's been incredibly fast to see the shift happening in front of our own eyes.

THE HUMAN PROTOCOL (07:36.475)
That's good to hear because all of us, practitioners, right, that are trying to help businesses to implement and deliver AI solutions to the organizations, you are not a magician. But I feel sometimes organizations look at businesses like yours like, hey, magic, use your magic to discover.

what I have in mind, what I'm trying to achieve for my organization. And this is not your role or your company. You can help, but at the end of the day, organizations need to tell you what is it that they're trying to solve first.

Larissa (08:08.936)
Yeah, and they do. I would say they've really come a long way and they do. in some conversations I'm having a few weeks ago, I was talking to someone at one of the top three or top five Wall Street banks. even though they have like tens of thousands of internal developers,

They still have a backlog of close to 2000 AI projects and they will never be able to do everything in-house. They will never be able to like buy individual point solutions for all of it. So it's like, what do they do? And they need to all be achieved by the end of 2026 calendar year. So it's a race against time to bring value to the business.

THE HUMAN PROTOCOL (08:49.071)
I love the racing because that's how we all feel about this AI conversation. It's a race from governance companies. Everybody's racing towards trying to achieve as quick as possible what they need from these projects. Super interesting. And Larissa, what have you seen as it's the biggest gap between AI strategy and AI operations?

Larissa (08:57.994)
Exactly.

THE HUMAN PROTOCOL (09:17.179)
Because we're talking about two different things, right? Strategy is what should drive the business outcome. But then after AI is implemented, that's when it comes to the operations side of things. What can you share with us about what you have seen from both angles?

Larissa (09:35.647)
I would say that the biggest thing that I have seen and still see every day, and it's also shifted a lot, is the decision of how are we going to tackle this problem? Is this a build? Is this a buy? Is this a consultancy route? How do we build? Do we build individual solutions and so on? At the start of 2024,

I still had very regular conversations with people that were like, our strategy is to train and fine tune our own model. And that was always like a bit of like, okay, do you really think you need to do that? And people would ask us a lot, do we build and train our own model and fine tune? And it was like an absolutely hard no, because there's a lot of great foundational model out there.

And they have spent a lot of time and a lot of money perfecting what they're doing. And that's a great partner to work with. We see that a lot less now, like the real medium sized business or even a larger enterprise trying to build their own models. I don't have those conversations a lot anymore. So people have realized that was not the right strategy to take. The next came the, are building our own platform internally.

THE HUMAN PROTOCOL (10:33.531)
Mmm.

Larissa (10:52.032)
which was interesting and it's still an approach for some use cases. But if we just look at the sheer amount of software that needs to be reinvented for the AI era, it's still creating bottlenecks in-house, right? Because even though you have a few developers that you put behind building this internal platform, you still have your business users basically queuing around the block and waiting for you to have time to build something in-house on this platform.

THE HUMAN PROTOCOL (11:19.579)
Hmm.

Larissa (11:20.79)
So you try to standardize a little bit, but it didn't really get to the outcome as quickly as possible. So business users started to not wait and go to the buy route again and find individual point solutions, which then in return created this very, very disassociated tech stack of different point solutions that do not talk to each other. They don't work together. They don't integrate into each other. So it's kind of been.

THE HUMAN PROTOCOL (11:43.799)
It sounds like an old problem, right?

Larissa (11:46.701)
A little bit, yeah. And I think that's really what shaped our business model a lot is to not fall into one of these buckets because it's totally like the pressure is immense. So I really understand that from the CEO down, getting pressure from the board to the CIO, to the individual business unit leaders and so on, like all of them have to deliver for the business because what are we doing against our competitors in terms of AI?

They are running. How are we running faster? How can AI help us? That is conversations that happen every day. And so I understand that companies want to try every single route and there's huge buzz going back to your word, like this buzz everywhere. they try, but ultimately it comes back to the business value. And many of the approaches that have been tried over the last couple of years here are not the ones that have been giving the business value. And so that's where people reshift the priorities and start again.

THE HUMAN PROTOCOL (12:45.211)
I'm listening to what you're saying right now and many things come into my mind, especially in all these conversations that some companies are rushing to cut people because of efficiency gain from AI, right? But I'm hearing you and this is also creating a lot of work. AI is not removing work, but it's also creating work. You need to maintain all this system.

Especially what they're implementing, people need to maintain, people need to train, people need to update. So it's like a whole new ecosystem that is being created because of AI in a way.

Larissa (13:19.725)
It could be, and if we look at the large enterprise again, and if they take this route of we want to use AI to maintain efficiencies and replace maybe human workforce that we cannot sustain in-house anymore, had to make some cuts there, replace it with AI, often the maintenance after the go live is something that is being forgotten about to a certain extent.

Ultimately, as you said, if the build in-house approach is getting very big, the costs are not necessarily going down. It's just being shifted from one team to the other. And then it's a numbers game because depending on who you're trying to replace, in many cases, the very brightest and best AI development talent is definitely not the cheapest head count on the market right now.

THE HUMAN PROTOCOL (14:10.927)
Yeah. It's like we're shifting from one cost center to the other in a way in some organization, because AI will create a totally new cost center in a way because there are expenses. It may not be human. But there is a conversation about the future for leaders that they're going to have to take into consideration the talking, the tokenization costs that involves all these AI solutions that we're not talking a lot lately.

I'm curious to see if you have any thoughts on it and if there isn't any concern or any insights you can share with us about what have you seen about conversations regarding tokens. Because AI works, but behind the scenes, there's a lot of tokens being consumed. And that conversation is still not a lot clear or a lot of it out there. Curious to hear your thoughts on it.

Larissa (15:06.626)
Yeah, as you know, we are in the first inning of AI right now. There's still so much that we don't know how it's going to play out long term. So I don't really want to give a recommendation or thoughts. I purely do not know. But what I can tell you so far,

is that the efficiencies of tokens and costs have really played to our favor over the last couple of years. Models have gone down. There's been a lot of great models and many different models that have come out. because we always, from the start, build a platform that is in favor of the customer, the CISO in terms of security, in terms of the CIO and the CTO's cost centers.

THE HUMAN PROTOCOL (15:37.083)
Mm.

THE HUMAN PROTOCOL (15:45.947)
Mmm.

Larissa (15:50.636)
we actually try to be extremely efficient with our token cost. And that's how we architect it. And we also have a shared interest. Many of our customers run us as SaaS, and we do not pass through the token cost because it keeps us all on our toes to be very efficient. And it's really not been that crazy the way that we architected the platform. If you have lot of like rag workflows implemented yourself and you

passing through a lot of heavy documentations as one-offs many times over, then it's probably going to get expensive. But there's a lot of things that you can do with really smart processing. again, I'm not the technical mastermind behind our platform, but we have some really, really great people that have spent a lot of time really perfecting this. And we've been able to achieve great results for our customers that have not been breaking the bank.

THE HUMAN PROTOCOL (16:30.715)
Mm-hmm.

THE HUMAN PROTOCOL (16:47.759)
That's great to hear that you guys are paying a lot of attention to that. And while I'm listening to you, I hear there were efficiency, of course, the models and everything should also get more efficient and, and, you know, platforms and companies like your that are paying extra attention to that. That's probably where, where it does makes a lot of sense. So I agree. And thanks for sharing that Larissa.

Larissa (17:02.222)
Of course.

THE HUMAN PROTOCOL (17:16.827)
And Larissa, from your experience, right? Working with companies and implementing and delivering these solutions. A lot of people always want to know what's my ROI, right? Let's say I'm a business owner. What would be the best way for a CEO or someone like me working with your company on you to explain to me

How can I get to an ROI? Is it too complicated right now? It's a step-by-step process. Because when you're putting projects, people put a budget for the project, right? I'm curious to see how these implementations are going as far as what is the budget, what is the ROI that you envision for it will depend probably project to project, what the business requirements. So curious to see if you have anything you can share with us on the approach.

Larissa (18:16.355)
For sure. I think you're spot on. You should really look at it project by project or use case by use case basis. And a hundred percent, if you are a business owner, you should not be doing an AI project if you're not expecting to get an ROI. You're not doing this for fun. I mean, it is fun to play around.

THE HUMAN PROTOCOL (18:33.135)
Mm-hmm.

Larissa (18:36.77)
But really what you want to do is get results for your business. I've seen all kinds of measurements in our normal sales cycle when we work with a new customer, we always work on an ROI analysis, like a whole business case together with them so they can also defend it back to their internal teams and their business. But I've seen it from CFOs coming to us and saying, we have to reduce 20 % of headcount in this specific team. Can we see which tasks we can automate?

THE HUMAN PROTOCOL (18:40.227)
Yeah.

Larissa (19:07.057)
or they've come to us with things like, currently we're doing all of our invoicing and our quoting completely manually and it takes our team about two to three days to do so. Is AI capable of helping us to reduce this time because quoting is a numbers game. The more quotes I send out and the faster I respond to a customer, the likely it is for me to win the deal. And so the faster we can be and the more quotes we can send out, it's something that's very easy to measure, right?

There are some AI projects where it's a lot harder to do that. If you have a search use case, for example, and it's a matter of like, I want my employees to find documents that are shared in all kinds of different systems, you can measure that, but it will be significantly harder, I would say. Actually, was a little side anecdote. I was at a conference the last couple of days and someone was telling me, the very large bank, that they measure purely revenue associated to a specific

THE HUMAN PROTOCOL (20:04.894)
Larissa (20:05.613)
project and that they have an extremely integral team there that calculates, they need a nine to 10 X return on every dollar that they spent within five years. And we had a very lively discussion around that because for some, for most, I would say, it's actually really hard to attribute that to a single piece of software. Even if it's great software, does it really translate into direct revenue? So I don't think you need to be as explicit. They probably have

THE HUMAN PROTOCOL (20:19.662)
Hmm.

Larissa (20:33.965)
quite a few full-time people that spend time measuring those returns. But you can start with things like, am I improving my customer response time? Am I able to still keep this job going or the lights on when the person who's in charge is on vacation for a week? Is AI able to automate those bottlenecks? And you can do those smaller things.

THE HUMAN PROTOCOL (20:42.276)
Hmm.

Larissa (20:58.183)
And even if you're a small company and you are the owner of the business and want to start doing something, it can be huge, huge value gain for you. And then the bigger the company gets, the more complex, but I would a hundred percent always measure no questions asked.

THE HUMAN PROTOCOL (21:14.489)
Yeah, it's very interesting. really no one size fits all type of thing, right? Depending on the project that you're saying and what the business wants to measure, it's probably what we're going to get on what they want and how they're going to measure return on investment. So interesting. All these especially probably different countries, different companies, different sizes, different needs, right? So in a way,

Larissa (21:41.426)
Even different departments, right? Like sales team and legal team, very different type of core measurements.

THE HUMAN PROTOCOL (21:43.404)
even different department.

THE HUMAN PROTOCOL (21:48.815)
Perfect that you get there because that was going to be what I want to get next with you. The human aspect of it. We cannot do these implementations without organizations and cultures to embrace this, right? Because for it to be successful, you need to bring people together, right? And understand that. So where have you seen the problem with businesses implementing these and companies

Larissa (21:53.514)
Not even practice.

Mm-hmm.

THE HUMAN PROTOCOL (22:18.469)
kind of lacking in communication or there are people not embracing this. Do you think that we should embrace our people in a way that they all feel part of the mission? So whenever the implementation is done, the actual acceptance of this new feature will provide really the efficiencies that we're looking for.

Larissa (22:45.685)
Yeah, it's definitely important, I think, to have on the one side a top down, but also a bottoms up approach. It needs to somewhere meet in the middle, right? Because when I hear from a lot of companies like our AI strategy is give Microsoft co-pilot licenses to everyone or let them play around with cloud code and so on,

THE HUMAN PROTOCOL (22:54.203)
There you go. Yeah.

Larissa (23:06.625)
That's amazing, right? Like that's what you want. You want them to be naturally curious. And if you don't not give them any licenses or like a chat, GBT enterprise license, something like that. If you don't give them that they will take your company data and do it themselves with their own personal accounts. That's reality. Like let's face it. So give them something they can play around with it. They get comfortable with it. They feel benefit. They're probably very used to working with AI and natural language from their personal lives at this stage. They might be building.

personal efficiency tools or their own to-do list or automate an email send, something like that that is relevant for them and that gets them comfortable. However, I would argue that they will probably not use those tools to vibe code the next SAP or a company-wide ERP system or CRM. The very heavy business focused application, that's something where you probably need a top-down strategy and how are we wanting to get there to make sure that those

core systems that keep your business running across departments, across countries and business units. You need a proper strategy there. And we always believe the best way to do that is to work on the one hand with the IT department, but also with the actual business user, because they need to be fully in line. Having just the IT team bring in a new solution for the legal department,

It is very, very likely. Exactly. Features are off. It's not really fitting into their workflows. It doesn't make sense. It just creates more burden. And that will kind of prevent them from actually trusting the IT team that they can fix anything for them. We've seen this before. At Unframe, what we decided is every single POC that we start, it needs to have both parties in there. Business user really for the functionality, for the workflows and all of it.

THE HUMAN PROTOCOL (24:34.139)
It's a recipe for failure.

Larissa (25:03.32)
but also the IT team to feel, okay, this is a well-governed, well-maintained solution that fits our tech stack, that can be integrated, that has all of the security clearances that we need. Because you also do not want the business user that goes rogue and buys software, and then all of a sudden the IT team is left there to maintain it, to integrate it, and it just does not fit the overall strategy. So keep them both engaged and also...

Don't scare them. In most cases, I don't think AI is going to replace all of their jobs in the short, medium term. Let's debate on the long term. I am even not sure about that. But it definitely gives them more time to do things that have been put on the back burner in the past, because the manual work and rework

becomes less like the manual copying something from this email into this system, downloading it, updating it here, sending it back, copy in, copy out. Like those tedious tasks that are also by the way, very error prone, they are going down. The quality should go up and time to actually think and do something more strategic for the business should definitely increase for the employee.

THE HUMAN PROTOCOL (26:01.711)
Mm-hmm.

Very.

THE HUMAN PROTOCOL (26:16.431)
Totally love every single word that you just said, because I think this is key for any successful implementation. And it's not only AI implementation. Every single project, this is the best practice. You need to have your IT partners and your business partners. There is no way, right? Because they are both adding value to this conversation, and you cannot implement one without the other. You should not, at least that. Let's put it this way, right? And especially with AI, I think the adoption part of it

Larissa (26:19.853)
you

Larissa (26:33.039)
So, thank you

THE HUMAN PROTOCOL (26:45.815)
It's also responsibility of leaders to make sure that they, you know, explain to their people why we are doing this. What is it for them, which is you're probably going to leave behind all the manual work that created human errors that you probably hated. You didn't want to do to begin with, but you did it because someone had to do it. And it gives you an opportunity, all of us to really add value to where the business needs our body.

on thinking, on bringing new ideas, on brainstorming how to grow the business. I think there's a lot of opportunities out there, but I think also part of the adoption problem, it comes from leaders not being clear and communicating clear what are the goals and how do we see our people in this new future with AI. That's what I think organizations and leaders, we all need to do a little bit more there, Larissa, to people understand what is this new future and what is in it for them.

Yeah, I totally agree with what you just said there. Perfect. Yeah. And my next topic that I want to cover with you, it's security. And I did not know that you had a background in cybersecurity, you're probably bringing a lot from that. It's a big topic for me, right? Not only on the personal side, but also at the enterprise scale.

Larissa (27:47.618)
Spot on.

THE HUMAN PROTOCOL (28:14.619)
on the government sides and depending on governments they're doing things differently but

What needs to happen for all of us to pay more attention to security, especially in AI products and systems being designed? I don't know you, but I feel like there's not a lot of conversation on this topic. Everyone is doing their own thing. There is not a lot of regulation, at least here in the US. It's a race, as you know it. Everybody wants to get there and be the first one. So I understand that from the business side, governments and companies.

But is it safe? Can we trust? Where is our data going? When they say this is 10 an hour, is it 100 % not living to the internet? Are they not using our data to train? There are so many questions I have that a lot of people also have questions about it. I'm not sure if you have some information or thoughts on that. It's OK if not.

but just something that we should all be discussing more.

Larissa (29:26.243)
Yeah, for sure. And as you said, like it really depends

country by country. There's obviously a lot of talks about the European AI Act and this and that. Obviously, like GDPR in the California data protection regulation is some of the higher ones. But I think for most enterprises, it really is their own benchmarks and their own internal systems and protocols that count most, more so than just the local regulations.

And that's a really, really important part. We've seen many companies that have formed AI governance boards that have their own AI policy that have very clear idea of which models they can and cannot work with. So when we started our platform and as we already discussed, as three co-founders we met in the cyber security world, that's like Shai's whole background. So we really built the platform from day one with a CISO mindset because

We did not want that to be the blocker of AI adoption. So some of the things that we did is that we're LLM agnostic. Any of the modern LLMs work great with our platform. We do not dictate that. We let the company choose that we work with. Or if they don't have any preference, we give our recommendation. But let's say some of our customers have approved certain models. They are hard blocking some other models.

That's not a problem for us. We can work with the ones that work best for them. And that bypasses a lot of the concerns around uncertainty of where does data go, which models would unframed use. So that's one thing. The other thing is the whole deployment model. For many of the large insurers and health care and financial services that are heavily regulated, the last thing they want to do is just put their data in some kind of cloud that they can't control to a

Larissa (31:21.366)
startup because many of the AI solutions that are on the market right now obviously come from younger startups. And if you are one of the leading health insurance companies working on patient data, for example, that's not necessarily the first company you would put trust in. So the way that we went around this is that we also offer full on-prem deployment. whether the customer has their own cloud environment or even a data center on-premise, we can install there.

not a problem and we do not have to have any access to it whatsoever. It's only by them and they pull us in whenever they need help or troubleshooting or whatnot. And that really gives them a lot of security. There's also a lot of use cases where with the architecture of our platform and our building blocks, we've managed to get around sharing data with any of the models and to kind of share like a metadata layer and so on.

So there's a lot of things that you can do that is of importance for the customer. And you should be doing because until they feel comfortable, you don't want that to be a main concern, especially for some of the use cases. If someone is brand new to AI, I always say, let's start with an internal use case. Don't automate all of your customer communications completely autonomously from day one because

THE HUMAN PROTOCOL (32:27.739)
Yeah.

THE HUMAN PROTOCOL (32:40.091)
Mmm.

Larissa (32:41.248)
You know, you might get concerns and you want to feel comfortable internally before you open it to the world.

THE HUMAN PROTOCOL (32:47.769)
Yeah, and that's good to hear. And I really think it's for the companies to make that decision. Especially depending on the company size and as you said, industry, it's probably safest to be on their own environment and use their own infrastructure so they feel comfortable. Because at the end of the day, they need to feel comfortable, And that's where this part of this conversation goes. But Larissa,

Larissa (33:01.352)
you

THE HUMAN PROTOCOL (33:16.635)
When people say AI, sometimes I feel like they don't even know what they want. When they say AI, what do you think they want? And have you seen people asking about AI, but really what they need is an automation to automate workflows? Because I feel, I don't know if you do feel like that, but on the personal side and even, you

professional science, sometimes people are just talking and saying AI for everything. But sometimes it's not AI what they mean. It's just a simple automation.

Larissa (33:54.299)
Every day, every day. And that's okay. You know, of course we are an AI company and our platform is front and center AI, but I would never tell a customer, I sell you AI.

THE HUMAN PROTOCOL (34:05.488)
Hmm.

Larissa (34:08.429)
What we're selling you and what we're providing to you are solutions. And our solutions have, in most cases have AI components. A big part of it is though that they're made with and around and for AI. There's always an LLM included, but sometimes the LLM and like for many people, natural language is AI. That's like the kind of what we know from, from chat, GBT. Exactly. That, that is for many people that is AI.

THE HUMAN PROTOCOL (34:32.398)
The main definition. Yeah.

Larissa (34:38.48)
Rightfully so, right? That is really what changed our way of interacting with software in recent years. But if you are doing P &L analysis, or if you are working on a slide deck or a report that is being generated, the last thing you want to do is just interact in natural language for everything you do. So sometimes building blocks that can be UI components are just as helpful. So in a given unframed solution, I would say we can see anything from

15 to 20 % AI and the rest being either traditional machine learning, UI aspects, or just very good old fashioned workflows to probably 90 % AI for real natural language interactions and triggers and multi-system type of automations like that can be fully.

AI and somewhere in between. So it's not a one size fit all. ultimately it really depends to, to solve this business pain for the user. And it's not always a hundred percent AI. I would agree.

THE HUMAN PROTOCOL (35:47.131)
Yeah, I was hoping to hear that from you because we all hear from everyone, everyone is using AI for everything. And I think sometimes it's really not what this business needs or what they have in mind. But do you guys also work on designing, developing, and implementing AI agents for organizations?

Larissa (36:08.385)
Yeah, absolutely. So we have three main categories that we operate in. The first one is observability and reporting. So think about BI, analytics, search, data that is spread across multiple desperate systems. The second one would be extraction and abstraction, which is all the work that we do across unstructured data.

So thinking of financial statements and Excel files, PDFs, contracts, emails, obviously in enterprise settings and legacy businesses, there's a lot of that type of unstructured data that sits in a SharePoint folder or some drive and doesn't really get the inside surface as much as it should because it has some of the core business data often hidden in those documents. And then the last one would be automation and agents.

all of the workflows and agents, if you really want to. It's not my favorite word because of kind of the stigmas around it. But all of the automated actions, decision making, so that's the third category. And I love agents, but it needs to provide a value. It's my, honestly, my pet peeve is when someone comes to me and says, I want AI agents. I was like, okay.

THE HUMAN PROTOCOL (37:23.226)
Yeah.

Larissa (37:23.466)
What are they meant to do? What is happening? Is this really just like a one trigger or is this a full blown solution that has maybe a front end that has data visible, that has traditional tables and still also has agentic workflows on top of it?

THE HUMAN PROTOCOL (37:34.043)
Mm.

THE HUMAN PROTOCOL (37:41.243)
Have you received requests for AI agents to do the whole thing without humans in the loop? Or majority of us still have a check for a human somewhere.

Larissa (37:51.351)
In most cases, would say, even though we give the customer the opportunity to choose what they want, in many cases, there is a human in the loop somewhere. I would say probably in more than 80 % of our use cases right now, there's probably a human in the loop. It doesn't mean for everything. You can send certain benchmark. We work a lot with AI accuracy rates.

If we're confident that the accuracy rate of a certain respond is over 95%, 97%, depending on the customer, it would automatically act. If it's any lower than that, we usually have a human in the loop step and we work with the customer individually on what is their level of comfort and what do they have resources to maintain. And then we can act accordingly, but reality in enterprises, most of it still has a human in the loop and honestly it should.

THE HUMAN PROTOCOL (38:50.393)
And how is that accuracy being calculated? Just because of how successful the agent was in completing the task that he's supposed to do?

Larissa (39:01.698)
Yeah, our engineers worked on a very heavily technical system there, a multi-model, lots of different agents recalculating each other's work and verifying, always tagging back to the original source to avoid really hallucinations because what you want is being able to actually verify that the action was performed correctly. it's something that we've become very, very strong in, I would say, but of like a

THE HUMAN PROTOCOL (39:13.339)
I

Larissa (39:30.828)
Sender of excellence and a lot of our enterprise customers just love that because they can really trust the output.

THE HUMAN PROTOCOL (39:36.569)
I love this conversation because you are in the middle of the action, Larisa, in the middle of all this AI agent revolution. Because it's really so interesting where this is heading and it's heading very fast. It's a lot going on, definitely, especially I think this year. We're going to see more AI agents being rolled out in organizations.

Larissa (39:51.15)
There's a lot going on.

THE HUMAN PROTOCOL (40:02.853)
consumption and adoption will be a lot more than we saw last year. And because this is accelerating very fast, it really just going like, wow. So very interesting topic and the whole technical things and how business are approaching. So thanks for that. And Larissa, can you share with us when you're working with businesses and organizations?

Larissa (40:08.334)
Lovely. you

THE HUMAN PROTOCOL (40:31.341)
If you can recommend what would be good question for people to ask you about what they want to implement that majority are not, for example, what are the best questions that you're getting from organization that you say, okay, this is going to be a good project and that others may not. So think for that you're looking for.

Larissa (40:35.15)
So I would say twofold. One is if customers want to know like,

what's a good starting point and working with us. So there's a few questions that we usually ask them to like kind of start them thinking about specific use cases. And those are questions around like, where do you have workflows that are spread across multiple systems? So think about your team have like six tabs open just for a simple task that they need to cross reference and check and copy paste, or where are you relying heavily on humans and unstructured data? So that's

THE HUMAN PROTOCOL (41:06.586)
Mm-hmm.

Larissa (41:29.91)
really requires something that maybe no software, no system is in place right now. Or for example, where are you struggling because there's a lot of human error in a certain aspect of your business or in a certain function right now. So those are questions that we would discuss on calls. Questions that come to us from the IT team, I would say I love when they're really considering how does this fit in with their existing tech stack?

They have a core AI program and they do not want to buy islands of point solutions, even though it is very appealing right now. Everything that you see on the internet is like AI for venture capital back and AI for marketing automation off. You know, it's like there's so many different type of point solutions, but if you're considering a large enterprise with hundreds of backlog AI projects,

It's not realistic to tackle like three POCs for all of them and then get them to production and then also have them communicate with each other. That's not the case right now. So really understanding what is the AI program that the company is running and how do they find partners that fit in, that are not disturbing to internal bills, that are not disturbing to the overall strategy and don't become very

remote islands because they are point solutions that are really thought through as a narrow simple use case.

THE HUMAN PROTOCOL (43:02.905)
Yeah, it sounds like from organization to organization having and knowing what they want to achieve and what they have in house, especially around data really helped to get things, you know, get things running and in a more efficient project way on these projects. Sounds like it. Yeah.

Larissa (43:27.659)
For sure.

THE HUMAN PROTOCOL (43:30.075)
And let's see what I also have for you. And yes, Larissa, I'm going to ask you a few questions. They are part of the same group, right? But each one of them is for a different, it's intended for a different group. Okay. And I'm going to start with the first one, which is two individuals. These are regular human beings building, trying to see what is the new future for them.

Larissa (43:43.31)
Mm-hmm.

THE HUMAN PROTOCOL (43:59.903)
and what is in it for them? What should they, what do you think people should focus on mastering? And these are regular human beings, okay, that may not be tech people or anything like that, just regular human beings. What do you think they should try to learn or master to be successful in the future?

Larissa (44:21.591)
In the context of AI or yeah. Okay. Yeah. I mean, honestly, I would say

THE HUMAN PROTOCOL (44:24.121)
Yes, especially in the context of AI. Yeah.

Larissa (44:30.21)
familiarize yourself with the basic things like the chat GPTs and how to prompt. We're really seeing differences there. This is nothing to do with on-frame specifically, but users that are 10 years younger than us and users that are 10 years older than us. The way that they use simple models and that the chat GPTs of the world are drastically different, I think from a very basic.

prompt training a little bit and playing around with it. You can do so much. I feel like until you get to a certain level of power usage, you probably don't even know the possibilities that are out there. There's always so much more than what meets the eye on the first glance. I think that's extremely important. And to also have this understanding to not just copy paste any output into an email and call it a day, right? Like how can you get the most

the best value out of your, LLM still make it sound human, you know, because nothing worse than just having inter change with the customers and colleagues. And it's just a pure of like copy pasting, chat, GBT emails to each other. It kind of like you're missing, you're lacking the human.

THE HUMAN PROTOCOL (45:31.483)
Mm-hmm.

THE HUMAN PROTOCOL (45:40.569)
Yeah.

THE HUMAN PROTOCOL (45:44.341)
It infuriates me. It infuriates me to be honest with you.

Larissa (45:49.281)
I cannot, I cannot read it. it's very obvious, try LGBT. I'm like, I cannot unpopular opinion. Probably. don't know, but, it's just, I need to feel like I'm talking to a human still.

THE HUMAN PROTOCOL (45:54.139)
Yeah, I know.

THE HUMAN PROTOCOL (46:02.907)
Yeah, it's especially, you probably noticed this on LinkedIn lately. People just copy and paste in what they get to respond to your post. It's just driving me nuts because the main idea of you engaging in other posts is for you to engage. What is your thoughts, right? If you want to have a conversation, but no, they are just because LinkedIn reach somehow.

Larissa (46:29.262)
Totally. I actually just read an article about OpenAI and Tropic, a few other like large tech companies. They have

THE HUMAN PROTOCOL (46:30.687)
It measures how people interact with others. They just copy and paste in like going wild in other cause and when you read that, what is this? Sometimes it doesn't even make sense to the conversation that you're having. It's just crazy sometimes.

Larissa (46:54.861)
think tripled the amount of communications professionals that they're hiring over the last few months here because all of the web, blogs, corporate LinkedIn profiles, everything is just so full of GPT generated postings that they're really looking for the human skilled communication professional to cut through that noise again. it's become one of the highest paid jobs right now in some of the big tech companies.

THE HUMAN PROTOCOL (47:15.079)
Mmm.

Larissa (47:24.457)
right after the AI developer.

THE HUMAN PROTOCOL (47:27.395)
I can only imagine. You posted that on LinkedIn? I'm going to check it out.

Larissa (47:32.495)
I did not post it, but I found it. I'll send it to you after.

THE HUMAN PROTOCOL (47:35.157)
Okay, yes, if you can send it because that's exactly what we're talking right now. It's interesting to see that. And Larissa, for companies thinking on implementing AI, what they should stop doing right now before they reach out to you.

Larissa (47:39.926)
Absolutely.

Larissa (47:56.12)
What they should stop doing, just.

THE HUMAN PROTOCOL (47:58.443)
Right. Thinking or implementing AI, what they should stop doing.

Larissa (48:04.898)
I think they should stop just trying to get tools and then finding a home for them after. it's as long as they have a lot of time at their hand and free time, but think about the opportunity cost of the just tinkering around and trying to a clear rollout plan, a clear strategy that is measurable. You don't have time to wait, so get it done now.

THE HUMAN PROTOCOL (48:32.65)
Good advice. Thank you. To AI leaders and people designing AI, what do you think the focus should be on these companies like Anthropic, OpenAI? What do you like them to focus on primarily if you can choose one topic?

Larissa (48:53.855)
I think no other response than them focusing on AGI makes sense because that's the ultimate goal for all of them, I would assume. Obviously, in the last few weeks alone, we've seen such huge announcements around the Entropic for PE and all of the cloud codes and co-works of the world. There's so much going on right now, but I would argue

THE HUMAN PROTOCOL (49:08.26)
Mm-hmm.

Larissa (49:16.65)
The biggest value are the foundational models and there's a few hype applications in between or front ends and wrappers and so on. But the core of foundational models walking towards AGI and then there's other parties that are working with those models to create the application layer.

THE HUMAN PROTOCOL (49:34.373)
Thank you. To policy makers and governments, what type of guardrails matters for you and what do you like to see? Thinking about on the business side and also on the personal side.

Larissa (49:46.541)
Interesting. That's a really heavy one. And it's really, it depends so much on the country, I would say, and even within countries on the current political climate within those, because I think from one party to the next and one administration to the next, I think those can look drastically different. So I think it's a really tricky one, but

THE HUMAN PROTOCOL (50:09.669)
Things change.

Larissa (50:16.007)
I always struggle with people that do not have real hand experience and I really have tried it out and have talked to business leaders, just making policy. So I would say talk to businesses, see what they need. If you're being too restrictive, you're probably hindering your country or your businesses to be competitive because it will be a huge differentiator of competitiveness going forward.

We always say the Fortune 500 or the Global 2000s of next decade, they are planning their AI advantage today. And if you block too much, you might risk them going elsewhere or just falling behind. That being said, I do think it's important to be close to the development that's going on and see what happens. There needs to be a certain level of protection, but it kind of

can't just be done once and then forgotten about because for that, this business, this type of technology just moves way too fast. You need to constantly reassess those.

THE HUMAN PROTOCOL (51:23.343)
Yeah, it's especially, you know, different governments, as you say, different countries, different parts of the world. And it is a wild west out there and it is a race. Everybody's racing for it.

Larissa (51:35.814)
Totally. When is efficiency enough? Well, you know, there's multiple ways of actually talking about efficiency.

THE HUMAN PROTOCOL (51:37.423)
And last but not least...

This one, Larissa, it's finally, when is efficiency enough?

THE HUMAN PROTOCOL (51:55.471)
it

Larissa (51:58.412)
In the traditional sense of efficiencies, I guess, it's like getting more out of the people. And there's the nice, very corporate political way of saying efficiency is comparing it to layoffs. So I'm not sure which part of the equation we're going to go here. But I think you want to provide efficiency to your people. I think that's an important part. Because you don't want to.

THE HUMAN PROTOCOL (52:09.861)
Mm-hmm.

THE HUMAN PROTOCOL (52:24.815)
Mm-hmm.

Larissa (52:25.953)
hire the best and the brightest and just bury them with a lot of human effort and a human work. Like you really want to get them to use their brains and provide the actual intellectual value that you hired them for rather than just the human active labor, if you will, doing manual work. So I think that that's an important part. If your top priority because of business climate and you had to do the layoff efficiencies,

See if AI can help out because ultimately that probably means that you're trying to do a lot more with a lot less people. And so you will kind of burden the people that are left. I think there's, there's a lot of things that you can do there. I, I'll repeat myself for the fifth time. Start with your objectives, start with your business pain that is most high priority for you. It is not going to be the same with your competitors. It's probably not going to be the same for any other business in the world right now. You can define what's important to you.

And just try it out with AI. There's so many ways now to see if you can get good value from solutions that are out in the market and just try it and hopefully find a partner that doesn't lock you in forever with extremely high cost so that you can experiment around a little bit.

THE HUMAN PROTOCOL (53:38.363)
Yeah. Thank you. Thank you for that. It's something that I have in mind all the time as we see things accelerating and things getting better. AI agents and more AI. I keep thinking, when are we going to get tired of automating things or if we're ever going to be? But, you know, we're all figuring out things, as you said. Companies are also figuring out things. We'll have to go through and try to find out.

But Larissa, thank you very much for being with us today. This has been a very interesting conversation and I learned a few things with you. So I do appreciate your presence with us today.

Larissa (54:07.853)
Glad to hear. Thank you so much for having me on. really cool conversation.

THE HUMAN PROTOCOL (54:23.663)
Yep. And if you are leading AI in a complex organization, share this episode with your CIO, your COO, your security leader, right? Your IT and your business teams for everyone that is responsible for making the pilot real. And follow us, the Human Protocol, because the code may change, but our humanity is the constant. Thank you, everyone. See you all next week. Talk soon.