SPEAKER_01

The friction to adopting technology for a long time now has never been technical. It's been purely human change management.

SPEAKER_03

Where do you see the businesses getting stuck trying to make that move from everybody's kind of got their personal way of doing it to now it's a unified effort?

SPEAKER_01

The place where this is getting complicated is there's a big trust component. How do you let the agent have the right information at the right time? Once you're able to remove that, where the agent is able to get its context on its own, then it is able to also take the right decisions without you crafting it and also tend to make changes to that context in whatever systems need it. And that's scary. So you gotta do it right.

SPEAKER_03

What would be some of the standards, like every company should have these context documents?

SPEAKER_01

To be fair, like I think everyone's trying to figure that out. At least a basic kind of directory of that information because everyone has to be the same using the same brand guidelines if you are the same strategy. And that might evolve over time. So how do you keep it updated too? If you want to move this up, you gotta have a system behind that. What are your sources of truth in your organization? And what is the mechanism by which you can get that information or that context to the right agent at the right place at the right time? Context becomes much more important than the prompt itself.

SPEAKER_02

Mark Boscher is the founder and CEO of Unido, where he helps companies reduce operational friction and turn AI from a personal productivity tool into a true business advantage across teams, systems, and workflows.

SPEAKER_03

Welcome to Using AI at work. I'm your host, Chris Dag. Each week we'll be learning how today's business owners, entrepreneurs, and ambitious professionals are getting more done with smart use of tomorrow's tech. Let's get started. Right now, every business leader is asking the same question. What are we going to do about AI? If this is you, ChiefAIOfficer.com has the answer. We give you a simple path forward where we provide executive and team training so your people know exactly how to safely use generative AI in their day-to-day. We also manage the deployment and implementation to make sure tools actually get adopted and deliver results. And we'll also guide company-wide transformation so AI becomes part of your operating system, not just another shiny object. The companies that act now will increase productivity, cut costs, and grow faster than their competitors. Those that wait will get left behind. So if you want to make AI work in your business, visit ChiefAIOfficer.com and see how we're helping companies of all sizes finally get results from AI. All right, everybody, welcome to another episode of Using AI at work. My name is Chris Daigle, and I'm the host. And today our guest is Mark Boscher. Mark is the founder and CEO of Unito. Is that how I pronounce it, Mark?

SPEAKER_01

Hi, Unito, Unito, Potato Potato. Okay. Up to you.

SPEAKER_03

But what the reason that we've got Mark on here today is because it's a topic that's I guess critical to you know operations in general, but specifically in the age of introducing new uh new ways of doing things, introducing AI into the processes and workflows. Uh Unito uh specialize in reducing operational friction. They actually built the infrastructure for keeping work and people in sync across the systems. So uh today we're gonna talk about what it looks like when it's messy, what it looks like when it's done well, uh, introducing AI into the systems and operations of the team. So, Mark, welcome to the call. And as I I warned you, I was gonna say, um, by the end of this episode, what is it that you want people to walk away with?

SPEAKER_01

You did warn me. Um I think uh I think the the big question that is on top of a lot of people's minds is how do we leverage AI today to go from a personal productivity use case to more of a business use case? Uh I know this is something that uh we're hearing a lot in the echo chambers of AI, but it is the reality in organizations. There's still a lot of there's a big gap from going from using AI locally or just for your use cases to actually rolling this out in a cross-team or cross-organization use case.

SPEAKER_03

So I don't know that every listener has made that distinction.

SPEAKER_01

Yeah.

SPEAKER_03

You know, like in the way that you put it, if somebody told me, oh, I've got a couple people in finance using AI, and I've got, oh, my ops team is using it, and marketing's definitely using it, to me, I would think, oh, like, okay, they're using AI in their company, right? But that could really be identified as as personal productivity because uh if they don't have a a unifying strategy or uh an approach or whatever, then it really is just boils down to individual individuals who are capable and using the tool to do the same job. Is that accurate?

SPEAKER_01

Yeah, I think the one way to put it is like, are you using AI in single player mode or are you using it in multiplayer mode? Um and I stole this from from some other vendor here, but but it's a good metaphor. I I think most people are using it in single player. The most you know, simple example is you're opening your your your Chat GPT or Gemini or Copilot and you're pasting in prompts or attaching documents, you're getting a response, and you're using the output and doing something with it. Nobody ever saw you were using this behind the scene, right? But even now, like with all the the crazier on cloud cod code and all that stuff, you're you're still building very locally, and there's actually a lot the tooling is being built for a lot of local usage, it's running off your machine. Um, and so it's great, it's like a huge unlock for productivity. You could build a lot of stuff, but it's not building agents that are actually then going, interacting, taking a workflow and crossing the boundaries of a role or a department. Um, the the only few ones we've seen are uh support use cases, right? Where a ticket, like an agent's answering tickets on on their own, right? Autonomously, and then pulling other people if they're needed or escalating the tickets. That's one of the rare actual broad deployments we're seeing in organizations of an indigentic use cases. Most of the rest, even the coding use cases, are mostly single player.

SPEAKER_03

So I would imagine that if you talk to any CEO though, they're gonna think that or a leader in the company, they're gonna think their company is doing AI because employees have ChatGPT licenses or whatever, right? So, where do you see the businesses getting stuck when they try to if they if they're listening to this and they go, Oh, I like this multiplayer mode idea, where do you see them getting stuck trying to make that move from everybody's kind of got their personal way of doing it to now it's a unified effort?

SPEAKER_01

Yeah, I I'm happy to get into that, but maybe like us taking a step out like why does it matter to go from single player to multiplayer, right? Because we are, and there's nothing wrong with it, I I would say. Like you are using, yeah, you are gaining product getting productivity gains. Your people are adopting these new ways of thinking, and I think that's that's a really good start. That's where most organizations are at. And it's already a big productivity gain on the personal side and getting rid of a lot of the the dredge work. Uh, but I think the the the the largest multiplier in terms of productivity or impact of the business is when you go that when you you take that step up, when when you're able to build um an genetic workflow that crosses, that is able to act across your organization just like humans are able today, right? So unsiloing the agent. And I think the the place where this is getting complicated is there's a big trust component. How do you manage, how do you let the agent have the right information at the right time? Because the reality is right now, the humans are the ones bridging, giving the context to that AI, right? The the most simple example is copy pasting. Hey, rewrite this for me, or here's my strategy, build a campaign for it, you know, and we're crafting a little context manually, we're attaching stuff, pasting stuff in the prompt, and then we're getting the output and we're putting it somewhere else, right? So we're acting as that middleware, but we're a lot, we're adding a lot of friction here of doing that copy pasting. And so once you're able to remove that, where the agent is able to get its context on its own from the right places and craft it its own, then it is able to also take the right decisions without you crafting it, and also tend to make changes to that context in whatever systems need it. That's scary, so you gotta do it right.

SPEAKER_03

Yeah, I I wanna when you say context in plain English, what do you mean for the CEO or the operator that's listening to this in context uh in relation to AI usage?

SPEAKER_01

Yeah, context is one of the big buzzwords these days. Uh I've heard a lot of like the biggest blocker today is called the context gap. You know, it's like how do you get the right context to your agent? So context is really just information or data. So when you're when you're giving uh attaching your strategy or a brief or your brand guidelines, you know, to your prompt, you're giving it context for it to execute, you know, your campaign in the brand guidelines, for example. Um, but it could also be, hey, I want to do, I want to do a pipeline review. The context is gonna be, well, what are the open opportunities in my pipeline right now? So context is a very broad term to represent any kind of data that makes sense for executing something. If you it's the same thing for a human. Like when you hire someone and you ask them to do something, if they don't have the right context, they're gonna do a shitty job, however smart they are, right? So if you give them too little context, they they can't they don't have enough information. If you give them too much, they get lost. The more senior people, you don't have to give them much because they'll go and find whatever they need, right? So that is the exact same metaphor, I think, with agents. Uh, you got to give them the right information uh or the ways to go and get it themselves.

SPEAKER_03

I I'm with you. The the more context I can feed the model, the more precise the result will be, or the less iteration required to get that. Oh, that's what I was looking for, right? So, what would be some of the standard, like the every company should have these context documents, and then maybe let's talk about a leadership role. What would be some context relevant to the leadership that might be listening?

SPEAKER_01

And and to be fair, like I think everyone's trying to figure that out, right? Uh so it is something that is in motion for a lot of businesses, but that's the biggest uh barrier you see when you try to move to multiplayer because now you have to have the foundations, you have to have that information in a place. Yeah, at least a basic kind of directory of that information because everyone has to be the same using the same brand guidelines if you are with the same strategy, and that might evolve over time. So, how do you keep it updated too? Yeah, so right now it's the humans that are doing that copy pasting. We're the we're the middlewares, right? But how do you, if you want to move this up, you gotta have a system behind that. What uh what are your sources of truth in your organization? And how what is the mechanism by which you can get that information or that context to the right agent at the right place at the right time? And I think there's there's quite a lot of ways to do it, and everyone's trying to figure that out. That this is where the real challenge is. Yeah.

SPEAKER_03

So this is interesting because in 2023, oh, I need a prompt library. In 2026, like a context library, a document of context elements that could be any any player on my team would have access to that same like this is how we define what the company does. That like I okay, I like it. I hadn't really put that into that context context before.

SPEAKER_01

But I could give you an example from like we were just building this week. Um a it's part of like having an agent that is able to help as almost a sales manager, uh, and and coaching reps and playing a lot of the roles in sales manager. And we're we're building a skill uh that is a deal advisor, an opportunity advisor, right? So its whole job is that you could ask it questions on an opportunity, uh, let's say as a rep or as sales manager, you could say, Hey, uh, I'm kind of stuck. What what would be tactics that could go about uh to unlock this deal? Um, or what did I miss? Like I feel like we're stuck. What did I do wrong in the past? Or if I'm a sales manager or a leader and I want to ask, hey, what should I ask this rep about this deal to challenge them, right? Or how could I coach it better? So this is a something that requires a fair amount of context if you want to do it right. Sure. It requires the information about the opportunity itself, like who's the customer, who are the the contacts associated to it, a lot of stuff that is in your CRM usually, and it requires, hey, transcripts of all the conversations that were had with this customer, emails, exchanges, but not just that. Hey, what what is in the LinkedIn profiles of these people? Are they of the contacts on the opportunity? Have they changed roles? What are they talking about? Um, have we had internal conversations? Is there a deal room in Teams or Slack, right? Is there open tickets or features they're waiting for in the development environments, right? So all of that context is really valuable. That the same LLM or AI will give you like an order of magnitude better result or insights if it has the right context.

SPEAKER_00

Yeah.

SPEAKER_01

But the work, all the work is just building that context. It's not running the prompts because the agents already know what a good sales manager is. They've been trained on that stuff. They just don't know what your business is about and what that deal is about, and that's what you got to feed it. And that's a lot of work.

SPEAKER_03

So one of the things that we uh touched on in our our previous call was this idea of static versus dynamic context. Can you kind of rehash for the audience like your thesis on that?

SPEAKER_01

Yeah, so when you're taking um you're doing that deal analysis in the exam that we had, right? Are you gonna take uh you need information from the CRM? When are you getting that information? Is it last week's update? Is it a snapshot from the CRM? Is it live going on the API through some form of MCP and fetching that data? Um, when it's getting our latest sales playbook to know how we sell in the company, is which version is it getting? Right? Is it a file I copy to my computer and attach to the agent? And maybe it's the version from last quarter and there's a new playbook that came out. Like, so first is what is the context it needs, how do I get at the context? The second thing is how do I keep that context up to date? And that's I think for a certain set of data that changes frequently, like systems of record data, you need that context to be dynamic or live, right? Um, and that's again gonna give you a much better performing agent. Because right now we're doing all that copy pasting to feed the context manually, but it's really painful if it changes all the time because it's already stale the moment you kind of snapshot it. So I think that's the notion of dynamic context. It it's still early, but that is that's what you quickly hit, you know, because it otherwise this your agent might be using stale data.

SPEAKER_03

Okay, so mechanically, what does that look like? You've got a uh database that sits in the middle of whatever communication nexus is occurring that you want to make sure is updated and yeah, so I think it depends on what you're running your agents on. Okay.

SPEAKER_01

If you're running purely on a you know ChatGPT or Claude, you're gonna need to have stuff, build stuff to keep that context locally, right? For for your agent. So it might need to build databases and things like that. If you're running your agent off of a software platform, like you're in Salesforce and you're gonna build an agent force uh agent, for example. Yeah, yeah, or a HubSpot or a Sana or any of the software that you might have, you know, NetSuite and Workday, they have already a lot of the context there, not all of it, but some of it. So then that is you get it for free if you're running your agent there. But then you still don't have the full picture, right? So the question you're just you still have the same problem. Um, and honestly, that's where we built Unido for it's really to bring the right context to the right place for the right people. Now, originally it was just for people, it was allowing people to collaborate across software without having to switch all the time. So we basically sync data between the systems of record, between two systems. So you could be uh in a project delivery team and a sales rep close a deal, and all that information is the CRM, it'll stay in sync and be in your project management tool. And then as you update progress on delivery, the CRM will have that information too. So the sales rep or the account manager sees exactly what's happening. And you could collaborate back and forth, one person living in Salesforce and the other person living in a project management tool like an Asana or Reich or Monday. And it feels to them like they're in the same platform, that they're collaborating with someone on the same tool. But behind the scene, a platform like Nito is doing this live to async. What's been really interesting now is that the same problem is replicating, but with agents now, because the agents might be running on some of these platforms, but they also only have access to some of the data. So the same use case that we're solving for suddenly apply for a human working with an agent on the other side. So a ticket gets escalated from a support system like a ServiceNow or Zenesk to an engineering team leaving in Jira. Maybe there's an agent that picks it up on the engineering side to triage it. And then it asks question back and it's going to show up back in service now, right? So we're that that bidirectional sync allows the context to stay in sync across any platform.

SPEAKER_03

Now, that would mean that all of us have this con we we we may not recognize that it's context, it just needs to be extracted and formatted, I guess, for consumption by the models at the appropriate time, depending on the condition of use, I guess. Is that exactly?

SPEAKER_01

Okay, I think context is like we're making the word more complicated than it should be. Yeah, it's the same thing as with a human, right? Like if you want someone to do something, they're gonna say, Hey, you asked me for this. Like, you you asked me to build a campaign. Like, can you give me the context for this thing? Like, you asked me to fix this book. Like, what's the context here? Like, when did it happen? And and in which condition? What was the airlock? Like, you need context to deliver on on anything. Um, and unless you're you're hiring an army of monkeys, like they're gonna need context, right? Yeah. And it's the same thing with your agents.

SPEAKER_03

So we had um, we have a uh chief AI officer certification that we've been doing since 2023. And the faculty that I had that was teaching prompting back then, he was a graduate of Sloan, smart guy, Christian Olstrup, Christian, if you're listening to this, shout out. Um really smart guy when it comes to AI. But his prompting that he taught was basically just dump into Claude or Chat GBT, just like record, just blah blah blah blah blah blah blah. And at the time, I was like, oh, that's novel, right? Because everybody else was teaching, you know, uh a structured prompt. It needs this element and this element, and you need to have markdown here, and you need to blah blah blah. But in 2023, you you did need to give the models those instructions, right? Is that still the case?

SPEAKER_01

So remember it started with like, oh, prompting, prompt engineering, you know, is the new thing. Yes. And I think that came and went pretty fast. Uh yeah, because the the they they introduce a lot of the thinking modes in the LMs that are able to iterate and extrapolate the meaning a lot better instead of having to spell it all out. Um, so I think that change, and I don't know if the the term then became context engineering, because it's all about it's not about the prompt itself, it's like what's the information you give around it. And again, I think the LLMs today, the the core AI models, are really quite smart enough to deliver a ton of value. Like I said, the in the example for the uh deal advisor, like I actually don't need to tell them much beyond, hey, you're a sales manager, you're a world-class sales manager, and your job is to give, you know, answer questions about a deal given this context. The LM already has been trained on what a great sales manager is. I don't need to give it a lot, but I do need to tell them, hey, this is the opportunity, this is the company, these are people involved, this is the history of interactions, this is their feature requests, this is where the stat like I do need to give it all that information so the context becomes much more important than the prompt itself. So I think the weight on the prompt has gone down, and it's all about giving the right context at the right time.

SPEAKER_03

You know, that's interesting because just like organically, I spend most of my like strategic prompting is me hitting dictate and just rambling to dictate, knowing that it's gonna be messy. And then the part that I actually structure is like one line. And the rest of it was just I me giving context by way of turning on dictate and just like verbalizing the issue or problem.

SPEAKER_01

Huh. That's because behind the scene, what they've done, they've they they've basically added a first step when they take your prompt, and there's a there's a prompt internally that says, okay, take in Chris's, you know, verbal message, rewrite this with best practices for prompt engineering, and then and then it takes that behind the scene, right? So it's already applied the skills, the skills and knowledge of writing a good prompt if you want to baked in. So you can give it much more loose stuff, and it'll rewrite it for you, just like a lot of the Claude experiences now are like, hey, I want to do this, and it starts asking you questions as if you had a business analyst or a McKinsey consultant next to you. It's like, okay, that's what you want to achieve. Let's work through how to get there. And they just kind of guide you through it. And eventually they have like the hey, here's the spec. And it's this like very thorough uh eventually spec or prompt, but you iterated through in a much more fluid way.

SPEAKER_03

This makes a lot of sense for sure, and it's uh internally at Chief A. Officer, we have been like I've got people that are just focused on what is the best practice of this, right? Because I know that I can ask for the best prompt in the world, but if I'm if I'm going to generic training data in Chat GPT, I'm not going to get an answer that's bespoke or customized from my situation. It's going to be good advice, but it's not like I'll still need to do some more work on it. I'll still need to wrestle with it as a human. But if I give it that context and ask that question, the likelihood of me getting, you know, like copy paste ship kind of quality stuff is so much higher. And that's that's really the so for the executives who are using this right now to uh write your emails, you know, write me an email or whatever, that's one thing. But if you were to give it the thread or you had context documents like we've talked about here, to where it could tap into, oh, it's this client or it's this specific vendor or it's this whatever, right? The quality of that email that was being generated, even something as basic as that, would be much less synthetic and moving much more towards the authentic, which to me is the ideal you know goal for uh any work is that they can't tell that AI did it, right? Chris is probably.

SPEAKER_01

But I think there's uh there's another unlock that happens when you what what you just described happens, which is the email that gets drafted is good enough, right? And that and so you're not actually reworking it. And then there's what happens, like you're like, okay, send, send, send, and you're you're no longer reworking it. You're still human in the loop in single player mode, but once you you like start trusting it, you're like, hey, this is good every time it's good enough, then you're like, you know what? Just do it automatically, right? Yeah, just send the email. If you have a hey AI agent, if you have a high confidence, this is you know good enough based on my experience, just send it right now. You're like, okay, well, for all these classes of email, vendor management, whatever, I'm it's doing it on its own. Let me just take that agent and make it a business agent to answer for other people in the organization, right? Yes, and and now suddenly that investment in that agent becomes multiplayer. Everyone's gonna give it their little flavor, but then so you that's when you see the business value because you're no longer not just copy pasting, but you're no longer even approving it, and now it'd be it applies to more people. So the multiplier effect of that productivity gain, it just goes way, way up. That's single player to multiplayer.

SPEAKER_03

So for Unito, for you guys, are you are you the platform where they build the agent, or are you the environment where the agents are interacting through at least some structure?

SPEAKER_01

We're the context feed, right? For the agents. So our general approach, and you know, a lot of people have been hey, build all your agents on us, or we're the orchestrator of agents. And I think right now there is a plethora of options for where to build your AI, run your AI, run your agents. Yeah, and I think it's gonna continue that way. And we should, in the spirit of allowing fast adoption, high, quick ROI, let your teams pick the right platform for them. If they already work in Salesforce and they know it very well and all the data is there, let them build agents there. Yeah. Service now, uh, and any of the more horizontal management tools really work well too. The Asanas, the Mondays, et cetera, because they have a lot of context. So let them build there, but then make sure that that platform has access to all the other information in your company that it needs at the right place at the right time. And that's where we come in. We basically set up these back-end integrations, it's no code, it's live syncing, and it means that in Salesforce you'll have everything that's missing. In Asana, you'll have everything that's missing. In uh your Jira or your last environment, you'll have everything that's missing that you might want to have visibility in the sales stuff. Well, now you can bring it in. So now the agents, if you build in there, they'll have full capabilities, right? They'll be able to do that.

SPEAKER_03

They'll be able to access uh an update to the account from Asana or something or the project, tie it into the Salesforce data related to that client and the service now request that came, like they're able to see all that one agent now, almost has like is crossing the blood brain barrier a little bit. Exactly.

SPEAKER_01

But it doesn't need to know because it's all the data is there in the platform, so it's accessing it very fast, always with late as data. Because right now, a lot of people are playing with MCP as a way to give the agents a way to access all these systems. The reality is it's it's very it's very slow, it's very clunky, it's very slow, it consumes a lot of tokens. Yeah, so you're not it puts a lot of friction. Um, like your agent could be instantaneous, but it's taking you know a lot of time and and it's very unpredictable as well. Yeah.

SPEAKER_03

Interesting. So some portion of the agent performance is outside of the agent's control, it's the delays on the MCPs or whatever that 100%.

SPEAKER_01

Um I I think it's just this week, the same I it's a recent example because I have a top of mind that uh deal advisor, the context building part, just gather all the information it needs. It was taking up to an hour with the MCP step. And we went to another approach, we kind of connected things through integration, and it takes seconds and it's always up to date at that point. We bypassed the MCP. And the reality is like 90% of the work was happening just to get the context, and just 10% was to actually get the intelligence out of it. So that's a lot of friction, that's what the context gap is.

SPEAKER_03

It's just an interesting parallel to humans, right? Like if I've got inefficiencies in my processes and I think, oh, I'm going to introduce AI, yes, but if you're introducing inefficient AI, you're like not getting the real magic is being missed.

SPEAKER_01

LLMs and AIs have been trained on human patterns, right? So it's it's actually quite practical uh in a leadership position to just treat a lot of these agents in the same way we we think about um you know staff or individuals. We have to manage them. We have to give them context of like what's the goal of the company here? Can they have some visibility into that when they even they do the smaller things, right? So it simplifies things if we apply a lot of the same management concepts that we're used to, uh, even at the agentic level. And the problems that we went off solving at Unito almost 10 years ago is the same thing. It's the human, you give them tools, but if they have to swivel chair across 10 tabs, 10 things to get the right context, oh yeah, I gotta go there and get it there. Like we all know how inefficient and frustrating that is. Yeah, we all know that's inefficiency. Well, it's gonna be the same for an agent.

SPEAKER_03

Ha. Interesting. So for the listeners that are out there that are like, okay, this makes sense. Better context equals better results from the AI, how do I know if my systems are good context environments or libraries for us to use?

SPEAKER_01

Well, I think the most companies uh it's not. Uh because uh if you think about the last strategy memo or you know, your sales play about gathering dust somewhere on a on a you know SharePoint or Google Drive, exactly. And I think a lot of like humans are good at building context and remembering it, um, but they're still like they they have a limit on the how much they can remember as well. So we do all these all hands and things to refresh, remind people, oh yeah, this is what we're trying to do. We're doing a sales kickoff to refresh the practices and the sales and the playbooks. You don't like you can you can't really do that the same way with an agent. You just have to have uh one source of truth for a lot of its things and give it access to it. So just doing a cleanup of what are the key pieces of context or information in your business, and I think there's a hierarchy to it. It's nothing new, it's like what are the company's strategic goals? Yeah, okay, what is this department's initiatives this quarter or this year? Okay, and then for each of the functions, what's the core playbook for that function? What are the responsibilities? Uh, and making sure that's just that's clean enough. It's you can use it for training your people, you can use it for like the same business practices ever, but now the value of having that documentation, it's a little bit less like we used to do it and kind of forget about it. Now keeping them up to date in a in a clean place for your agents to consume is gonna make them way more efficient. And it's like instead of having to go through a training program for two weeks, you just give them the file, right? Like that's like it, it's like it's ridiculously fast training. You just swap the context. Yeah, a little bit like that, yeah.

SPEAKER_03

Plug it into the back of the head. Um so okay, I've got a lot of systems. Uh I can kind of like use some connectors in my LLM to get some information out of it. I can add a bot to my Slack that is integrated with a little disjointed and very um environment specific, but an integration traditionally has indicated big bills, uh technical um there's gonna be delayed gratification because of all the technical stuff, and it's we're integrating and it's gonna break. Is is what should I expect if I want to do this in the Unito fashion or the ways that you guys suggest?

SPEAKER_01

Yeah, I mean, without going too much into pitch mode for you, Chris, here, but the when we set out to build a company, what we were observing exactly what you were saying is like software technology is super easy to adopt. Uh back in the day it was SaaS, you could just sign up, right? Instant deployment, adoption, no servers, no nothing. But the integration bit was was still like high friction. Like you needed technical skills to do it as a big project, and often more expensive than the software you're trying to integrate. So we that was a design criteria for us is like, how do we make this as easy to set up as it is to spin up you know a new software or sign up for new software? So it's like you have to have integration that is no code or that an agent can set up themselves, and it has to be no technical skills required. So the premise that we went after was we need to do deep two-way integrations, which are fairly complex, but package them in a way that any non-technical users can set up. So really raise the abstraction layer where you're saying, hey, this data here, these opportunities in Salesforce, they I want them in this other system. Here's how I want to represent them in my uh you know reporting tool or in my um work management tool as opportunities. Here's how the data is kind of connect, and then you figure out how to make it happen, right? So it's a much more this and this, let and the system figures out how to do it. Um, we it was really important for us that business users could set these things up and that you could evolve the integrations as your workflows change, because otherwise you don't do it and you get break it broken integration all the time. Um, it is not an easy problem to solve, but I I do think that is a my like that is you have to lower that bar. And there are some platforms, like there's not a lot of players that that do this in a non-technical way, and that is one of our core selling points because IT people don't have time, they have technical resources, but if you ask them for an integration, they're gonna be here's open a ticket and they're gonna be able to do that.

SPEAKER_03

They know it's messy, yeah. Yeah, so this would allow a user to on demand be able to say, Oh, I really I'm working on this issue in the business, but I get some of that info from Drive, I get some, I need to be able to look at the QuickBooks or whatever the financial tools are that we're using, and they will just be able to type that in, and then all of a sudden the agents will on demand kind of create as long as it's an approved source or whatever, the agents can on demand compile my integration.

SPEAKER_01

So the way we work today is you you look at the systems you already have, yeah, and via Unito you can enrich them through two-way sync with data from other systems. Nice so that every system can have a full picture of the truth that it needs. Then your people working on those platforms get a lot more value because they have a lot more context, and any agent you build on it will have a lot more value. And I think we're gonna see a lot of that happening where uh a lot of the plat software platforms that we already have deployed in organizations are all gonna become agentic platforms. They are all becoming agentic platforms, and they will. You're gonna have agents like your HR team is gonna build agents inside of Workday, and your IT team is gonna build agents inside of Service Down, and that's okay. Like, let them build, they're gonna have the best results building their agents that is on the right platform. You just gotta make sure the right data is also in that platform, and that's what we specialize in.

SPEAKER_03

I want to get your thoughts on this, right? So there's a study that I heard about in December. Maybe MIT and the Bureau of Labor Statistics got together and they were looking at how much of these jobs could AI handle today, right? And at the task level, it was about 12 to 14 percent of the tasks. But 12 to 14 percent of tasks of those jobs aren't being handled by AI. It's that human element, like somebody's still gotta plug in the AI, right? Even if the agents can do this, do you see the I don't know even what that concept is, is it human friction or but do you see that being an impediment for businesses? And if so, I'm listening to this, I'm the CEO, I say, hell yeah, let's do this thing. But just because I call you doesn't mean my people are necessarily gonna think in a way where they're like, oh, I could do this. Are you guys encountering that friction at all?

SPEAKER_01

So let me make sure I understand uh rephrase it. So the your question is like, eh, how do we get from where we're at at 14% to the next kind of stage?

SPEAKER_03

Or the question's more like just because it can doesn't mean my comp my people will do it.

SPEAKER_01

The friction to adopting technology for a long time now has never been technical. It's been the purely human change management. Yeah, yeah. Uh I would say since the arrival of SaaS and cloud-based software, it's it's be it's shifted from a technical barrier. Before you needed supervision, the software and servers and all that stuff. So there was technical barriers to even adoption of software. Now that's that's effectively gone, and now we can even build software on the fly with with these LLMs. So there's gonna be even more software. But the barrier has always been and will always be change management. Um, so how do we lower that bar for change management is always is always the what we what we need to aim it for. So the more we give people easier access to it without changing the ways they go about it, the easier it will be. And I think it comes back to what I was saying. Like every software platform is an agentic platform, yeah. Okay, or is becoming one. Yeah. Our humans have developed expertise in the software that you work in all the every day. So the lowest friction, the lowest barred adoption of any of these agentic use cases is to leverage the platforms they already know best. It's not necessarily to go learning you know how to code via Claude. Yeah, yeah. Um, and I think so. I think that's leverage your vendors. Like everyone has written off a lot of the software as the old world. Yeah, yeah. But the reality is they all have massive teams that are all working day and night on on making, you know, adding those AI capabilities inside of their own platforms, and they're gonna be really powerful because they have direct access uh to the data of that domain they're experts in, right? Like Net Suite has direct access to the system of record for for finances, right? And they know they have domain expertise on what finance people need. So in theory, they're in a really good position to build financial agents.

SPEAKER_03

Well, Mark, this is uh this is a topic, the the whole concept of context has been something that I've like you've even got some of our chief AI officers doing research on it because like I want to figure out how does this become an expectation of a client that we're working because we do the services side, like yeah, yeah. I first I'd love to educate them on the importance of it because I know that when I give more context to even just just the models to work on something, I get to the aha a lot faster, right? But I also this this is is is it an easy concept to explain to a business, like the importance of it?

SPEAKER_01

I I mean I don't think it's a hard concept to explain, but it it's a hard concept to explain the importance of operationalize or operationalize okay. So what does it mean, right? Yeah, so I think you guys should consider things like if you come in to see a customer or a cus uh someone's listening to this and like how do I get started? Just start by surveying what you think are are the pieces of context that you need in your organization. It's all the classic stuff, it's just that they're hidden away, or they're all in the shelves here and there. So it's again, it's the strategy, it's your sales playbook, it's your brand guideline, it's your it's your voice, it's your market positioning, uh, it's your CRM. It's nothing exotic, it's just that they're all sallowed off in our business, and so the eye doesn't have easy access to them. Some of them you'll want to do deep integrations, maybe using something like you need others, it's just hey, dusting it off, maybe taking that that word file and you know making sure it's at a place where the agents can really access it more easily. Uh, or it's like, hey, here's the gold standard for this for our brand voice. Let's make sure everyone knows and knows where to get it. But there's really low-hang fruit there, because what every time you give it this one nugget of of like this the gold print of your your objectives or your playbook or whatnot, it's like so much stuff you don't have to tell the eye on your prompt. It's like it's it's like instant, it's your your neo, you know, let me learn how to play, uh how to fly in a helicopter instant download, right? So it's not that mysterious. It's all the standard practice of every business that we kind of forget that we've done or we've let kind of gather dust on the shelves on the other side. And and we don't need to, you don't need to change everything and rip and replace or introduce a whole new platform. You already have it all. It's already used by your whole organization in different pockets. You just have to unsilo them off, right? And I think that's the it's really exciting times because it's it's just common sense um that is forcing everyone to just be more can't be more transparent in the organization, right? To to to be clear about our intents, to be clear about how we do things, because the ROI has just gone up of doing it, right? It's yeah, it's no longer just locked in people's minds and experience.

SPEAKER_03

So for the listener who's got a bunch of chat chat GPT business licenses, or maybe you went and got enterprise licenses or whatever, you know this that if you enter a prompt in it, somebody sitting right next to you enters a prompt in it, the answers aren't gonna be exactly the same. And if you've got people using AI in external communication or investor relationship, whatever those things are where they're communicating with people, and they don't have that, it's gonna be a little different every single time. But if you have the context documents that are part of that process, the likelihood of uniformity with the output as it relates to what the context documents were increases significantly, which means you, as the business owner or operator, don't have to be worried as much about man, what are my people sending? Like what is going out in those AI emails?

SPEAKER_01

I think you put your finger on it because for AI to be adopted at scale, from solo hidden behind the scene to multiplayer, let's say, you have to have trust, right? And how are you gonna get the trust? Well, you you need a reliability, you need to know that this is not gonna mess it up. And uh the there's a direct, like the better the context, the higher you're gonna get trust. Yes, it's just gonna it's just it's a direct, direct correlation, and it's exactly the same thing. I think working with human, if you're like, hey, you write the press release um for this, you're and then next week you ask them again, you're not gonna get the same exact same thing. Yeah, but that person, because they know the brand, they know that they've done it before, they have the same context, so it'll still be within the boundaries of what it should like what it's correct and won't mess anything up. So it's just you're just trying to reproduce this concept of the right context, there will still be variability, but it's a good variability, right? And I think this is something I've seen a lot of people struggle. I'm curious if you counted the same thing. You know, LLMs, the non-deterministic nature, and it's almost a mouthful to say, versus the deterministic nature of code or systems of record or workflow software, uh, like a deal pipeline, you know, it's always gonna work the same way, it's the same stages. Software is deterministic by nature. LLMs are non-deterministic by nature, and you have to know when to pick one or the other to get the best results. And right now, if it comes to the examples we're giving, if you're using a non-deterministic LLM to build the context, you're gonna get variability in the context. Do you want variability in your the information about your deal? It's the same deal, it's the same information, it should always be the same. So deterministic approaches for context, non-deterministic approaches or LLMs for interpretation, for guidance, for creativity, for next steps. And if you're if you're able to use the right approach for each, you're gonna get much, much better results, much more reliably, and that's gonna lead to trust and to using this in multiplayer mode.

SPEAKER_03

And to add on to that, if I was a business owner, I I had a mentor once that told me he wants to remove discretion from the process for employees. Yeah, like they don't need to decide, they have a way of doing things. If I'm a business owner and I'm giving My folks, which IGBT, maybe we're getting them training and that sort of thing, but there's still some wild stuff. Like, who knows what what AI AI could glitch, they could be doing it, whatever. But if I have if I can remove discretion by introducing determinism as compared to probabilism, right? Then as a business owner, I feel a lot safer with my people using it and will introduce more and more trust as they earn it, kind of thing, right? And having that context document.

SPEAKER_01

Yep. I love it. Exactly. It removes because if you if you think about it, that means you if you have a good con you don't need to put as much in the prompt. If you don't mean to put as much in the prompt, that means there's a lot less chances of getting you know wild results. Yes. It's like shifting a lot of the stuff is preset, right? It's training wheels almost. It's uh yeah. I think it's a form of training, but it's the same thing when you're training people or they gain experience, right? It's just packaged. Yeah. Um, it requires a lot less cycles for these agents to do. So I think what you're talking about, removing discretion is really interesting. I think any role that you where your boss would want you not to have any discretion, AI can do that's a role away. That's gonna be a role that's gonna disappear pretty quickly. Um but judgment still has a lot of value. I think these LMs can have good judgment as well. Yeah, uh, but judgment is where like this is where you want typically humans to be. Yeah. Uh so I think roles that you know have discretion uh are the most valuable ones, yeah.

SPEAKER_03

So if you're listening to this, that's a very important nugget to take away from this. Make sure that you're somebody where they still need, well, what does Chris think? Right? Not like, oh, just go push a button, go use a GPT, go whatever, right? That you don't need Chris anymore.

SPEAKER_01

I mean, I'm not a I'm a I'm an entrepreneur, so like I'm on the optimistic side of the curve. So like I I do think like roles will change and evolve, but I do think it's gonna be much it's gonna be a new breed of roles that are really interesting too. And it's gonna the demand is gonna increase. I I'm not on the uh, you know, all these jobs will be lost. I I just think they will change. Um and it it's more of an embrace it than try to resist it kind of thing.

SPEAKER_03

Being an entrepreneur as well, in the startup world as well, I've kind of ready for universal high income. Yeah, startup world stuff, man. You know that. Uh, we're working hard over here, people. Um well, awesome. So so, Mark, for those who want to maybe dig in a little bit more about this, they think that their environment's ready for that their team is ready and that they understand the concept well enough to get serious about this introducing determinism into their process through the this context play. What what are their next steps for learning more about Unito or what you're the way you're thinking or approaching this?

SPEAKER_01

Yeah, I mean, uh Unito.io, have a quick look. Uh we support quite a lot of uh the core platforms and software. If there's any that you are using and and and want to invest on the Ionit and it's in our catalog, just you know, ping us through the chat or uh ask uh for a meeting and and we'll give you uh a demo of the platform and we'll guide you along the way. Um our approach is really show the best, show teach how to fish, show the right approach, but then give the tools for people to set it up themselves.

SPEAKER_03

Nice. Are you do you have time to share your thoughts and discoveries on this stuff, like on LinkedIn or X or anything?

SPEAKER_01

Yeah, so I do uh quite a lot of posts on LinkedIn. Uh I'm trying to stay uh away from the the uh the ex zoo right now. Uh but yeah, do follow me, uh Mark Boshe uh on LinkedIn. I'm posting like three times a week on exactly those topics. Uh would love to hear from uh from comments from the show.

SPEAKER_03

Thank you so much for your time, Mark. I um this is a really uh uh salient topic for me because, like I said, I've got our chief AI officers doing research on what is the zeitgeist around context. And this, like, I had no idea I was gonna get so much uh from this conversation. And um, I love the chance to talk to people who are like just plugged in on AI, right? And I can tell that you are plugged in.

SPEAKER_01

There's it's uh it's a big fire hose. We're trying to serve we're trying to keep up, and honestly, I don't know if anyone can keep up, but uh we are looking at specific parts and uh happy to share it. Thanks for having me over, uh Chris. Appreciate it.

SPEAKER_03

Right, and thanks everybody. So um obviously we'll have another episode coming out uh every Monday. We're getting close to episode 100. When that drops, it'll be a big deal for us. Yeah, it's pretty exciting.

SPEAKER_01

Congratulations, uh Castillion. Not a lot of people make it to Undrip.

SPEAKER_03

So yeah, congrats, big it just you know, step by step, I guess we got here. Um, but if you're listening to this and you're getting a lot of value out of it, one of the ways that you could help make sure that others are getting this message is to go and leave a review on whatever your podcast platform is of choice. Forward this to a buddy or a peer. If you have questions, we'd love to hear from you. And again, thank you so much for uh investing your time in the Using AI at Work Podcast. Thanks for tuning in to Using AI at Work. Don't forget to subscribe for more conversations about how to use AI at work. And a special thank you to our sponsor, Chief AI Officer, for empowering businesses with AI education and training. Visit their website for a free AI readiness assessment and AI strategy guide to help you get started using AI at work. That's www.chiefaiofficer.com. Follow us on Twitter at the handle UsingAI at work and visit www.usingai at work.com for free resources to help you harness AI in your role.