Access to LLMs, like ChatGPT, make many believe that they are a complete automation solution. This belief, however, is misguided. Using these tools without full automation behind them keeps workflows slow and manual. Krista Chief Geek John Michelsen cautions, “If you’re still pasting documents and pulling answers manually, you are doing the work. Automation should be doing it instead.” Relying on LLMs alone may look productive, but it’s an illusion that holds back real operational efficiency.
“If you’re still pasting documents and pulling answers manually, you are doing the work. Automation should be doing it instead.”
Access to LLMs, like ChatGPT, make many believe that they are a complete automation solution. This belief, however, is misguided. Using these tools without full automation behind them keeps workflows slow and manual. Krista Chief Geek John Michelsen cautions, “If you’re still pasting documents and pulling answers manually, you are doing the work. Automation should be doing it instead.” Relying on LLMs alone may look productive, but it’s an illusion that holds back real operational efficiency.
Manual AI Interactions Are Inefficient
Current workflows involving LLMs, where users manually upload documents, formulate questions, and extract answers, miss the mark on true automation. Michelsen illustrates this with an example: “An executive asked a question about paternity leave in ChatGPT, uploading their company’s HR policies for context. They got an answer, but it wasn’t complete because ChatGPT didn’t know if the person asking was an employee, a contractor, or when they joined.” Without specific context, LLMs cannot deliver actionable responses that integrate with organizational processes, leaving employees to interpret and fill in gaps themselves or follow the response to contact the relevant HR representative.
The Real Impact of Slow Workflows
Using LLMs in a piecemeal fashion can boost individual productivity, but it doesn’t drive organizational transformation. Michelsen points out, “Most outcomes sit in a queue for the majority of their time. Getting to answers faster as an individual isn’t transformative if those answers still have to wait in line for action.” Companies that fail to adopt end-to-end automation remain stuck at human speed while competitors advance at machine speed.
Automating for True Business Outcomes
For business to transform, AI must do more than answer questions—it must drive outcomes. “It’s not enough to answer, ‘Am I eligible for paternity leave?’” Michelsen explains. He goes on further to say, “The real question is, Can you record my need to take paternity leave, process the paperwork, and trigger follow-up actions?” Platforms like Krista enable this by integrating LLMs into workflows, automatically updating relevant systems without human intervention.
This orchestration is key to reducing time spent on manual tasks and ensuring the process flows seamlessly from start to finish. As Michelsen puts it, “You can increase engagement with conversational AI, but if you’re not orchestrating each step to accelerate the outcome, you’re not truly moving faster.”
The Cost of Stagnation
Failing to automate beyond isolated tasks comes at a steep price. Slower decision-making, missed opportunities, and higher operational costs are the consequences of a half-hearted approach to automation. Michelsen warns, “Those that don’t solve for complete process automation are at a disadvantage. A leader’s job is to integrate and accelerate business outcomes, not just individual tasks.”
In a competitive landscape, those who fully adopt automation will capture opportunities faster, minimize human intervention, and scale more efficiently. The future belongs to companies that understand this distinction and act on it now.
Moving to Machine Speed with Full Automation
LLM limitations become starkly apparent when viewed in the broader context of enterprise automation. Michelsen asserts, “The capabilities of AI need to be woven together with existing systems and human expertise to drive meaningful outcomes.” Platforms like Krista allow businesses to operate at machine speed, leveraging AI as part of a cohesive, automated workflow rather than a manual, ad-hoc assistant.
Michelsen sums it up aptly:
“The more AI capabilities you see, the greater the need for orchestration. Without it, you’ll find yourself doing the work manually while the competition accelerates at machine speed.”
Links and Resources
Speakers
Transcription
Scott King
Hey everyone, thanks for joining this episode of the Union Podcast. I’m Scott King and I’m joined by John Michelson today. Hey John.
John Michelsen
Hi Scott.
Scott King
How are you doing? It’s bright and sunny back behind you in Dallas, glad you’re home.
John Michelsen
It is, yeah, thanks. We’ve got some storms lately, but clear for now.
Scott King
All right, perfect. We’re going to talk about kind of a polar opposite belief when we talk to people about Krista, and specifically LLMs, right? Half of the people have tried an LLM, they’ve tried some type of automation, and they can’t get it to work, and they understand the limitations. The other half of which you’ve had more of these conversations than I have is when people say, well, I can do all of these things in an LLM. I don’t understand why I need something to help me, we call it orchestration. I don’t understand because I can ask ChatGPT questions, or Claude can write code or Gemini can help me search the web, and things that. What’s the genesis of this polarization, and who are these people?
Give us an idea of which role it is because I think that’s important to understand someone that’s using technology versus implementing technology.
John Michelsen
Yeah, Scott, this is really kind of a, it’s almost a joke now internally. It’s, okay, when we meet someone and we introduce them to Krista, we’re either gonna get the, “I totally know that’s exactly how I can now operationalize all this great stuff that’s going on in the world” or the “Why do I want to do that when I could just use the other stuff directly myself? I can do it.” We can really, from an audience perspective, see it across the board. There are pretty much, it just has, I think, a lot to do with whether you’ve really gotten into the weeds on how would I really put together something that is transformational and outcome-oriented with these great capabilities that we’re seeing. And if you’ve thought through that sufficiently, you will very quickly realize the whole premise upon which Krista was built. And if you haven’t yet,
and you’re getting to know Krista, you’re about to. We’ve had a couple of unfortunately awkward moments, but funny stories is the way they end up with folks who are about to have that aha. And if it’s all right with you, I’ll share one. This is kind of the classic one I’ve told you before.
Scott King
Yeah, I’d to hear any specific examples without disclosing probably who gave you the pushback. Because I personally use several of the LLMs to do little tasks, but the amount of work I’m doing in the background to make it work is incredible.
John Michelsen
Yeah, it’s not fair. Exactly. Of course, and that’s not to take away from the fact that it is capable of doing something that heretofore either was much harder or sometimes not even really technically feasible. With that, our personal productivity can improve. But we really don’t affect, and this is, potentially provocative, but we really don’t affect an organization’s transformation with an incremental productivity gain on behalf of a person doing a task. We really don’t because most outcomes sit in queue for the vast majority of their time already. If we’re trying to accelerate a business and we’re trying to accelerate the pace of technology adoption and exploitation and we’re trying to build new things with great new capacity that’s given to us by AI, one of the thousands of tasks you do being more efficient with, name, AI technology, whatever related, getting to work faster is certainly a productivity boost, but your organization didn’t transform if it’s now faster for you to get to the office. That’s just not going to do it, right? I’m just saying it’s a general statement, that’s the case. Now when it comes to these LLMs and some of the euphoria around what they either can or maybe soon can do,it gets a awkward. We’re giving a presentation to an executive team of a company that’s now a customer, and we’re thrilled to have them as a customer. And one of the senior executives, in fact, in the technology side of the business is, of course, engaged, listening, asking a couple questions, but then goes quiet for a few minutes and then says, “Look, guys, I got to admit, I’m just not, I’m not trying to be ugly, but I just don’t get it. I really don’t get it.”
Back up and give some context. We’re talking about, at this particular moment, one of the many use cases. Krista goes enterprise-wide and does use cases around pretty much any part of the business, and we’ve done pretty much every part of the business, including parts of her business I never even thought existed. And in this particular moment, our sales rep who has to have an HCM background, a human capital management background, goes into a dialogue around enabling a higher level of employee self-service and a better employee experience at a lower cost and a blah, blah, by using Krista to do a set of things the classic employee handbook and managing a lot of those tasks and outcomes around it.
And at one point he says, “Look, may I just share my screen real quick and kind of explain to you why it is that I’m just not sure we need to go forward with this. It just doesn’t make sense. While you were talking about how you guys can manage all that content and you guys can make sure that the right answer gets produced and you can do this, look, I just went into OpenAI to ChatGPT. I uploaded a PDF of our HR policies. I asked the same question you asked, am I eligible for paternity leave and sure enough, look, I got this great answer. Look, three paragraphs worth of answers. What on earth do I need that I don’t already have from Krista?” And this is where it gets awkward, right? Because the fact is, I had to point out, this is when, by the way, our sales reps get really quiet because they know, and they have been walked through countless examples of this type of situation. But for some reason, they just aren’t quick to the mic at that particular moment. It ends up being who says, “Well, mister, if you look at that answer, and I can’t read it from your screen because you’re just showing me a big, you know, I don’t have to at it. It’s not an answer. It’s a whole bunch of questions because it has no context of who you are. Are you even an employee of the company? Are you a contractor? Did you join this week?
The answer to do you have, and how can I know this without even reading your answer for that opening I gave you? Because it doesn’t have the context of who you are. How could it possibly know the answer? It can know your policies. Know being a word we’ll put in italics if you don’t mind. It can have the factual body of content for your policies. But without context of who you are, it is not going to accomplish that goal. And all you’re going to get is, well, and I’m not kidding, I’m trying to read the screen while I’m saying this and I see at the bottom where what I call is lawyering up with all due respect to lawyers. These responses, they know they don’t have the answer, they refer you to the people that trying to get an ROI by not having to use much. They go and say, “Be sure to talk to your HR representative and get a complete understanding of the paternity policies that would be applicable to you.” Essentially, almost word for word.
That was the killer, right? That was when the aha happened for that particular person. But, Scott, that’s not even half of it. I am not curious about a paternity leave policy. I don’t care what our paternity leave policy is unless I need to take paternity leave, right? The biggest irony is that’s not even the half of the difference between what a Krista is capable of doing versus what just an LLM alone is able to do. It’s not just context. Ididn’t ask the question about paternity leave out of some random curiosity because I just wonder and I’ve got that kind of spare time on my hands. I probably just learned that I’m going to be a daddy and we have to figure this out, right? I’m trying to get something to happen. I need an outcome, which is I need to be processed out of the organization for a certain period of time to help take care of our brand new baby. Just giving me information is not addressing. That’s not what the question was, right? The question wasn’t, “Hey, I’m curious. Under what conditions do people get paternity leave paid versus unpaid, is it just you don’t get paid?” That wasn’t the question. The question was, “Having a baby, don’t know how I’m gonna handle this, I’m gonna need some time off, and I’m not really sure how that all works, but can you get me there?” That was the question, was it not?
When someone who hasn’t really thought through, how do I get context to a capability an LLM? How do I get integration into the systems to know what are the particulars of that particular person’s employee status, location, a variety of things that drive what the answer to that could be or not? And then even more important, orchestrate the outcome of producing what the question was, which is, “Can you help me record my need to take paternity leave? I’m going to need to be away from the office for a period of time, and I need that to be recorded. I need that process followed.” And of course, who knows what that process is. I exhausted this topic. I hope I didn’t exhaust you guys in the process. But this is a super important thought.
The capabilities that we get in technology are the very reason we built Krista. When you look at the pace of innovation in technology, we love it. We want more of it. We’re here to make it possible for you to take advantage of it. Because if it takes a person going to a screen and typing something in or uploading something every time you want to take advantage of a piece of tech, you’re never going to hit the velocity that you need. And you’re never going to get to outcomes. You’re only going to get to yet another set of things that people now have to know to do, be aware of and be trained in order to do, in order to even take advantage of it. That’s not the world we’re looking for. We’re looking for the world that is, there’s a great new capability, an AI that is able to, you name it for your business. Plug it in, it’s available within minutes. No one had to be retrained. No one even necessarily knows the name of it. Yet we all get the benefit of it. That’s where we go when you leverage Krista. For those who are less familiar with Krista, we sort of jumped into the middle, didn’t we? The magic that we think we’ve produced with Krista is that it orchestrates human capabilities, system capabilities, and existing AI capabilities, and it creates the AI that your business needs uniquely. It is unique to how your organization functions and decides things.
It can’t be bought and it can’t be borrowed from your competitor. What we do with that is create an outcome out of that interest in taking paternity leave, not just resolving the curiosity of here are the set of policies under certain conditions and then go talk to your HR rep. That’s an example. And of course we do this across the whole organization, that’s where we think we’ve done some pretty cool work.
Scott King
John, think it’s where people have set their own expectations of this happening. Because maybe they are looking for a personal productivity tool or most of the AI point solutions are sold that way. I do this one, I help you do this one thing, Whereas orchestrating a business process, quite possibly is just completely outside of their expectations. With the paternity question, it’s a good example to show both processes, right? Because there is a follow on chain of events for a paternity question, but they may not realize that. But maybe, employee onboarding say a new role starts, okay, they need access to these systems. They need, the payroll set up and, the tax and the I-9 or whatever you got to fill out, right? Maybe they’re just not thinking that way because of the market kind of just provides these myopic views. Do you think that’s it?
John Michelsen
Yes, it is absolutely. It is absolutely because in fact, have a we’re in a market and I’ll try to avoid too much time on this big soapbox. Kick it off from under me if you need to, Scott. The entire tech industry is oriented around building the smallest. We call it MVP, right? The smallest amount of functionality that can get someone to start being willing to pay for it. Expand it to a point where some large company wants to come and buy it.
That is the motion that I as a founder have been pressed into for my entire career and I’ve resisted it to the best of what I could for my entire career. And here, I had the opportunity to say, go away that guidance and we built a massive platform and we took years to do it.
But the pressure on anyone is anyone outside of us, even within a huge company, they’re only going to get funded for that, “give me an MVP that you can go hook a customer with,” then this kind of thing. And by the way, that automatically breeds this explosion of point solutions.
By the way, there’s a capability there. We want it. We need to orchestrate it with software, not people, because it’s an overwhelming number of point solutions that we end up with because of that motion. The second problem is that they measure their success and their stickiness based on how many people they land on a desktop in front of. If they feel they’re not enfranchised with their customer, if they aren’t, my logo is in the upper left-hand corner of all those people, I know my renewal is going to work. And I totally get that, businesses want to be in business and they want to be in business next year. They want to make sure that they promote their relevancy to their customers. But in doing that motion, right? And right now, AI companies are doing this. AI is a function. It’s better consumed by a platform Krista.
In many cases, now there’s personal productivity aspects, right? If I could get a model to generate an image while I’m constructing a presentation, that’s awesome, right? And I would directly interact with that AI to do it. The transformation, the world isn’t going to be all that different. The organizations financials aren’t going to be all that different if my image is the same old ugly stuff I usually come up with, or if it’s a really cool one that an AI created, not going to do that. But the capabilities that same AI has woven together with your existing systems, your tremendous human capabilities and the orchestration of all of those things, and of course the bespoke, unique to you AI that must be created in order to fill in those gaps, that’s where the financials change. Software companies and AI companies and a lot of AI point solution companies understandably are trying to get in front of you and make sure you see their logo and you know that we’re the one and that way our stock price is good and our renewals are safer and all that but what you really need is an invisible fabric that’s running all this stuff because, my analogy always goes back to Star Trek. How many times did Captain Kirk or Picard have to go pull off a panel and go dorking around inside of a whole bunch of things trying to figure out what icon did we press for that or what is what what was that system name again? Right. Or or where do I have to go? What do I need to install to to be able to ask for it just never was even in the lexicon. That’s the mission for us is to create that. And back to our premise of this particular podcast was the more AI capabilities you see, the greater the need for a Krista, which is what I described, right? If I use Krista as an adjective, you need that approach. You will not capitalize on it the way you really need to without that approach. I think that’s defensible regardless of whether you were even potentially working with us or not. That is what we have to do, right? And that’s why we’re having these bipolar conversations where we’ll introduce someone to Krista and they go, “That’s exactly it.” Because, it’s the herding cats of my people that is my greatest challenge. I’ll give you one more really quick example of that story, of the story of that, and then we can make sure if you have any thoughts to wrap it up. But, we’re talking about a company that needs to better engage with recruits to get them on board fast. Because it’s a really challenging labor market, and these are entry-level type positions where there’s not a whole lot of brand loyalty. It’s not they’re out, they’re not using recruiters, they don’t have a resume, they’re replying to text messages. And there’s lots of point solutions that are trying to help customers get those employees engaged and to the point where they can get them an offer. But there’s so much orchestration that needs to happen and so much human workflow that needs to happen in order to make that really possible. You lose the real value and outcome of onboarding new employees if you only think about it as, I’m going to use some computational AI to get that engagement level up.
But I’m still going to take three days because people aren’t getting the workflow of processing the new employee into a qualify, then do the background checks, then get the offer letter out, deal with any other circumstances, maybe unique to that individual that need to be addressed, and then get that signature back.
That is a heavy duty human workflow and integration of systems orchestration kind of thing. And yes, AI can do some really great work to take out some of those human steps where it’s clearly pattern matching on what the behavior of the organization already does, right? That’s the unique or bespoke stuff. All of that is what it really is going to take to accelerate your ability to bring on employees. If you just increase the engagement with using some conversational AI and get some gen-AI stuff in there and get a good, personality,and you still don’t get the rest of the thing. You have to do systems thinking, I guess, right? You have to think about the outcome. If I’m not orchestrating the acceleration of every step, I can accelerate one step, but that just means it sits in queue longer waiting for the next step. It doesn’t mean I went faster, right? That’s really where we go.
Scott King
Yeah, there’s book about that.
John Michelsen
Yeah, yeah. Then, well, we’re ex DevOps guys, right? When I, it was just a reaction to how do we take constraint theory and the whole idea around the, what happened in manufacturing physical goods and the need to orchestrate there. You take some of those, concurrent engineering principles and you take some of that stuff and you apply it to how do we adopt technology and you realize, we do it the exact opposite way than you would want to do it in order to get the kind of results that we’re looking for. Take a step back and what does that look ? And that looks a lot Krista. That’s at least our position on it.
Scott King
I think with kind the myopic view, your audience does have to be someone that sees the entire process because all of these systems and people and budgeting, they’re all built in silos, right? But the process is spanned across the different silos. If you’re thinking is not outcome-based, you’re not going to get it. I guess make sure that you understand the process and then you can see it, right? But I think it takes that kind of person.
John Michelsen
And you’ll, right, and that’s where you’ll find that even if this is sounding a Krista commercial and I don’t, of course I’d love for everyone of you guys to be customers, but I really don’t intend to try to do that nearly as much as just enlighten you on the mission that we have that I think that every organization must undertake. That, I think it stands regardless of if I was producing this product or if I was just an interesting bystander, interested bystander. This reality of we adopt technology in the wrong order or in direction is the way we usually describe it is fundamental. It’s really, and that whole dichotomy of how we get introduced to a customer and their initial reaction, is a reflection of that very thing, right? Because those who are attempting to think about outcomes or who are at least still signed up for that’s even a, they’re even willing to give it a try. They see that. And those that aren’t, you’re absolutely right, Scott. They’re thinking, “Wait a minute, I’m, I’m going to keep this as small as I can and I’m going to do just this bitty piece. And then I’m going to do lots of bitty pieces and then I’ll have done my job.” Well, depends, right? A leader is going to have to find a way to integrate and accelerate the outcome of a business, not a particular one task in a business. Otherwise, those that do solve for this have this incredible advantage in the way that they are able to take advantage of technology that you will not. And that’s not the position that anybody wants to be in.
Scott King
All right, well great, John. There’s a lot of great points. I mean, the LLMs are great, but what we’ve said before in previous podcasts, they’re only 5 % of the solution. I hope people, I hope they experiment and understand the limitations, because they will find them, and hopefully they’ll find us afterwards.
John Michelsen
Yeah. And that, and, and again, it’s funny because I’m an NLP guy forever. It’s why I entered the software category and have done it forever. And it sounds we’re trying to say negative things about LLMs and we’re absolutely not. We, we’re using this stuff constantly and it’s a, it’s a key piece of some of our most important use cases. But listen to what I just said. A key piece of some of, and that’s what’s odd. Maybe the, the, the beating the same old drum over all the time is software that can predict dates and numbers, that can identify activities as X, Y, or Z, whatever they happen to be, can do those sets of things, that’s tremendously valuable for a business. Think about what’s happening when full autonomous driving occurs, right? Because we’re on the cusp and we have been for a while. When that occurs, that’s not a generative AI model, that’s not GenAI tech. Now, GenAI tech may help us get across that final line and it will be five or 10 % or whatever and a very important five or 10%. The point is there’s a ton of other AI techniques at play to get autonomous driving to happen and that will have an enormous impact on most businesses more by the way than gen AI will on many businesses. But we’ve got to broaden that perspective right the euphoria around what the LLMs can do is great but we’ve got it then now recognize that in the context of what we’re trying to outcome and then take advantage of it in that regard, right? And make sure we don’t think of it as, “Well, now that I’ve got that hammer, I’m going to run around hitting everything with it to see if it’s a nail,” which is what we see a lot. Anyway, appreciate it always. Folks who are willing to listen in with us and get our perspective on things, I do, as I say, think that the argument here, the position here, I think it stands the test of time and rational thought, regardless of whether you think I’m self-serving to position Krista in a certain way or not. It’s interesting to see just how our industry has turned in a very odd way to get myopic, as you said, Scott, and that even some significant percentage of our customers start looking at the world and even adopting transformative technology in that myopic way, which isn’t going to get you where you’re transforming.
Scott King
Perfect. Thanks, John, and I really appreciate it. If anybody has any questions, you can, of course, contact us over at Krista.ai. And until next time.
