Summary
The session, sponsored by Camunda and featuring their CTO, Daniel Meyer, as a special guest, focused on re-engineering business processes for the high-velocity enterprise in the AI era. Meyer explained how legacy processes with manual handoffs can be transformed by first orchestrating them end-to-end and then selectively introducing AI, citing a case in which loan underwriting sped up by 90x. The panel, including Joanne Friedman, Joseph Puglisi, and John Patrick Luethe, discussed the continuum of trust in AI, the need for phased adoption, and the importance of guardrails, observability, and human oversight. They explored the differences between orchestration and choreography, with Daniel advocating for orchestration as the default. The future vision included more conversational, integrated experiences for customers and employees, with AI handling complex tasks across systems, potentially dissolving departmental and even corporate boundaries. The conversation also touched on compliance, auditability, and the dynamic nature of learning systems.
Attend The Great Process Re-Engineering
NYC on September 10. Click here to register!
Isaac Sacolick will keynote this event.
AI Is Reshaping Business — But Not Transforming Yet: How World-Class Enterprises are Re-Engineering Work
AI is everywhere in your business. Transformation isn’t, yet. Most enterprises have deployed AI, but few see it move the bottom line, and the gap isn’t the models. What separates the few is that they stopped bolting AI onto legacy processes and started re-engineering how work gets done. Isaac unpacks what world-class enterprises do differently.
Speakers
- Host – Isaac Sacolick
- Guest speaker – Daniel Meyer, CTO of Camunda
- Digital Trailblazers – Derrick Butts, Martin Davis, Joanne Friedman, John Patrick Luethe, Liz Martinez, Heather May, Joseph Puglisi, Elena Putilina
Discussion
- How are most enterprise business processes “legacy,” and when should leaders add AI to an existing process versus re-engineer an AI-driven process?
- What are high velocity enterprises doing differently in the AI era around governance, improving agent accuracy, developing trust in AI, and recognizing AI’s boundaries?
- What does “the orchestrated enterprise” look like in 3 years?
Research
Whiteboard
Transcript
[00:00:00] Speaker A: Welcome to this week’s coffee with Digital Trailblazers. So glad you are here for this really special episode.
I have a special guest, Daniel Meyer or Meyer. I keep going back and forth on that. Daniel, he is the CTO at Camunda.
Camunda is sponsoring today’s event and Camunda, really interesting company.
This episode is brought to you by Camunda, the open platform for agent orchestration.
And we’ll be getting into AI agents and orchestrations and what engineering the high velocity enterprise looks like here today in the AI era. Thank you for joining. If you are here, please do say hello in the comments stream. Let us know that you’re here and let us know your questions, your thoughts on AI agent orchestration.
And we’ll be getting started in just a few minutes. I’ve got Daniel Mayer here from Camunda. I’ve got Joanne Friedman here. I’ve got John Luth here. John Patrick Luth is here. I’m expecting Joe. And we may have a special appearance by Heather. That’s our group today, this August 14th. And welcome to this session on Beyond Legacy Processes Engineering the High Velocity enterprise. There’s a lot of language in here so I’m going to try to unpack it for all of you, but are listening today. Let me bring up the research slide for today and talk about some of the research I found on Camunda’s website. They have a paper here that you can link to State of Agent Orchestration and automation in 2026.
And what you can see here is a pretty sizable gap. We have a lot of organizations reporting they have process complexity stemming from regulatory complexity, branching and conditional logic. This is my favorite one. This is like that nice A to Z process that looks great on paper.
And then as you start getting into implementing it, you start finding out all the exceptions and all the rules that need to go into it and flash forward two, three, four years later and you end up with a situation that you have lots of branches and conditional logic and in some cases what I call spreadsheets and emails in the middle, things that the process just can’t accommodate. So just some of the process complexity that we see out there, we have legacy systems, some that are API, some of them are orienting and providing MCP connection.
But we still have a lot of legacy systems that are just not API friendly. And so we got a lot of process complexity. We have organizations just telling us, hey, we’re not ready for agentic orchestration, 85% of them. And that can come from not just about the state of their processes that can Be compliance factors, it could be trust factors and it certainly can be change management factors. I hope to get into all of those issues today in our discussion around beyond legacy processes engineering, the high velocity enterprise. And then what I share with you here, agentic orchestration across industries. This is coming from a graphic on Camunda site sharing things like in banking, hundreds of thousands of hours saved through AI contract analysis. Right? You get these big long documents and now you gotta extract metadata around it and know what the actual policies are in those contracts. And insurance, I love this one. 25% improved customer satisfaction through AI powered risk assessment. This is just good for the company and good for the consumer. This is coming from insurance and then in healthcare, 30% reduction in unnecessary procedures with AI diagnostics. Amazing data coming from here. What you’re seeing on the right hand side. Camunda is having an event on September 10th in New York City. The great process re engineering. And for those in New York City you have a treat. I will be keynoting that event.
My topic for this will be AI is reshaping business but not transforming yet how world class enterprises are re engineering work for the AI era. I will share the URL for that event in just a little bit and then you’ll be able to go out and access it. But for now I want to welcome our special guest, Daniel Meyer. Daniel is the CTO of Camunda. And Daniel, I want to just start with a base question for you. You know we talk about processes, we’re talking about high velocity enterprises, we’re talking about AI agent and agentic orchestrations. A lot of jargon there. But I want to start at the beginning where most organizations are struggling and talk about legacy.
What do legacy enterprise businesses processes look like? And then we have a situation where the easy thing may be to just bolt on AI. Take one box of the process and say we’re going to use AI and replace it in this process but not change anything about the whole end to end workflow. Or we’re going to get into situations where we want to relook at a process from the ground up and say let’s re engineer it and make it an AI driven process. So I want you to ground us in legacy versus when does it make sense to re engineer and when does it make sense to start just replacing parts of the engine. Daniel, welcome to the show. Just give a brief introduction to yourself and Camunda to get started.
[00:06:06] Speaker B: Fantastic. Thank you Isaac for that great intro. So I’m Daniel Meyer. I’m CTO at Camunda Camunda. We’ve been in the first business process management and then orchestration space for a long time. So we’ve been helping customers with improving their business processes, orchestrating their business processes end to end for while. In my, in my case it’s more than 15, 16 years now that, that we’ve been running Kamunda and been active in this space. And what’s really interesting is it’s not a new problem, right? This whole aspect of legacy processes, we’ve been seeing this for decades now. Basically a lot of work being manual.
You have a lot of manual handovers. So an email comes in, somebody’s got a look at it and then they need to input the data into a system and then somebody needs to pick it up from there, input it into the next system.
Now a lot of automation has happened over the last years, right. So there’s been a lot of bots being built, point to point integrations, integration platforms and so on and so forth. But the concept of orchestration, where I’m really looking at that business process could be processing a claim, could be onboarding a customer, could be underwriting a loan. So really that end to end process, looking at that and then orchestrating that end to end, there’s been very few instances where we’ve actually done that across the business. But whenever we do, then the value we unlock is on another order of magnitude than by just automating point to point or individual tasks. And now with AI, this becomes more and more relevant and quite crucial actually to effectively adopting AI.
[00:08:04] Speaker A: Wow.
Maybe let’s, maybe let’s dive into an example I’d love to hear, you know, pick something in financial services or insurance and you know, walk me through that sort of, here’s what legacy looks like, here’s what you know, you know, dropping AI in might look like and here’s what a re engineered process might look like.
[00:08:30] Speaker B: Yeah, I can give you an example where. So we’ve been working with a commercial lender in Australia and they had the typical situation that we often find where they want to underwrite new business.
And typically what you find is, and we all know this from our own experience, the, you know, when you want to get a new loan or something like that, you might try out multiple lenders at the same time, right. And then the one that replies to you the quickest and the fastest might get the business. And this was their situation as well. So they were losing a lot of business because they were just slow at underwriting it. Right. Going through that end to end process. And the reason they were Slow was because that end to end process had many manual handoffs in there, right? So they had the classic situation of you have people involved who don’t only need to complete a task, but who have to also actively move the process forward.
And they were just falling behind the competition. And what we did with them is we said, hey, let’s look at this process end to end. Let’s look at what are all the necessary steps that are there in order to underwrite alone.
And then first step, let’s orchestrate that end to end. So keep doing what you’re doing, let’s keep doing the same steps, but let’s orchestrate it end to end. So let’s model that process.
Let’s put it into our orchestration engine. And now instead of a human always triggering the next step, the automation does it, the orchestration engine does it. And that in and of itself unlocked massive increase in speed. And once we had that in place, that orchestration, then we could go in and look at, well, which of these steps actually have to be done by a human.
Can we assess risk using AI and then present the result of that assessment to the human to then approve it or reject it? Can we use AI to research information about the applicant and so on and so forth. So we put that in and then in the end, the end to end application process was 90 times faster compared to what it was before. So not 90%, but 90 times. So it was down from many, many days down to basically under an hour. And those are staggering results that you only get if you combine these two things. So orchestrating the process and then also looking at each individual step of the process and thinking about where can we actually introduce AI here and where is it not a fit?
[00:11:30] Speaker A: Wow.
So does that take us into RE engineering or is there another step beyond that that starts really getting into RE engineering after that?
[00:11:42] Speaker B: Yeah. The thing with the RE engineering, first of all, why are we using this term?
When I talk to business leaders, they’re all, you know, I must say scared to death about AI. And it’s actually two fears, right? The first fear is, well, if I don’t use AI, I will fall behind the competition. So it’s like this fear of not using AI, but then that is immediately followed by the second fear is, well, if I use it, is it safe, can I trust it?
And that combination is the state of mind of many business leaders currently that we’re talking to. So why are we using the term RE engineering? Because while now that we have AI, the latest wave of generative AI, all the business processes that we have in our organization that determine, well, how do we work, how does work get done, who does what?
That’s all legacy. Because all of those were designed in the world before AI existed doesn’t always mean they’re bad. They might be good, but they weren’t designed with the current generation of AI in mind.
So what we have to do now is re engineer them to understand, well, what does that process look like? How would we actually design it today now that we have AI? How would we be doing this if we were implementing it today?
And then also how do we make it safe? Right? How do we implement AI in a way that it’s trustworthy and that we can rely on the results? And then that is re engineering those processes.
[00:13:24] Speaker A: It’s something I actually encourage a lot of organizations to do. And the reason is exactly what you said. There are vast capabilities today that are beyond orchestration, connecting end to end steps in a process, beyond just automating a step in a process, whether it’s a bot or some other tool that you’re using to take a step in just making it more efficient, faster, improving quality around it. Now I can take a whole bunch of edge cases and start asking an AI step to do the analysis and to come up with recommendations and then start thinking about, well, which types of recommendations have low risk that I’m just going to give it that agentic thumbs up, start using this, start doing this, and we start getting into a completely new way of working.
Thank you, Danielle. I’m going to go around to our other advisors and experts here at the coffee hour. Joanne’s first to raise her hand.
Joanna, why am I not surprised here? First we’re talking about AI and agent orchestration and process re engineering.
This is straight in your wheelhouse.
How do you want to separate when organizations should be looking at re engineering versus just putting AI to automate a step in the process?
[00:14:51] Speaker C: Well, I think I want to recount a bit of a conversation that, that I had with someone yesterday.
And this was not.
I’ll get to agentic choreography versus orchestration in a second. But the conversation kind of went like this.
The question was asked, well, isn’t trust something I can just use, like an on off switch? You know, flip the switch, I let it run autonomous and the agent run autonomously. It’s either on or it’s off.
And my answer to that is that’s a fool’s errand waiting to happen. And if you want to blow up your facility, be my guest and I didn’t mean that to sound as sarcastic as it actually does. But the point that I was trying to make to this individual is don’t make the mistake that many executives do, which is thinking that an agent or Trust in AI is an on off switch.
You don’t go from 0 to 100.
You go through a phased gate approach.
If you use the analogy of a switch that it’s on or it’s off, you’re already dooming your project.
If you look at it more like a gate that has a floor and a lift, then you can start seeing where things actually come into play. So here’s the example that I gave this individual who said to me, well, give me a for instance. And I said, okay, think about this.
You’re looking, if you look at the switch, you either give an agent permission to do something or you don’t.
And you can put human in the loop, human on the loop, human on the hook, anywhere in that process. But if you really look at something from the point of view that a process step or a process is not an island and it has dependencies and co dependencies and cross references to other processes or other parts of a process, then you start getting the impression that the gate has to have a floor and that floor is the worst possible situation.
Can you take it? And then you have a thing called reversibility. Can you take it back?
So first there’s think of a gate opening and closing more than a switch going up and down where you either are working with it or it’s, it’s dead.
In that worst case scenario, it’s very easy to take something like a claim that’s been sitting in limbo for a while and suddenly execute upon it. But if it happens to be torque on a part in an automobile engine like the recent recalls that Ford had, what you’re doing is taking the worst case scenario, someone dies because of it and you look at reversibility. Well, if I allow this agent to execute on a trusted path, which you know is something that we built into Gail, that trust is earned on evidence. And it increases with evidence showing that it’s reliable, it’s safe, and so on. In the case of the Ford situation, you’re not going to give an agent carte blanche, go do without humans in the loop because human safety is involved in healthcare. It’s built into the law. You can’t do it. But what you can do is start looking at these things. Well, what would a codependency be to giving that agent permission? Well, has it Passed quality.
Well, what, what happened to the manufacturing process if we’re already talking about quality and keep going backwards? So a lot of times the conversations go around. If you had to go back to square one, where are all the gates?
Where’s the floor of each gate? Where’s the lift of each gate? And start reconciling your business processes around that. Now, the difference between my perspective and Daniel’s might be that we look at choreography because it has more flexibility and more adaptability and it’s not a central controller the way many agents are designed.
Choreography allows you to have shared resources and bring some of these dependencies and codependencies into play while you’re dealing with the governance and authority agency issues, which is what you’re really talking about.
[00:19:22] Speaker A: Joanne, a couple really interesting concepts there. I want to go back to Daniel before we bring Joe up.
Daniel, I just want to get your thoughts on this like continuum of trust, especially as we take something that had a lot of those handoffs. I mean, what you got in handoffs is somebody eyeballing something in between point A and point B.
So number one, your thoughts on the trusted path that Joanne was just bringing up and then this notion of choreography rather than control, are those two of the elements that we should think about when we’re redesigning re engineering an AI driven process?
[00:20:04] Speaker B: Yes, absolutely. So first of all, yes, trust is, trust is critical.
And I love the way you, you, you phrase this. Joanne. It’s a continuum, right? It’s not an, it’s not binary.
And we need to, we need to develop the maturity and we need to.
Basically, it’s like climbing a ladder. Right.
For me, the, the test is always, these days it’s very easy to build an agent. Everybody can basically build an agent. What I’m always asking people is, well, you probably have a brokerage account and in my case, mine is on public.com. it has an API. The test is the agents you’re building, you connect those to your brokerage account and then whenever I ask, people always say no, of course not.
When you ask why then, well, I wouldn’t trust it with my money. Right. But that’s the task. So if we want to bring AI into mission critical enterprise operations, decisions where stakes are involved, then we need to build it in a way where we have trust that we can actually delegate some of that work to those agents. And well then the question is, how do we build that trust? And to me, it is about the following things. So I need to design for it, so I need to be able to define structural guardrails. So not just the prompt where I say, hey, please, please, please always ask me before you trade my money. And then it’s up to the LLM to follow that instructions or not. But I need to be able to design a guardrail where the LLM cannot override the guardrail that whenever it trades, it first asks me for approval.
So that’s the design part I need to verify. So I need to be able to test and verify that whatever guardrails I have designed are actually being followed. And then I need to have observability and visibility. So I need to be able to observe while the agents are running, whether what they’re doing is actually what they were designed to do, whether they’re performing, meeting their qualitative gates and so on and so forth. And then that leads into the last piece is I need to be able to intervene if like the, you know, the off switch or take corrective action if an agent does not do what it’s designed to do. And if I have those pieces, I can design for trust, I can test for trust and verify, I can observe and then I can intervene.
That’s the framework, right? So I need to have these pieces in place, then I can start building incrementally agents and orchestrations that I am then willing to actually delegate mission critical and high stakes work to. So that’s the first piece.
The question with orchestration versus choreography.
This is an age old question, so to say. This has been discussed for many years. It was also a big topic during the microservices era where basically what’s the difference between those two? When I’m orchestrating, then I have an orchestration that delegates work to other parties. So now with agents, that’s agents. So I have almost like an orchestrator agent that is like the conductor in an orchestra and says to other agents, now you do this and now you do that. So that allows me to break complex high level tasks into smaller tasks and then delegate those to for example, specialists.
The choreography pattern is, is a bit different there. I have shared resources, I have a way of communicating, often asynchronously, where then independent agents, if we’re now in the agent world, can collaborate to then collectively solve a problem.
And both patterns have their, their pros and cons and also have their applicability.
We always say the default is orchestration.
That’s what you should use by default. And if you reach certain limits with that, then choreography can be helpful.
This often a good fit with orchestration because it’s also how we are typically organizing work within teams when we’re working on tasks together. You typically have a manager might break down tasks and then other people are working on them. Could also be a project manager but the larger the organization gets, the more you’ll want to then also have more loosely coordination.
That’s then where choreography can fit in very well. So you can combine those two methods basically.
[00:25:29] Speaker A: Thank you Daniel. I do have a follow up question but I want to hear from Joe.
Joe, welcome to the floor. We’re talking about going from a legacy complex mission critical type process and now thinking about going into AI.
You know, how do we rethink our future and re engineer an AI driven process? Your thoughts today Joe?
[00:25:56] Speaker D: Yeah. To achieve a high velocity enterprise infusing AI is to recognize where the humans involved are either truly adding value or introducing friction or stumbling blocks along the way.
And you know, I was reminded of the old adage of paving cow paths or the story of the holiday roast. I’m sure we’re all familiar with the legend of little Mary asking mom why you cut the roast in half and her not knowing.
She goes to her grandmother who ultimately explains that in her day the ovens were so small you had to cut it in half to put it in. So you know when you go through your legacy systems you find a lot of places where you’re cutting the roast in half because the ovens used to be too small.
And it doesn’t matter whether we’re talking about AI or any other technology and whether we trust it or we don’t trust it or we can infuse it or not infuse it into the process. You really have to look at where does the process add value, where is it necessary or where is it simply introducing friction for perhaps for no good cause.
[00:27:15] Speaker A: Thank you Joe. I’m going to go back to my follow up question with Daniel in just a second. I want to welcome everybody to this week’s coffee with digital trailblazers.
Our episode today is sponsored by Camunda. I want to thank you for joining with Camunda CTO Daniel Meyer. We’re having a great discussion on on process automations and I want to just share Catch me keynoting the great process re engineering event on September 10th in New York City. To continue this conversation, you can see the banner for it.
It’s on the right hand side. It’s in New York City in Midtown. My keynote will be about AI is reshaping business but not transforming yet how world class enterprises are re engineering work.
I will Paste the UR URL to register for that event in just a second. And Daniel, I’m just thrilled to have you here join us because we’ve been talking about AI agent orchestration, you know, how much agentic we’re going to be in the future, how much human augmentation are we going to be and you know, where does it make sense to have human in the middle some of these processes. In order to build trust, I want you to go deeper and just share with me what high velocity enterprises are doing differently around governance, around accuracy, around trust and around building boundaries. And let me just go a little bit deeper here, Daniel.
You know, when I think about trust and financial services and insurance, I’m thinking about compliance. So even if I have, you know, a CIO and even a CISO and some business leaders really excited about using, you know, an agentic process around underwriting or around fraud detection or around asset allocation, I’m going to have somebody in compliance in risk management who is going to ask a lot of questions about trust and process and audit ability. I think that’s, you know, one side of this, you know, you know, what are high velocity enterprises doing differently? The other side of it is, you know, when you start getting into this notion of choreography and you think about process re engineering in the past, you know, I could show a six Sigma black belt, a tool with an end to end process with a lot of bunch of boxes up in there. It’s pretty much flowing from left to right. It’s got some branches in there to deal with things. It’s got some exception handling processes. It’s pretty easy to follow that linear flowchart. And now we’re talking about choreography, we’re talking about asynchronous operations, we’re talking about multipathing and looping. It’s just not that easy to go to people and say here is exactly how the thing works. So how do you gain trust when you go back to those operating business leaders and say, you know, let’s, let’s build this up together and go from where we were today in a legacy process into a more AI driven process. So let’s start with that first question. Let’s talk about compliance.
[00:30:38] Speaker B: Excellent. Yeah, with compliance, typically you have multiple requirements there, right? So there’s requirements like, let’s say KYC. Okay. Each decision must be auditable. So we need to be able to prove how we made the decision based on which data and we need to be able to prove that we did the kyc, that we did the right steps and in order to do that, we need to, again, getting back to the trust framework, right? We need to design the process to make sure that in each case it does the right thing, the right actions and steps are executed. And then we need to have a complete audit log to prove to an auditor or to ourselves that we have actually executed those steps. And when you have an orchestration, then that’s exactly what you do, right? You will design this process visually very similar to what you actually described. Now, Isaac, right, so you, you design your flowchart where you say, this is where it starts.
Here are the steps in the process.
And now what’s new with AI is you can have part of that process, part of that visual flowchart. You can say, well, here I have AI deciding on the next steps, right? So I’m not connecting activities with, with arrows and conditions, but I’m saying, here’s the list of possible options. And now here AI is deciding which of those options to pick.
So could be something like, well, we will always validate the data, but then we will have AI deciding, well, based on the data that we have, which steps do we need to, which checks do we need to perform, do we need to do a sanctions check, do we need to do an advanced enhanced due diligence, and so on and so forth.
Then at some point it says, okay, I have checked this customer, here’s the results. And now again, via the flowchart, I’m enforcing that a human now reviews this result, for example. And if I’m doing it like that, so if I can model exactly where do I want to have what we call deterministic orchestration, where modeled process logic, a modeled flowchart governs the process and blend that with what we call dynamic orchestration, where AI in a goal driven way decides on next steps, then I have the full flexibility. And then I can also control where do I enforce structural guardrails and where do I give AI the freedom to decide? And then I can apply that depending on the decision or depending on the use case in a way where I’m balancing autonomy with trust in a way that that use case requires it. Does that make sense?
[00:33:52] Speaker A: It makes sense. There’s a question here from a really good friend of mine, Wayne. I’m so glad that you’re here today. And he asked this question, you know, about ownership, like, you know, this notion that we’re starting to hand off decisions to an AI and creating a continuum around it.
You know, how do we, in your system, how are, how is ownership managed? How do we Think about that.
[00:34:22] Speaker B: Yeah, well, you have.
So you have a design time and a runtime, right? So at design time I’m defining the process, I’m defining the agent, the prompts and all of those things. So the person designing that has the ownership to make sure the design is sound and it’s verified and compliant with the requirements that we have. I think that is unchanged. That is the same we’re familiar with from classic let’s say business analysis, requirements, definition, implementation, and so on and so forth. So if you maintain a design time, then you can also have that design time responsibility for people to design agentic orchestration correctly. Then you get to runtime and then it’s about, well, which decisions in my process are made by AI and which decisions are made by humans, which decisions are implemented as code. Right. As just rules and so on and so forth. And where is it a combination of, of the above. Right. So where do I, for example, have AI preparing a decision and then a human reviewing it and then that’s your toolbox, right. And then you, you need to think about for each decision, well, is this something that I’m actually comfortable to delegate to AI or is this something where I want AI to prepare it, do the research and then present it to a human?
Or is this something which always a human needs to do?
So you have that flexibility and then you can apply it in a way that makes most sense.
Ultimately, if decisions are wrong, you can’t blame it on AI. We own what we build and we own what we design and we own what we run.
So only if we design it correctly, then we can also make good on that ownership, if that makes sense.
[00:36:46] Speaker A: It does. And built into designing the process. There’s a question here, I forget who asked. I’m looking at the common stream about quality of output.
I think it’s from Orhan. The agent can be trusted if set properly, but not producing high quality output. How does that factor into the overall re engineering process?
[00:37:11] Speaker B: I mean, I don’t know what you think. I think that hasn’t fundamentally changed. Right. In terms of if I’m, you know, if you think about AI as a technology, a tool that I can, that I can use to implement decisions or bring reasoning into a process or for example, understand intent.
What does the user want?
You often have that in customer support.
What’s the user’s intent? What’s their problem? What do they want? And then mapping that intent to a solution or mapping that intent to a process that then provides the user with that solution.
I think this hasn’t fundamentally changed.
If this technology works, then it’ll produce the right results and if it doesn’t, it doesn’t. And then we need to build the maturity in terms of, well, when does AI provide the right results? And I think we’ve collectively learned a lot about that in the last years. And it is really about ultimately context and guardrails. Does it have the right context then? Are there the right guardrails in place so that I can de risk bad decisions?
Then within that space it can be an extremely impactful tool.
[00:38:50] Speaker A: Thank you, Daniel. I actually have a follow up. I’m going to hold off and bring in Joanne. Joanne, we’re continuing going down this thread about what high velocity enterprises are doing around governance and developing trust and you know, what guardrails and boundaries look like. Just your thoughts, Joanne?
[00:39:10] Speaker C: Well, when it comes to guardrails and boundaries, one of the things that I said earlier was if you think about a gate with a floor and a ceiling or you know, a ladder, you can’t get to the top of the ladder without evidentiary proof.
And that’s on the reliability of the agent, that’s on what it’s tasked with doing, how well it communicates with others, if it’s a swarm, etc. So there’s a lot of factors. But one of the things that Daniel said I just wanted to key on because it is about two things that one has been mentioned, the other has not. Context rules because even though that’s an oxymoron statement, context needs to have a semantic spine.
And in that comes intent, it comes perspective, it comes so many different variables that if you don’t have a really, really well balanced semantic spine, you’re going to run into difficulty when it comes to reliability because agents drift.
And the part that has not been mentioned is the fact that this whole setup of agentic AI is very much dynamic. And one of the guardrails that people don’t put around, which they really should, is this notion that it’s a learning system. Because every time an agent runs it should be collecting more data and more knowledge from its environment, from its context and so forth, so that it’s reliability improves, trust improves and speed increases. So that the situation that many people in high velocity organizations get into is something that’s very well known in business, not so well known necessarily in technology, but it’s about trade off management.
And really that’s what it comes down to. What trade offs are we willing to make with what cost and what consequence? And that’s the ultimate point of where you have to decide the rules and the governance for that agent. What is, what is the last mile? So execution is actually AI’s last mile. And this is where all of the work and everything that we’ve been discussing for more than a year, really, the rubber hits the road.
[00:41:38] Speaker A: I think there’s a lot to that. Joanne, I want to hear. John. Yeah, it’s good to see you, Joan. We’re talking about guardrails and agents and orchestration.
Your thoughts today?
[00:41:49] Speaker E: Yeah, I was going to say a couple, couple thoughts. What is on guardrails? And I know from personal experience that agents can do anything that they actually have permissions to. And so if they have permissions in the system to do something, they can actually do that thing. And so because I’ve seen situations where the agents weren’t built good enough, but you give them read only tasks and after they have read only tasks for several days, they just decide that they’re going to make the change themselves. And so when people want to know what their agent can do, you really have to look into the system and see like what are the computer permissions, system permissions, what is this span of all the data can it get access to? And what’s the span of all the things, changes it can make in systems? And we’re also seeing that in the news too, because I think it’s a source of pride with all these different frontier models saying that their model escaped something and hacked some website to obtain some goal that it was going after.
And I personally have seen agents go beyond what you’ve asked them to do.
And then another comment, people are asking what does it make sense for us to do? Should we put this investment in?
And when we started talking about insurance, that’s near to my heart because you know, several, several years ago I ran activity based costing on several thousand claims people. And what we quickly learned is that there’s tasks that add value and there’s tasks that don’t add value.
And what you really want to be doing is automating the tasks that add value and the tasks that don’t add value. You got to figure out how to get rid of those things.
And so my favorite comment comic is actually one on does it make sense to put an investment on this automation that saves time?
And if you can do the math and you can eliminate something based off how many people are running it and how frequently you run it and how much time savings, it gets really clear whether or not you should, you should do this investment to save time. And one of the things I’m really happy about just seeing lately is that 20 years ago people started building business applications on web browsers and, and then start companies started using SaaS and so they really weren’t running their own software in many places. But the places where there’s no integrations, the agents are so proficient at logging into websites right now. I just, I see that like there’s going to be so much opportunity to save time across processes that involve different SaaS applications and different web applications. It’s just there’s going to be a lot of neat savings going on while
[00:44:23] Speaker A: we’re learning all these best practices. And you know, we got about 15 minutes left.
I want to bring all of you into the future a little bit because we know we’re sitting on a lot more legacy processes. We’re still using some of our old tools of the past. We still have a lot of change management activities. We have to build trust now we’re bringing AI in. We can instill AI in steps.
We’ve been talking today about what high velocity enterprisers do to create AI driven processes. I want to look into the future a little bit and stay away from the hype.
Daniel, I want to get your sense of what does an orchestrated enterprise look like in three years. And I’m just going to throw a couple things out there. One thing, something you were touching on but didn’t directly say.
You know a lot of our decisions when we thought about linear processes were pass fail were on off types of decisions did are you approving a loan or not? Is this fraud or not? And I think what these agentic enterprises are going to agentic decisions are going to do is give us more of a continuum. Right? Maybe the thing you asked for is a no, but maybe the agent is going to come back and give you a lot more options to consider and trade offs to consider that maybe are available and yes to you. So maybe that loan that you won at 2%, that’s tenfold your net worth is a no. But, but here’s what I can do for you. Just as an example, I also want you to comment on Joanne’s point about the learning enterprise as part of this and I love this idea that we’re building agentic processes that are going to have feedback loops and get smarter. Except we have this issue of maybe we’re learning the wrong behaviors just like our children do. So how do we put feedback loops on feedback loops to make sure that the right behaviors are amplified and the poor decisions or the wrong ones are managed as we’re trying to get into more of these re engineered processes. So maybe comment on both of those.
[00:46:51] Speaker D: Oh wow.
[00:46:52] Speaker B: Yeah, there’s a lot there.
It’s interesting right now. I think we collectively always fall into the same traps, right? Every, I don’t know, every 10, 15 years there’s a new transformative technology and then we think, well, that technology by itself is going to solve it, right?
And then we figure out, oh wait, we actually have to do the hard work of transforming and so on and so forth. And I think that’s what we’re also discovering again collectively now with AI. And so AI is only as good as the people who are training it and wielding that thought, so to say. Anyway, so where does all of this lead to?
I always try to think backwards from well, what is the experience we want to provide to people? So when I’m a bank, what is the experience I can provide to my customers and what is the experience I can provide to my employees?
And today that experience is often still very much disconnected, right? We’re all very familiar with that. Just try to cancel your cable or try to, I don’t know if your cable and your TV contract are linked, try to cancel one of them and then you’ll soon see how disconnected the organization is you’re interacting with, right, as a customer.
So the opportunity is now with AI is that we can provide a much better experience to those customers and employee.
And ideally we can hide that underlying complexity and disconnectiveness, so to say, if that’s a word that we have in the organization, by providing a more, let’s say, a smarter engagement layer for those employees and customers. Because what is AI very good at? It’s conversational, right? So if I don’t have to log into a bunch of tools to get something done, but if I can just have a conversation with something that then gets it done on my behalf, and then that AI logs into all those tools and does it for me, then that’s a huge win. And I think that’s the pattern that we’ll see. So organizations will provide an increasingly integrated and conversational experience to customers and employees.
Below that you need an intelligent orchestration layer that then can map what the user wants to intend. So it can map their intent, can map that to processes, agents, rules, decisions and so on, which then orchestrate across existing systems that are there. And those systems are not going to go anywhere for the foreseeable future.
So organizations will be keeping their CRM, will be keeping the ticketing system will be, you know, we’ll be keeping some of their legacy systems and so on and so forth. But we can provide a very transformed experience on top of that if we can orchestrate and that’s for us the key technology. Now in between. So in between that AI driven intelligent experience and those existing systems and that existing legacy, I need to be able to, to orchestrate in a way again that is adaptive, that is intelligent, but that I can also trust by enforcing deterministic guardrails and decisions and so on. And that’s at least how we are thinking about it and what we see resonating with many of our customers.
[00:51:04] Speaker A: That’s a really elegant and expansive vision. I like how you’ve brought together that sort of divide between customers and employees with complexity in the middle.
You know, you feel it every time you, you are getting help from your, your bank, around your wealth manager, around investments, all the way down to your car mechanic. Right. What are my options here? And you’re getting conflicting information and leaving to you as the customer to make decisions that may not fully understand all the trade offs on. And so I think this notion of conversational experiences, when it’s really done well between an employee information that an AI is providing back to that employee information that the AI is bringing back to the customer and we’re having a real dialogue around this gets really interesting. And if you want to try that on a personal level, just go plan your next vacation with an AI and you’ll see and start feeling what a conversational experience can look like. Joanna, your thoughts on what the orchestrated enterprise looks like in three years?
[00:52:20] Speaker C: Well, the choreographed enterprise is fluid. There’s fluidity across the organization, not the compaction that comes with departmentalization.
Because Agentix will allow us to have a holistic view of what’s going on. You know, processes are connected to each other.
None of them are islands. Now as we get rid of silos and I, I think that’s one of the biggest things we’re going to see is the ability for the AI to from a contextualized perspective, pick up all of the relevant information and give us the whole picture. So a blue ocean view if you will, rather than a minuscule. In this system, in this process, this is the answer. So that we will have a lot more capability based on the enterprise as an ecosystem of groups of people and systems, et cetera. So when you want to make a decision, you’re better informed. You, you’re not using dashboards that basically say here is what just happened. And everybody’s left to say now what? Because knowing is not enough. Being able to have all of the factors to make better decisions is one aspect, I think, because, you know, coming from complex industries or complex workflows of manufacturing and supply chain and whatever, we’re going to see the beginning of companies that are collaborative with each other. So agents will traverse boundaries with the right permissions, with the right guardrails, with the right compliance, the execution of information flow will change. So you’ll know what your suppliers are thinking if you’re a manufacturer, and you’ll understand the feedback from your customers all being rolled into one. I shouldn’t have to wait for someone to say to me, you need a spare part to replace something on a production line.
Oh, we don’t have a part in stock. Oh, now I have to wait for another person to get involved to figure out who the supplier of that part is, and then go through that whole process.
The agent will be able to say, we don’t have a part in stock here, but we have one at another facility. Do you want me to have it shipped?
That’s fluidity within one organization.
If they can’t do it in time, here are the nine other suppliers who can provide it in a reasonable period of time. And here’s their cost structures, and here’s what expediting the shipping is going to cost you as well. Do you want me to fulfill this process that removes downtime, that changes the game completely, and industry by industry, we can go across the board. So I would see that happening in not far distant future. I would say probably within the next three to four years.
[00:55:18] Speaker A: Thank you, Joanne.
Lots unpacked there like that, you know, from silos to blue ocean views.
It’s a very pictorial way of showing what the orchestrated enterprise might look like in three years. Joe, be a futurist.
[00:55:34] Speaker D: I think Joanne is spot on.
I was going to say pretty much the same thing really, that I think in the near term, as Joanne estimated, you know, three to five years, we’re going to see department boundaries dissolve.
The notion that things are done in this department and that department, you know, networks came along and made the information flow among departments work better.
But AI has the capability of sort of eliminating departmental boundaries. I think Joanne’s illustration was, was. Was quite spot on. If we go a little bit further out into the future, I think corporate boundaries begin to dissolve and the whole economy starts to interplay with, you know, the supply and demand of goods and services in a very highly orchestrated or choreographed manner.
Daniel’s point about customer service reminded me of how many times each of US have called into an IVR system to be handed off. Ultimately, when you get frustrated enough to a human who says may I have your account number?
And then after understanding your problem, transfers you to another department that handles that. And what’s the first question?
May I have your account number, please?
[00:56:56] Speaker A: We’re going to start fixing the stupid and the simple and get into all the way to. Corporate boundaries dissolve and the economy transforms.
Somewhere in that three to five year time period. Look, I, I think that we’ll be somewhere in that continuum. My, my point of view. John, very quickly, your thoughts on the future.
[00:57:19] Speaker E: I see a future for the, for the people that primarily work, you know, with computers is that a lot of their time is, is really spent giving tasks to the agents and, and having the agents run run and do these tasks and, and then the humans really kind of checking in on the agents and seeing, seeing what they’re doing and kind of making sure that they’re going in the right direction. And so I just, I think, I think we’re going to have a future of. It’s the, the office work is largely going to become a lot more supervisory than it that it is today and that a lot of the heavy lifting is really done by agents and that the humans are at a higher level making decisions, reviewing things much in a a way that a manager word or director, but, but that’s going to be the entry level people actually. And so that’s, that’s my view of the future.
[00:58:07] Speaker A: Joanne, you got 30 seconds and I gotta give the mic back to, to Daniel. Go ahead, Joanne.
[00:58:12] Speaker C: Yeah, sure.
No, to John’s point, the notion of forward deployed engineer.
Try this. Forward deployed expert because engineering is one thing, but expertise, domain expertise in a topic, whether it’s compliance, healthcare, insurance, manufacturing, construction, irrelevant.
There’s expertise that’s required and small language models and specialized language models will start to really come to the surface. I mean, think about it as like hugging face on steroids, but with real guardrails and real experts behind them. That’s what’s going to make that gap between organizational boundaries going away and corporate boundaries going away.
[00:59:00] Speaker A: Daniel, this is what happens when I ask people to be futurists.
We are becoming futurists. Daniel, I’ll give you a last thought today.
[00:59:09] Speaker B: Yeah, I think we covered a lot today.
That was fantastic. I like the idea of organizational boundaries dissolving over time.
We’ll see how far we get with that.
What I want to add from our side Camunda side is I think you brought up the event, Isaac. So if you’re in the New York area, please join us there.
And then we also have an upcoming webinar about a tool which we call Process os, which allows organizations to build an operating system around their processes and they can use that to transform their processes rapidly to join this orchestrated enterprise future. So thank you for having me today and it was a great conversation.
[01:00:01] Speaker A: Thank you Daniel CTO of Camunda, our event episode sponsor, for joining today. 71% of enterprises have deployed AI agents. Only 11% have reached production. The problem is not the AI, it is the processes underneath. If you are in the New York City area, I hope you’ll join us at the great Process Re Engineering. Scroll into the common stream on LinkedIn to be able to find the URL to join it or DM me and I’ll send you the link. Our upcoming episodes on the 21st we’ll be talking about shadow AI at work. Real risks, practical guardrails, hidden innovation. The 28th we’ll be talking about SAS cloud and AI contracts where transformation leaders lose leverage. And on the 4th of September, we’ll be taking a week off to enjoy our Labor Day weekend. Thanks again Camunda, for being our episode sponsor. Do Visit them@camunda.com that’s C-A-M u n d a dot com. Thank you for joining us this week and we’ll see you here for next week’s episode.
Leave a Reply