Camunda Execs On How They’re Navigating Through Agentic AI’s ‘Trough Of Disillusionment’

‘How do we get from this trough of disillusionment to the plateau of productivity?’ Camunda CEO Jakob Freund discussed at the 2025 CamundaCon event.

(Camunda CEO Jakob Freund)

Agentic AI has “huge potential,” according to CEO and co-founder Jakob Freund of Berlin-based Camunda.

Freund said during his keynote at the 2025 CamundaCon event Tuesday in New York City that despite his thoughts that agentic AI “completely redefines how organizations operate and compete,” there are still two fundamental issues that are stymieing agentic AI adoption in the enterprise: trust and lack of orchestration.

At the event, digital process automation software maker Camunda unveiled new capabilities in its business process and orchestration platform which is designed to help organizations integrate AI agents into their existing systems. The new features include an AI agent connector that “enables agents to operate autonomously within end-to-end processes” and integrate with various LLMs. Other new capabilities transform AI agents from isolated task-doers into “continuous, context aware” more autonomous processes.

The new capabilities address the trust and orchestration complexities around AI agents, the company said.

Trusting AI Agents

“Agentic AI is certainly a high topic these days. However, at the same time, you already entered this trough of disillusionment,” Freund said.

“There’s plenty of sources confirming that, yes, agentic AI projects look very exciting. However, [they] might actually fail, [they] might hit a ceiling, [they] might not deliver the value that executives are hoping to get out of it. So that sense, for example, only 8 percent are actually scaling AI at an enterprise level.

“The biggest issue, they’re not trusted,” he added.

“Twenty-seven percent of organizations only expressly trust fully autonomous AI agents. And that is actually a negative development compared to 12 months ago. How do we get from this trough of disillusionment to the plateau of productivity?” Freund said.

Daniel Meyer, chief technology officer at Camunda, took to the stage alongside Freund and offered insight into how Camunda builds trust into its agents.

“An agent that we use at Camunda ourselves is the GitHub Copilot agent,” Meyer said.

“What you can do is you can go to GitHub, your code repository, you create a ticket issue where you describe this is the code change that I want and then you assign that issue, not to a developer, but you assign it to the agent,” he said.

“And then what the agent is doing is it’s checking out the code. It is reading my text-based description of what is the code change that I want. Then it’s trying to understand what do I want. It’s changing the code, and then it’s sending me a pull request which I can review, ... that’s how that copilot is working ... It is using a number of tools ... checking out the code, changing the code, understanding the ticket, and then it is orchestrating those tools in a certain order, to achieve the goal that it has.

“Agents are orchestrating the tools that we give them, and they’re orchestrating those tools in a certain order to achieve their goal.

“We need powerful enterprise grade orchestration to build those agents,” Meyer added.

Orchestrating To Power Enterprise AI Agents

A daunting issue with deploying agentic AI is integrating agents with legacy systems.

“[If AI agents] don’t integrate well with your existing systems, your ERP, your CRM ... the customer journey, the employee journey, the end-to-end business process is going to explode in your face,” Freund said.

Meyer touched on the type of orchestration needed to build powerful agents.

“In the case of the GitHub Copilot ... it is sending me a pull request, it is not actually making the code change without telling me, right? It’s sending me a pull request, which I then review, and that is critical. If it wasn’t working that way, I would not be using it. I would not trust it to just change my code, push and done, right? And that is something that is built into that orchestration. But what is also happening is it’s dynamically trying to understand what I need and then making changes to the code which it then proposes.

So, this agent is bringing two types of orchestration together. And at Camunda, we call that deterministic orchestration. That is classic orchestration, where I create a flow chart to define what should happen in which order, and then the other paradigm is dynamic orchestration, where AI is deciding what should happen in which an AI is in the driver’s seat.”

“By having this orchestration layer, I can then power a new generation of agentic engagement layer, which doesn’t then only power chatbots, but conversational multimodal experiences that we will build in the future to transform how we are engaging with customers and how we are engaging with employees. This can only be achieved if I have an orchestration layer that can actually do that, and that is possible with Camunda,” Meyer said.