After a year or more of AI investment, something predictable and preventable starts to happen. Agents are working. They are delivering value. They are also scattered across the organization, owned by different teams, doing useful things in isolation. And the people responsible for end-to-end business processes are starting to ask the same question: how do we bring them together?
At CamundaCon 2026, we closed the day with a panel of practitioners hitting that exact inflection point.
- Dr. Simon Eisbach, Head of AI and Automation Platforms at Provinzial
- Jonas Viehof, Consultant at Audi
- Lee Kiew Seng, Head of Orchestration and Integration at Danica, part of the Danske Bank group
- Leon Strauch, Senior Manager, Agentic Orchestration at Camunda
Three different industries. Three different starting points. The same realization underneath: the agents are there. What is missing is orchestration.
The pattern: capability without coordination
Audi has multiple agents running across business units, each built by a different team, each delivering real value. None of them are yet connected to the end-to-end process they belong to.
“They came from different entities, different organizational units, and they all add a lot of value,” Jonas explained. “But they are not yet planned to be really orchestrated into the process. That is something we, as the process design department, need to take care of.”
Danica is in a similar place after twelve months of foundational work. “We have all the building pieces on our table,” Kiew Seng said. “What we need to do now is realign with the rest of the stakeholders and come out with a stronger plan.”
Provinzial has gone deep on one specific pattern, agent-assisted employees, and is being deliberate about scaling further. “Where we are very strong is agent-assisted employees,” Simon said. “A sales agent has an AI agent on its side that has tools and can support, answer questions, or help guide through software.” That path moved faster than embedding agents in core business processes for a reason: “It’s an easier entry, because you have pre-designed the human in the loop. With regulatory limits, you’re a lot faster on this end.”
Three enterprises, three industries, the same shape. Pockets of working agents. End-to-end processes are still fragmented. The next move is not more agents. The next move is not more agents, it is agentic orchestration: the layer that coordinates agents, people, and systems into processes that actually complete.
The pattern is bigger than these three companies. As noted in the Camunda’s State of Agentic Orchestration & Automation 2026 report, 71% of organizations are already using AI agents, but only 11% of agentic use cases have reached production. Nearly half (48%) say their agents operate in silos and aren’t woven into end-to-end processes.
Why “more agents” stops working
The instinct, when individual agents are delivering value, is to do more of the same. Build more agents. Deploy them in more places. But every agent added without orchestration adds fragmentation faster than it adds capability.
Leon described what we see across our own customer base and our own internal work.
“If you ask an LLM the same question 10 times, you’ll get 10 different answers. How do you produce reliable output at scale using LLMs? That’s one of the core challenges, and it’s what distinguishes a nice-looking demo from a production-ready solution.”
— Leon Strauch, Senior Manager, Agentic Orchestration, Camunda
A demo can tolerate variance. An end-to-end business process cannot. Claims handling, order-to-cash, customer onboarding — these are processes where the same situation has to produce the same outcome, audit trails have to be complete, and exception handling cannot depend on whichever model the agent happened to pick that day.
That conversation has to shift from “what can my agents do” to “how do my agents operate inside a process I can see, control, and trust.”
What changes when orchestration enters
Three things change when you put orchestration around your agents.
- One: agents become participants in a process, not standalone tools. “It’s actually easier to build a dynamic flow than to build a deterministic flow,” Kiew Seng said. In insurance, with thick rulebooks and conditional logic, that matters. A dynamic flow lets the agent handle the variability while the process holds the structure, the guardrails, and the audit trail.
- Two: the compliance conversation becomes possible. This is where the panel was clearest, and most aligned.
“Getting the technical stuff to work, that’s the easy part. The difficult part is the legal, the compliance, all the questions you’re going to get. That, to me, is the most difficult one.”
— Lee Kiew Seng, Head of Orchestration and Integration, Danica
Jonas described the same reality from Audi’s side. “We’ve got all of these legal requirements that we need to take care of. We need to document where we are using AI, how we are using AI, which data is going into the AI.”
An agent in isolation is hard to govern. An agent inside an orchestrated process inherits the governance of that process: approval steps that must happen, escalation paths that must execute, every decision and exception captured automatically. That is what makes the EU AI Act conversation tractable rather than terrifying. It is also why agentic orchestration is becoming the foundation enterprises build on before they scale AI further.
- Three: the value compounds. On the business side, agents handle dependencies, classification, and repetitive work, so operations teams get time back for complex decisions. On the IT side, dynamic flows are faster to build than deterministic ones for rule-heavy industries. On the customer side, faster responses and higher satisfaction. Simon added one more, often overlooked: knowledge transfer. “We have a lot of people going into retirement, and it’s difficult to find people who know the right answer.”
Leon framed it the way we increasingly do at Camunda: augmentation, not automation. The point of orchestrating agents into business processes is not to remove people. It is to take the mundane work off their plates so they can spend their time on what actually moves the business.
What this looks like in practice
The transition from agent islands to orchestrated processes does not have to be a re-platforming.
Audi started with structuring unstructured customer data. The agent reads incoming messages and routes them into the process. It does not yet send replies. “If I were to imagine actually replying to the customer, that’s something where we would always have a human in the loop, at least at the beginning.” Agents inside the process. Humans at the decision point. Systems connected end to end. That is what orchestration looks like from the inside out.
Provinzial runs a hub-and-spoke model: the platform team owns the foundation (LLM access, governance, the orchestration layer), and business teams own the processes. “Our job is to make it really easy and comfortable to do it right,” Simon said. The foundation is what lets the rest of the company move.
Camunda is doing the same internally. After a phase of smaller use cases spread across the company, the team is now re-engineering core business processes (quote-to-cash, employee lifecycle, accounting) and systematically injecting agentic AI where it adds the most value. Not as a parallel “AI process.” As part of the actual process the business runs.
The takeaway
If your organization is in the “we have agents but they’re scattered” phase, the next investment is not another agent. It is the orchestration layer that turns those agents into participants in end-to-end processes you can govern, measure, and improve.
The companies that get this right will not be the ones with the most agents. They will be the ones who treated agents as part of the process from the start, with the deterministic guardrails, the human checkpoints, and the audit trails that make it safe to give AI real operational authority.
That is what agentic orchestration is for. And from what we heard on that panel, the enterprises that figure it out first will set the pace for everyone else.
