Agentic orchestration: A Computer Weekly Downtime Upload podcast

For Jakob Freund, CEO of Camunda, business processes are legacy, because they were introduced before enterprises began using artificial intelligence (AI). Speaking to Computer Weekly during the company’s CamundaCon conference in Amsterdam, he says that although they have enabled business to get to where they are today, they are missing an opportunity. “They were designed for a world without AI, and now that AI is here and it's omnipresent, we can leverage it to be a lot more effective,  and a lot more efficient.”

This, according to Freund, is not about injecting AI here and there into an existing business process like the order to cash process. He says: “We have to rethink how that process really works structurally so that we can then fully leverage this magnitude of change in productivity with AI.”

Looking at the quote for cash process at Camunda, he says: “In the age of AI, we see the opportunity to make that process a whole lot more effective and efficient so we can scale the business further without needing to grow the headcount.”

Freund uses the term agentic AI orchestration to describe how it becomes possible to drive work across multiple steps, automatically much more than before. “We can now actually also automate the kind of work that required human judgment,” he says. While people were previously needed to decide what should happen in a certain process, a large language model (LLM) is now used to make those decisions. “Agenda orchestration allows us now to do LLM-driven work control paired with programmatic or deterministic workflows, which then makes sure that in situations where we cannot just rely on the LLM's judgment,  we have laid out and defined how work is supposed to happen under certain circumstances.”

This enables business processes to be driven by AI with guardrails and governance through deterministic workflow driven orchestration.

At Camunda,the account executives who are responsible for customers manage their customers through Slack rather than having to log into enterprise systems like Salesforce for customer relationship management and the NetSuite ERP system. This changes how organisations consider the systems of records that holds customer data and transactions.

Freund says: “We don't want to be the system of record. We want to be the system of process. These are two different things. An end-to-end process usually spans across multiple systems of record.  So if you tie yourself into a certain ERP system like SAP or CRM system like Salesforce, then you’re always containing yourself in a certain box. However, almost any enterprise out there is using more than just one system of record, so it makes sense for them to have an independent orchestration layer that runs agentic and orchestrated end-to-end processes across those systems of record.”

He says enterprise software providers like Salesforce have recognised this shift in how their products will need to be used. “Salesforce announced it was going ‘headless’ and exposing its functions through MCP (Model Context Protocol),” Freud says. “Increasingly, those functions are consumed not so much by humans, but by agents which perform work automatically.”

And if AI agents rather than people are interacting with enterprise IT systems, this shifts how software will be licenced going forward. It shifts from per user to some form of indirect access and if Freund is right, the balance of power then shifts to the system of process.