.jpg "The Great Process Re-Engineering: Who’s Running the Agents?")
The Great Process Re-Engineering: Who’s Running the Agents?
AI agents are moving beyond experimentation and into critical business processes. But with more agents making decisions and taking action, a more urgent question is emerging: who – or what – is actually in control?
For Camunda’s regional vice-president APAC, Jared McMahon, the challenge isn’t simply managing more AI.
Instead, he said it’s ensuring organisations can see how agents, automation tools, people and systems interact across an end-to-end process, and understand the decisions being made along the way.
“Humans don’t necessarily automatically trust what they can’t see,” McMahon told iTnews. “If those AI agents or automation tools are doing these activities and tasks in isolation, you can’t trust what you can’t see.”
Indeed, it’s a challenge becoming more acute as AI moves from personal productivity and isolated use cases into business processes involving multiple systems, decisions and participants.
Camunda’s 2026 AI Process Gap research, conducted by Sapio Research among 1,000 senior IT, operations and transformation decision-makers at large organisations in the US and Europe, found that enterprises have an average of more than 50 endpoints within their business processes.
“Adding AI as an active participant introduces another layer of complexity into environments that were already highly interconnected,” he said.
While individual AI agents and automation tools can be highly effective at specific tasks (answering a question, responding to a query or processing a payment), McMahon said complex enterprise outcomes rarely depend on a single task.
Instead, they involve a series of activities and decisions across an end-to-end process.
“The question then becomes: who’s orchestrating the outcome for the organisation, and what is the connective tissue across that end-to-end journey?” he said.
Visibility by design
That question becomes particularly important as AI is given greater autonomy.
For organisations operating in regulated environments, knowing what decisions are being made, how they were reached and where human oversight sits isn’t simply an AI governance aspiration.
McMahon pointed to APRA’s CPS 230 requirements around board-level oversight of critical operations and processes as an example of the broader accountability facing Australian organisations.
Without an end-to-end foundation connecting processes across the organisation, he said technology leaders may struggle to provide the necessary visibility when questions come from executives, boards or regulators.
Camunda’s position is that auditability should therefore be built into the process rather than reconstructed after something goes wrong.
“The end-to-end process should be visible by design,” McMahon said. “It should be human readable, not just machine readable.”
That includes maintaining a clear audit trail of the activities, calls and decisions leading to an outcome: information he said should be available “at the click of a button” rather than requiring a weeks-long forensic investigation.
Where should humans step in?
But greater orchestration doesn’t necessarily mean handing every decision to AI.
McMahon said the appropriate level of human involvement will vary between organisations and processes, but one consideration provides a useful starting point: the cost of getting a decision wrong.
Where that cost is high, human judgement remains particularly important.
Increasingly, Camunda is seeing organisations combine deterministic workflows (where the required steps are known in advance) with more dynamic workflows involving AI and judgement-heavy tasks.
In those cases, AI can operate within defined entry and exit criteria, with humans providing oversight where appropriate.
From four days to 17 minutes
New Zealand non-bank lender Harmoney offers one example of the two approaches operating together.
Harmoney has used Camunda to orchestrate its end-to-end credit decisioning journey, including identity verification, income assessment and credit decisions.
Some applicants, including people with multiple jobs or gig-economy income, can fall outside conventional income-assessment processes.
When that happens, an AI component called BankBot can take over communication and income verification, with human oversight available when required, while remaining part of the same orchestrated end-to-end process.
For the subset of loan applications using the approach, McMahon said approval times have fallen from four days to 17 minutes.
The model also allows the lender to scale without increasing headcount at the same trajectory while delivering faster decisions to customers.
Building trust before increasing autonomy
Another Camunda customer, a telecommunications provider in Singapore, is taking a deliberately staged approach to AI autonomy.
Within a customer support process, multiple AI agents make recommendations to a human agent after the customer has been verified. For now, those recommendations still go to a person for approval.
But that isn’t intended to be the permanent state.
“By tracking the agents’ recommendations and outcomes over time, the organisation can build evidence about where the technology can be trusted. Certain categories of decisions could then move to greater autonomy, while others continue to require human oversight,” McMahon said.
The important element, he said, is that the underlying orchestration maintains an audit trail showing what decisions were made and the reasoning behind them.
The Great Process Re-Engineering
The bigger opportunity may ultimately go beyond orchestrating the processes enterprises already have.
McMahon argues that most existing business processes were designed for a world in which humans managed them from end to end. Simply layering AI over those processes risks using fundamentally new technology to do essentially the same work slightly differently.
“Today’s processes were built for humans to manage end-to-end, and we’re now layering in AI over the top of that and asking them to do the same thing, but with a slightly better outcome,” he said.
“We think there’s a better path forward than that.”
Camunda describes that shift as The Great Process Re-Engineering: reconsidering how enterprise processes themselves should work as AI becomes capable of reasoning, deciding and acting within them.
Agentic orchestration provides what McMahon describes as the “shared frequency” connecting those environments, coordinating AI agents, automation tools and disparate systems while providing structure around where decisions are made, how they are escalated and how they can be audited.
Watch Part 1 of The Great Process Re-Engineering: Who’s Running the Agents? above. In Part 2, iTnews explores an even bigger question with Camunda: Is Every Process Now Legacy?
Camunda is the leader in agentic orchestration, automating complex business processes, including high-value knowledge work, across agents, people, and systems. By creating production-ready, enterprise-grade agents with built-in governance, Camunda uniquely delivers trusted AI agents for business-critical processes. Over 700 leading innovators like Atlassian, ING, and Vodafone, rely on Camunda to slash time-to-value from months to days, boost operational efficiency, and elevate customer experiences. Ready to become an AI-first enterprise? Visit camunda.com.
About Author
