The Automation Ceiling: Why AI Isn't Delivering Enterprise Results (And How To Fix It)
Jakob Freund is the CEO of Camunda, a software company innovating end-to-end process orchestration and automation with agentic AI.
gettyMost enterprise leaders get excited when an AI pilot works. The demo impresses the board, but then the project stalls somewhere between the proof of concept and the balance sheet. The promised transformation never quite arrives.
Camunda’s own research named this problem “the automation ceiling.” Enterprises are adopting AI agents faster than any technology in recent memory, yet the results stay flat. Our 2026 State of Agentic Orchestration and Automation Report found that 71% of organizations already use AI agents, but only 11% of those use cases reached production last year. Many are puzzled by the reason, since enterprises have more powerful models and more capable agents than ever. The real issue involves how work gets done inside the enterprise, not so much the agent itself.
Most enterprises built their technology stack one function at a time. The claims team got a claims system. The procurement team got a procurement system. Finance got an ERP. Each of these systems automates its own slice of the business well, but none of them owns the process as a whole. A customer request, insurance claim or loan application still has to travel across several of these systems, and someone or something has to carry it from one handoff to the next.
For years, that someone was a human being. Now companies are trying to hand that job to an AI agent, and often to several agents at once. A hospital might deploy an intake agent, a scheduling agent, a documentation agent and an imaging assessment agent—with each one genuinely excellent at its narrow task. The intake agent works faster. The scheduling agent optimizes appointment slots. The documentation agent cuts paperwork time. Individually, each agent looks like a win.
Yet collectively, the problems pile up. The intake agent captures information that the scheduling agent never sees. The imaging agent finishes its analysis without telling the assessment agent it’s done. Each agent optimizes its own task while the patient’s actual journey through the hospital stays just as fragmented as it was before. Even though leaders add AI agents to automation, they’re still seeing problems multiply.
This is the part enterprise leaders find counterintuitive. Adding intelligence to a broken process does not fix the process. It accelerates whatever was already happening inside it. If handoffs were fragile before, agents make those handoffs fail faster and in more places at once. If nobody had a full view of the process before, nobody has a full view of it now. In fact, the view is even harder to reconstruct because it is scattered across a dozen agent logs instead of a dozen human inboxes.
Adding yet another agent creates a new version of the same problem. Now there is one more system without a shared, structured view of the process, tasked with an almost impossible job: getting independent AI systems to reliably hand off work to each other without any framework governing how that handoff happens.
The uncomfortable truth is that the ceiling was never really about AI. Enterprises hit the same wall years ago with robotic process automation, and before that with point-to-point system integrations. Each wave of automation technology got layered on top of processes nobody had redesigned. AI is simply revealing the structural gap faster and at greater scale than previous technologies did.
Applying AI to an unchanged, legacy process tends to deliver modest gains, which are far short of what leaders were promised when they approved the AI budget. Getting past the ceiling requires a shift in how leaders think about their overall processes and how AI should be involved.
1. Understand the process as it runs today. Most enterprises discover significant gaps between their documented process and their actual one once they look closely, and any redesign built on the wrong starting point will fail regardless of how sophisticated the AI is.
2. Design the target process around the outcome you want. Ask what the process would look like if you were building from scratch, with agents and automation available from the start, rather than asking where you can insert an agent into the process you already have.
3. Build a single layer of visibility across every system and agent involved in the process before you scale. Leaders need a real-time view of every case in flight, a record of which agent or person handled each step and a complete account of what happened and why, especially in regulated industries where auditability is key. Without that shared view, AI initiatives stay siloed pilots with no way to prove their business impact.
4. Treat the process as something that keeps improving. The process should be monitored continuously against the outcomes that matter, whether that is cycle time, cost per case, error rate or compliance, with changes tested and rolled out based on what moves those numbers rather than on a quarterly review cycle.
The enterprises that make it through the ceiling share a common trait. They stop asking how to make individual tasks faster and start asking how to redesign the process those tasks sit inside. That shift in question, more than any single tool or model, determines whether AI investment turns into pilot fatigue or into a genuinely re-engineered, AI-first business.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
