Workflow and AI Deployment: Re-Engineer the Process Before Agents Run #AIg

On October 13, 2026, Camunda moved ProcessOS into early access and shipped Camunda 8.10. The company says early ProcessOS pilots compressed process re-engineering by 96 percent, and that enterprises are discovering that re-engineering the process is how AI earns the trust to run it. A week earlier, ServiceNow made AI Workflow Factory globally available, framing workflow improvement as a continuous loop under AI Control Tower. Both releases treat the same failure: agents stall in pilots because the process underneath wasn’t built for AI.

Principal Basil C. Puglisi reads that stall as a governance design problem rather than a model problem. His paper Human-AI Collaboration Needs More Than a Human in the Loop defines the collaboration as human direction, AI execution, and a named human who decides at a checkpoint and answers for the result. Checkpoint-Based Governance (CBG) puts the named reviewer in the path with binding authority to accept, modify, or reject before the agent gets production authority over money, customers, or regulated outcomes.

What Camunda 8.10 and ProcessOS put on the process itself

Camunda’s October 13 press release reports a Nordic bank case in which ProcessOS compressed work that traditionally takes 9 to 12 months into just over a week. That cut is 96 percent faster and 94 percent cheaper per process. Across about 200 processes, the company says the program points to roughly $26 million in savings. Combined ARR from customers with $1 million or more in ARR expanded 87 percent, and the count of those customers grew 67 percent. Regulated growth led the way, with insurance up 53 percent, public sector up 48 percent, and banking up 40 percent.

Camunda’s Process Gap research finds that 78 percent of process leaders say re-engineering is essential to staying competitive. The Camunda 8.10 product blog organizes the release under Speed, Trust, and Scale. Deployed processes can become callable agent tools over the Model Context Protocol, and they keep their governance and audit trails intact. Decision trails and an agentic control plane show how agents behave over time. Before an agent moves funds or sends a customer a final decision, the process can require a policy check or a human sign-off. The blog’s line is direct: the agent reasons, and the process decides what it’s allowed to do.

What ServiceNow’s AI Workflow Factory adds as a continuous loop

ServiceNow’s October 6 announcement describes AI Workflow Factory as a continuous loop under AI Control Tower. Process Mining spots which processes should change, Autonomous Engineer and Build Agent help teams build and check the improvements, and App Engine runs the updated workflows at scale. People set the direction and approve the outcomes while AI moves more of the work forward, and Action Fabric extends the same governed loop to third-party agents. AI Workflow Factory is globally available, with Autonomous Engineer Early Access available on request.

Puglisi treats both vendors as evidence that the market is moving the decision surface into the process, not as interchangeable products. Camunda emphasizes re-engineering the path an agent inherits, while ServiceNow emphasizes a continuous improvement loop with people approving outcomes. The shared claim is that bolting an agent onto yesterday’s handoffs doesn’t put the agent in charge of the outcome, because the handoffs still are.

Why Puglisi puts CBG checkpoints in the process path

Human-AI Collaboration, as Puglisi defines it, is not presence near a dashboard, and a person watching AI output isn’t the same as a person deciding it. Responsible AI, in his vocabulary, is machine checking machine with no human review. AI Governance begins when a qualified named human holds binding authority at specific checkpoints, with personal accountability for the outputs that pass through.

Applied to these launches, the first checkpoint sits where the process is redesigned. Whoever decides which steps become agent tools, which MCP-callable capabilities stay off-limits, and which outcomes need a human sign-off is already governing. Puglisi’s advice is to put a name, a written rationale, and a review date on those design choices so the rule the agent later follows is not tribal knowledge.

The second checkpoint sits where an agent action becomes consequential. That covers moving funds, issuing a final customer decision, changing a case routing rule, or publishing a workflow that other units will inherit. Those actions belong with a named reviewer who can accept, modify, or reject on the record. Decision trails and control towers make that review possible, yet they can’t replace the person who answers for the result.

What to measure once the process starts deciding for agents

Teams that adopt ProcessOS-style re-engineering or an AI Workflow Factory loop should measure the process design as carefully as they measure agent throughput. Track how many production agent actions require a human checkpoint and how often reviewers modify or reject rather than accept. Also track how often the decision trail supports the claim the agent made. A process that never slows an agent for a named decision may be efficient, and it may also be Responsible AI wearing a governance label.

Camunda and ServiceNow both advertise human sign-off and outcome approval inside the path. That matches the Named-Human Test in Puglisi’s Human-AI Collaboration paper: is there a named human who can accept, modify, or reject, with accountability that survives an audit? If the answer is only a prompt queue nobody reads, the process still decides what the agent may do, and it decided to give that authority away.

What to watch as process platforms absorb agent authority

Watch whether early ProcessOS customers publish independent numbers that match Camunda’s pilot claims outside the vendor case. Watch how ServiceNow customers use AI Control Tower when Action Fabric brings third-party agents into the same loop, because that’s where governance either holds across vendors or fractures by connector. Watch whether MCP-callable processes keep human sign-off when the agent is faster than the reviewer queue.

Agents will keep getting cheaper to run. The process that decides what they may do is still the expensive part, and it is the part that makes the difference between a pilot that demos well and a production path someone can defend.

Sources

• Camunda. (2026, October 13). Camunda’s Great Process Re-Engineering accelerates agentic orchestration adoption [Press release]. https://camunda.com/press-releases/camundas-great-process-re-engineering-accelerates-agentic-orchestration-adoption/
• Camunda Product Team. (2026, October 13). Camunda 8.10: The Great Process Re-Engineering starts now. Camunda. https://camunda.com/blog/2026/10/camunda-8-10-the-great-process-re-engineering-starts-now/
• ServiceNow. (2026, October 6). ServiceNow launches AI Workflow Factory to turn workflow improvement into a continuous, agentic AI-powered loop [Press release]. https://newsroom.servicenow.com/press-releases/details/2026/ServiceNow-launches-AI-Workflow-Factory-to-turn-workflow-improvement-into-a-continuous-agentic-AI-powered-loop/
• Puglisi, B. C. (2026, September 29). Human-AI Collaboration needs more than a human in the loop. basilpuglisi.com. https://basilpuglisi.com/human-ai-collaboration/

Questions readers ask

What did Camunda announce on October 13, 2026?

Camunda launched Camunda 8.10 and moved ProcessOS from closed beta into early access. The company says early ProcessOS pilots compressed process re-engineering by 96 percent, including a Nordic bank case that cut 9 to 12 months of work to just over a week, and reports strong growth among its largest customers in insurance, public sector, and banking.

What is ServiceNow AI Workflow Factory?

AI Workflow Factory is ServiceNow’s continuous improvement loop that connects Process Mining, Autonomous Engineer and Build Agent, and App Engine, governed by AI Control Tower. People set direction and approve outcomes while AI moves more of the work. It became globally available on October 6, 2026, with Autonomous Engineer Early Access available on request.

Why do AI agent pilots stall according to these vendors?

Camunda argues that enterprise AI pilots often stall not because the technology fails, but because underlying processes were built for a world without AI. ServiceNow frames the same gap as fragmented legacy infrastructure and one-off transformation projects. Both releases treat process redesign, not a smarter model alone, as the path from pilot to production.

How does Basil C. Puglisi define Human-AI Collaboration here?

Puglisi defines Human-AI Collaboration as human direction, AI execution, and a named human who decides at a checkpoint and answers for the result. Checkpoint-Based Governance requires binding authority to accept, modify, or reject, plus a decision record. Presence without that authority stays in Responsible AI, where machines check machines.

Where should checkpoints sit in a re-engineered process?

Place one checkpoint where process designers decide which steps become agent tools and which outcomes need human sign-off, with a named owner and review date. Place another where an agent action becomes consequential, such as moving funds or issuing a final customer decision, so a named reviewer accepts, modifies, or rejects on the record before production runs.

#AIg