Every company is working hard to build a software workforce. Few people know how to manage it. How do you manage, audit, measure, and adjust your workforce when it becomes an artificial intelligence agent? Most companies don’t. The problem is too new.
So Karen Webster asked Zafin CEO Charbel Safadi directly. Will generative AI rewrite enterprise operating models? Is governance the bottleneck that will determine the winners?
Mr. Safadi went beyond the premise. The launch of Zafin AIOS, which he announced in this PYMNTS exclusive, is not about agents at all.
“It’s not just the agents,” he said. “That’s the way it works.”
His argument is to stop treating AI as a bunch of tools bolted into workflows built for humans. These workflows are designed around human decisions and manual handoffs. Introducing autonomous systems there reduces productivity and creates chaos. What we need is a layer that sits on top of all of that. Safadi reaches for an apt analogy: airports.
“What AIOS is trying to solve is to become an airport for planes,” he said.
Advertisement: SCROLL TO CONTINUE
The airport does not fly planes. They route, monitor and manage thousands of movements and make sure nothing collides. Companies now need the same for AI systems running simultaneously across departments. Zafin first built AIOS in-house as Customer Zero and used it to speed up its own product development and research before sales. Lessons quickly surfaced.
The biggest barriers to AI adoption are cultural, not technical.
The next problem with AI is not intelligence. It’s management.
Factories used to drop robots onto lines designed for human hands. The robots could not benefit until the line itself was redesigned. Companies are repeating the same patterns with AI, and the lessons are here to stay. Productivity gains will never occur unless you redesign your operating system to accommodate your new employees.
“We need to rethink the way we work, and that requires a cultural transformation with AI,” Safadi said. “Many organizations think of this solution as adding tools to human capabilities, but not reinventing the way work is actually done across the enterprise.”
Deploying AI in enough places will have the opposite effect on efficiency. Teams spend their time managing a patchwork of apps, agents, and workflows rather than running a single system. That’s why many banks have invested large sums of money into AI, and have tread carefully. They buy tools and run pilots. Few companies have redesigned their end-to-end operating models around autonomous systems.
“What we are promoting is a new operating structure for the organization,” Safadi said.
On the other side of that structure is what he calls “The Company of Jesus.” An organization that can test ideas, prototype products, and respond to customers without hitting the usual resource walls. Because once autonomous systems are able to explore, design, write code, perform analysis, and perform real-world workflows, the old scaffolding begins to creaky. Approval chains, budget cycles, functional silos, and reporting lines cease to be safeguards and begin to become bottlenecks.
Clearing the governance bar unlocks competitive advantage
If AI changes the way work is done, the way products are manufactured, and the workforce is deployed, this is not a technology sourcing problem, but an enterprise design problem. And, especially in regulated industries such as financial services, enterprise design is directly related to governance and compliance. Banks need to satisfy regulators, auditors and risk managers who need to see how decisions were made and why they occurred.
Safadi said the same demands surface every time he has conversations with bank boards, CEOs and CIOs.
“They all said we need a control plane,” he said. “We need the ability to organize structures, look at efficiency, look at productivity, look at routing, look at cost and ultimately make sure we can meet regulatory requirements.”
That control plane is the heart of AIOS. Zafin doesn’t assume anyone will standardize on a single AI provider. It assumes the opposite. There is a mix of cloud models, open source models, purchased agents, and homegrown agents, all running at the same time. Its job is to make that mix behave like one coherent system through orchestration that runs end-to-end.
This leads to a question Webster repeated over and over again. When autonomous agents make decisions and begin work, whose head is at stake?
The proof of work becomes the moat.
Safadi’s answer is to shift governance from cost to function. As AI penetrates deeper into operations, transparency itself becomes an asset. Being able to show how the work was done starts to become as important as the work itself.
Here’s how it works: “Humans put compliance artifacts in AIOS, they put knowledge artifacts in AIOS, they put guardrails in AIOS, they put policies, restrictions, standards in AIOS,” Safadi said. Agents then operate within those constraints while another compliance agent checks their work against the rules in real time.
He calls the deliverables “proof of work.” All agents leave a record of their actions, decisions, validations, and compliance checks throughout the workflow, and all outputs are tied into an auditable set of activities.
“My argument here is that this is even more secure than the way we operate today,” Safadi said. “Most people don’t document why they choose to do something a certain way.”
And the line on which all the banks we’re watching are structuring their bets is, “Banks that can move quickly will compete. Banks that can’t will effectively become public utilities.”
That’s real change. AI doesn’t just create new products. We are building a new operating model, giving an advantage to those who can redesign, manage, and execute work the fastest.
Or, as Safadi puts it, a new operating structure for the organization.
