Enterprises Explore A New Orchestration Framework For Success With Agentic AI

Enterprise AI is a watershed moment. On the surface, a frantic race to adopt generative AI is underway. Just beneath the hype, however, operational chaos is brewing. In this "pilot purgatory," most organizations wrestle with security and governance issues, and few AI projects ever reach production. Now, instead of model intelligence, leaders face a new challenge: orchestration.

Some see the current AI boom as entirely new. But Matan-Paul Shetrit, Director of Product Management at Writer, sees familiar patterns. An accomplished leader with over 16 years of experience building platforms at companies including Brex, Flexport, and Square, he views the chaos of AI adoption through an expert lens. In a market with no agreed-upon metrics for success, this deep, enterprise-first DNA is what gives him and his team the "privilege to be very opinionated on what is good," he said.

From Shetrit's position, the primary purpose of AI should be to augment the systems that make a business successful in the first place. By mapping what he called "orchestration and judgment graphs," organizations can create an explicit blueprint of their hidden operating systems. It's these unwritten rules and informal power structures that dictate how work actually gets done, he continued.

• Make good better: But AI isn't a silver bullet, nor is it meant to replace teams, Shetrit clarified. "AI's primary purpose is to make what your enterprise does really well even better by learning from how you do things and how you got to where you are." Still, most teams approach the problem from a technical perspective when the real value lies in a less obvious solution.

• Hidden in plain sight: The orchestration graph already exists, he explained. It's just hidden in plain sight. "If you want AI to be effective, you have to map that hidden organizational knowledge and tribal knowledge, so agents know how to navigate your organization effectively." Instead of hyper-focusing on the org chart itself, his approach is about discovering new forms of responsibility, accountability, risk management, governance, and more.

Success with AI is more about the human element than technical capacity, according to Shetrit. Beyond processes and tools, the real leverage comes from tapping into the invisible information systems already embedded across traditional hierarchies.

• Orchestration ≠ flat org: Enterprise decision-making tends to be more challenging when organizations are "flat," he explained. "Organizations that aren't transparent about the decision-making process create teams and accountability structures that are frustratingly fragile." Decision-making still happens, he pointed out. But without the ability to trace who decided to implement what and why, the process lacks transparency.

• Military leadership: In the military, service members learn regulations through incidents, Shetrit continued. The enterprise certainly isn't a battlefield, but a regimented approach can be practical here as well. "Otherwise, employees and leadership will forget why a regulation or guideline was implemented. They'll assume it's cumbersome and remove it, only for the same negative result to occur again."

But making these hidden systems explicit and visible is a task that extends far beyond workflow optimization, Shetrit said. When organizations automate low-stakes tasks, it affects employees across the enterprise, elevating them into a new class of decision-makers.

• Your new job title: For enterprises to orchestrate AI across agent-agent workflows and human-agent workflows effectively, they must learn how to manage non-human co-workers first, Shetrit said. "We're all going to become managers. Managers of humans, managers of agents, and managers of hybrid teams. The decisions humans make will have a much higher impact as lower-stakes tasks are outsourced to autonomous agents."

Such a framework is necessary to support the economic claims at the heart of Shetrit's argument. As AI drives the marginal cost of creating work to near zero, the primary bottleneck for growth moves from creation to governance, he explained. Without a new management framework, companies face a "scaling cliff," similar to the historical freight forwarders who couldn't grow beyond a few hundred thousand employees.

• Cliffside scaling: Most enterprises today don't see agents as increased headcount, Shetrit explained. Unfortunately, that could create a host of challenges. "Hypothetically speaking, let's say you're a company of 1000 people. And you decide to deploy 100,000 agents. Now, you're no longer a company of 1,000. You're a company with over 100,000 employees." Some enterprises already hit the same roadblocks they would if they tried to scale their company with humans, he said. "Clearly, we need to rethink how we run and manage our business."

True enterprise AI won't come in the form of another static tool, Shetrit concluded. Instead, it will emerge as a dynamic partnership that evolves with the business. As more teams adopt AI, the orchestration graph can be a catalyst for that adaptation, he said. Without a standardized approach for human-agent interactions and orchestration, how organizations decide to move in this moment will set their course for the future. "The enterprises that will thrive in this new era of competition, tech, and talent will be those that don't treat their internal organization as static. If it's static, you're dead."