“We assume we know how our businesses run, but the reality is often messier than a teenager’s bedroom,” observed Maxime Vermeir, Vice President of AI Strategy at ABBYY, to UC Today, in a striking indictment of the current state of enterprise readiness for the agentic AI and automation revolution.
We are standing at the precipice of a momentous transition from copilots, assistants that wait for a prompt, to autonomous agents that can reason, plan, and execute. The vendor ecosystem is marketing this as the arrival of the “autonomous enterprise,” a frictionless future where digital workers handle the drudgery while humans reap the rewards.
However, the view from the ground is far more complex. Knowledge work is rarely a linear assembly line. It is a tangle of undocumented process steps, tribal knowledge, exceptions, and judgment calls that do not translate cleanly into binary code. Even when AI saves time, those gains often manifest in fragmented shards that do not automatically convert into productivity. Whether in enterprises, SMBs, or contact centers, “more efficiency” can quickly metastasize into “more load,” creating an AI productivity paradox that fuels burnout rather than relieving it.
The winners in 2026 and beyond will be those organizations that treat digital labor as a rigorous operations discipline. They will define scope, instrument quality, and redesign work rather than bolt agents onto broken processes.
The “Tribal Knowledge” Trap With Automation
The primary friction point for agentic workflows is not the models’ capabilities, but the opacity of the work they are asked to perform. Corporations are built on “Standard Operating Procedures” (SOPs) that often bear little resemblance to how work actually gets done. When an autonomous agent attempts to navigate these undocumented waters, it crashes against the rocks of human intuition.
Vermeir recalled a deployment for a large European financial institution aimed at automating a loan approval process. On paper, the SOP was clear. “But once we started implementation, huge gaps came to light,” Vermeir said. “We found missing data, systems refusing to approve valid requests, and customers waiting indefinitely in digital limbo. The so-called ‘Standard Operating Procedure’ truly depended on the human intuition and understanding, nuances that their AI automation plan hadn’t accounted for.”
The lesson was expensive but necessary. If you cannot see the reality of the process, including the invisible roadblocks that only humans know how to bypass, automation will fail.
This disconnect is not limited to complex financial instruments. It also plagues routine IT operations. Jeremy Rafuse, Head of Digital Workplace at GoTo, suggested to UC Today that patch management is a classic example where the absence of “tribal knowledge” can cause chaos:
“If important workflow exceptions, like month-end close in Finance or end-of-quarter for sales, aren’t documented or well-known, automation can disrupt critical business moments. We have learned firsthand that missing these blackout windows meant automated reboots happened at exactly the wrong time, affecting productivity and performance.”
The implication for IT and operations leaders is that the pre-work for agentic AI is anthropological rather than technical. You must capture the “shadow process” before you can automate it. Rafuse noted that “successful automation depends on capturing and utilizing the human knowledge that guides key operations.” Without this, companies are automating their own confusion, scaling errors at the speed of silicon.
The Productivity Mirage With Agentic AI: Redefining ROI
For years, the industry has measured the success of automation in “hours saved.” It is a metric that appeals to the CFO, but it is often a mirage. Saving an employee five minutes ten times a day does not necessarily result in fifty minutes of new value creation. It usually results in a fractured workflow, where the employee is constantly context-switching and unable to enter a state of deep work.
Shawn Spooner, CTO at billups, outlined to UC Today that we must look beyond raw throughput to the quality of the workflow:
“The difference between theoretical ROI and actual bottom-line impact comes down to this: are you fragmenting work into AI-assisted micro-tasks, or are you redesigning entire workflows so humans stay in their ‘zone of genius’ while AI handles the mechanical execution?”
Spooner offers a compelling case study from his own organization. Before integrating AI, the billups analytics team could build two custom targeting maps per person per day. “We can now build 14 per person, per day, giving us a seven-fold increase in capacity without adding a single person,” Spooner said. “That’s not theoretical productivity. That’s seven times more client deliverables, seven times more strategic conversations, seven times more revenue opportunity from the same team.”
The key to this success was not simply speeding up a task, but automating a contiguous block of work. Spooner emphasized that the goal is “preserving a flow state so our people can stay in the creative or strategic work without constant context-switching, rather than to simply save five minutes here and there.”
This sentiment is echoed by Vermeir, who dismissed the standard metric of time savings entirely. “Time saved is the vanilla ice cream of KPIs; it is fine, but it is not what keeps executives up at night,” Vermeir argued. He suggests that true ROI lies in precision and risk reduction, citing a frozen goods manufacturer that used AI to cut customs clearance times from over an hour to 5 minutes. “Hours saved may increase efficiency over time, but the real value lies in precision and reliability,” he added.
Rafuse added that while productivity remains the dominant metric, noting that 72 percent of IT leaders in GoTo’s recent survey measure it, sophisticated buyers are looking beyond it. “55 percent say they also measure both improved customer satisfaction or retention and increased revenue,” Rafuse explained. The consensus is that if your business case relies solely on shaving minutes off a timesheet, you are likely missing the transformative potential of the technology.
The Automation Operations Playbook: Unbundling and Governance
If we accept that digital labor will not simply replace jobs but unbundle them, the challenge becomes one of management and governance. How do we design the “handoff ” between silicon and carbon? The most successful deployments treat digital labor as an operations challenge, requiring a fundamental redesign of roles.
Spooner advocated a “Wizard of Oz” approach to prototyping, in which a human simulates the agent’s role to test the workflow design before any code is written. “Review the prototype with the people who’ll use it. Let them tear it apart,” Spooner advised. “They’ll find the edge cases you missed, the context you didn’t understand, and the reason step seven can’t happen before step five.” This human-centric approach ensures that you are not just “automating your assumptions.”
Once the agents are live, the “Golden Rule” is context. We are moving from a world of deterministic software to probabilistic “units of production,” and that requires a new safety net. “The goal isn’t zero errors. It’s catching errors before they compound,” elaborated Spooner. He warns against bolting on human review as an afterthought, which triggers alerts that “yank someone out of deep work to debug an agent’s confusion.” Instead, the handoff must be workflow-native, a designed pause point where review happens naturally.
Vermeir reinforced this, using a relay race analogy:
“An agent that hands off a task without context is like a relay racer throwing the baton at a spectator instead of the next runner. We need the right person, with the right context, at the right time. It is about augmenting human intelligence, not replacing it with confusion.”
This “human in the loop” strategy is about training the system, not catching mistakes. Rafuse emphasized that “human employees should be working closely with agentic AI purpose-built for tasks. By confirming the accuracy of all AI agents’ outputs, human employees further help to both eliminate potentially serious errors and refine the AI models.”
Deciphering the True Value of Agentic AI as Digital Labor
The trajectory for digital labor in 2026 is trending up, but it is not the straight line that vendors promise. We are entering a period of disillusionment where the “magic” of AI clashes with the friction of enterprise reality. The organizations that succeed will be those that stop waiting for a miracle cure and start doing the hard work of operations. This includes mapping their processes, unbundling their jobs, and governing their digital workforce with the same rigor they apply to their human one.
As Vermeir astutely noted, the failure of AI initiatives is rarely technological. “AI is not rejected because leaders do not understand. It is rejected because nobody translated it into the language of business,” he said. “We need to stop talking like engineers and start talking like strategists.”
The future of work is about elevating humans, rather than replacing them. However, that elevation requires a solid foundation, one built on documented processes, meaningful metrics, and a relentless focus on the flow of value.
