NEW YORK — AI is pushing companies to hand more everyday operations to software, Forbes contributor and tech executive Sanjay Srivastava wrote in a Feb. 16 commentary.
Srivastava said the biggest near-term change “will not be ‘better content.’” Instead, he wrote, it will be “the conversion of operational work from human-executed to AI-executed, with people supervising only the edge cases.”
What agentic processing is
Agentic processing refers to software that can take “an operational case, decide what should happen next, and then execute the work across systems using tools and workflows.” Srivastava said it can “gather missing information, apply policy, route approvals, perform transactions,” and then “escalate to a human when confidence is low, or risk is high.”
He stressed this is not a chatbot and not simple automation. “The defining shift is closed-loop execution,” he wrote. In plain terms, the system doesn’t just suggest what to do; it does the steps, checks the result, and asks for help when it’s unsure, all while keeping records and following rules.
Why this is happening now
Srivastava said this is becoming practical because “the stack has matured enough to operate inside enterprise constraints.” He wrote that the tools to test, monitor and manage these systems have improved, and companies have more ways for software to connect to their systems.
“In short“we’ve moved from interesting demos to systems that can plausibly carry production work under supervision.”
He also argued that past automation left a lot of work behind. “Most digital transformation has automated the ‘happy path’ while leaving the messy middle to people,” he wrote. Over time, he added, “exceptions expand, handoffs multiply, and unit economics hit a ceiling because the operating model is still ‘more volume = more labor.’”
Where companies may start
Srivastava wrote that early wins tend to come from work that is frequent and guided by clear rules. He pointed to cases where people spend too much time “reconciling issues” and chasing missing information. Examples he listed include claims intake and triage, AP/AR exception handling, order-to-cash disputes and follow-ups, KYC refresh workflows, benefits enrollment changes, and field-service triage and routing.
Build it or buy it
Srivastava said some companies should build these systems themselves, especially if the work is a competitive advantage or involves sensitive data and high risk. He wrote that building can be strategic when a process is “differentiating, tightly coupled to proprietary data, or central to how the business wins,” and when “regulatory exposure, brand risk, or data sensitivity demands maximum control.”
Others may choose to buy a managed service.
“The hard part is often not the model,” Srivastava noted. “It is running the full production system: instrumentation, exception operations, human takeover, reliability, and continuous improvement.” For companies that want speed or lack capacity, he said, buying outcomes can make sense.
A new kind of outsourcing
Srivastava said the market is moving toward three supplier paths. Including a new category he expects to grow: Agentic Process Outsourcing (APO).
“APO is different because it starts AI-first,” he wrote. In APO, he said, “agents are the throughput engine,” while “humans are designed into control and escalation.”
This means a type of outsourcing where the provider is judged on results delivered by software. And the provider is responsible for keeping the system working when things change.
How leaders can move ahead
Srivastava urged leaders to start with a small, high-friction piece of work and prove it can run safely before scaling. He wrote he would “require an evaluation harness before scaling anything,” then “run a controlled production pilot with explicit guardrails and human takeover,” expanding only when “the economics and controls hold.”