Dominic Jainy stands at the forefront of the next great shift in enterprise automation. With a career spanning the rise of machine learning and blockchain, he has spent years helping organizations navigate the complexities of digital transformation. Having witnessed the initial surge of Robotic Process Automation (RPA) over a decade ago, Dominic now argues that we are entering a post-script era where AI agents don’t just mimic clicks, but actually understand the work they perform.
In this conversation, we explore the transition from rigid, pattern-based automation to goal-driven agents. Dominic explains the mechanics of “Skills” and “Computer Use,” the challenges of dismantling legacy RPA estates, and why the C-suite must prepare for a future where software can reason, decide, and act autonomously across existing corporate systems.
RPA scripts often break when software interfaces or input formats change. How should organizations identify which legacy scripts are now obsolete, and what specific technical hurdles arise when transitioning from these rigid, pattern-based scripts to flexible, goal-driven agents?
Identifying obsolete scripts begins with a cold look at your maintenance logs and escalation rates. When you see a bot that requires constant intervention because a vendor changed a PDF layout or an ERP updated its UI, you are looking at a prime candidate for retirement. The technical hurdle in moving to goal-driven agents is shifting from “how” to “what.” In the old world, we spent hundreds of hours hard-coding every single mouse movement and keystroke, creating a brittle chain of commands. To transition, teams must learn to define “what good looks like” by providing high-level objectives and constraints rather than a step-by-step click-map. It’s a move from deterministic programming to probabilistic reasoning, which requires a fundamental change in how developers document and test their workflows.
Pairing reasoning “Skills” with “Computer Use” allows agents to navigate ERPs and handle exceptions independently. What is the step-by-step process for encoding a company’s internal policies into these digital playbooks, and how can you ensure the AI doesn’t misinterpret a subtle or complex business rule?
The process starts by distilling your standard operating procedures into what I call “Skills”—reusable digital playbooks that contain the templates, examples, and logic of a specific business function. First, you gather your best human performers and document their decision-making criteria for tasks like contract review or invoice triaging. Second, you translate these into structured prompts and constraints that an agent can reference. To prevent misinterpretation, you must use a “chain of thought” approach where the AI is required to explain its reasoning in clear language before it commits an action in the ERP. By reviewing these explanations during a pilot phase, you can tune the guardrails until the agent’s judgment consistently aligns with your 10-year veteran employees.
This evolution expands automation from repetitive clicks to complex knowledge work, such as drafting reports and triaging contracts. What new governance frameworks are necessary to manage these autonomous decisions, and how does the required skill set for the average office worker change as a result?
We are moving away from managing “bots” toward managing “outcomes,” which requires a governance framework focused on intent and verification rather than just uptime. Organizations need to implement an “audit-by-design” layer where every autonomous decision—such as how a GL account was selected—is captured and searchable. For the average office worker, the role shifts from being a data entry operator to being an “Agent Orchestrator.” Instead of spending 6 hours a day on swivel-chair tasks, they will spend their time defining the goals for their digital assistants and handling the 5% of edge cases that require true human empathy or high-stakes intuition. The new essential skill is the ability to specify and govern outcomes rather than performing the manual labor to reach them.
Modern agents can now interpret invoices and cross-reference them with purchase orders before executing transactions in existing software. What are the risks of moving toward this “straight-through processing” model, and how can teams build enough trust to let AI operate inside sensitive financial systems?
The primary risk is “silent failure,” where an agent might process a transaction that looks correct on the surface but violates a subtle financial nuance. To build trust, you cannot simply flip a switch; you must implement a “Skills-based” verification system where one AI performs the task and a second, independent Skill checks the work against a different set of constraints. We saw in the early days of RPA that cycle times were reduced significantly, but the bots were blind; today, we gain trust by using “Computer Use” to let the AI actually “see” the screen and cross-check data across multiple tabs, just as a human auditor would. Trust is earned through transparency—by having the agent log its “thought process” for every invoice and purchase order it reconciles, providing a clear trail for any human supervisor to review.
Transitioning to agentic workflows often requires unwinding significant past investments in traditional automation. How should the C-suite approach the financial and operational trade-offs of replacing an established RPA estate, and what does a successful organizational redesign look like in this new era?
The C-suite must recognize that the economic logic has shifted; the cost of maintaining hundreds of brittle, breaking scripts often outweighs the one-time cost of replacing them with adaptable agents. A successful redesign doesn’t happen in a vacuum—it requires moving automation out of a siloed “Center of Excellence” and into the core business units. Leaders should identify the “high-maintenance” 20% of their RPA estate that causes 80% of the headaches and use those as the initial pilot for agentic workflows. This isn’t just a technical upgrade; it’s an opportunity to redesign how work flows end-to-end, potentially eliminating entire layers of middle-management reconciliation that only existed because our previous systems couldn’t talk to each other.
What is your forecast for the future of robotic process automation?
I believe we are witnessing the sunset of RPA as we know it, where the “robotic” part of the name finally gives way to true “intelligence.” Over the next decade, the vast majority of traditional RPA scripts will be unwound and replaced by a smaller, more powerful fleet of goal-driven agents that possess both the “brain” to reason and the “hands” to use any software interface. We will move to a world where software is no longer a passive tool we drive, but a proactive partner that understands our business policies as well as we do. My advice for readers is to start experimenting with “Skills” and “Computer Use” capabilities now, while it’s still a competitive advantage, because soon, being able to delegate complex knowledge work to AI will simply be the baseline for staying in business.