In 2025, every analyst declared RPA dead. They were wrong — and the companies that agreed are now behind.
The 'RPA is dead' narrative that swept enterprise technology in 2023-2024 was wrong in a specific and important way. RPA didn't fail — it succeeded at exactly what it promised: automating deterministic, rule-based processes at scale. The mistake was expecting it to do something it was never designed to do. The essay argues that the companies now winning in agentic AI are the ones that kept their RPA discipline and added AI judgment on top, rather than abandoning process rigour for AI flexibility.
Sometime in late 2023, "RPA is dead" became a consensus take in enterprise technology. Analyst reports declared robotic process automation had been superseded by large language models. Vendors pivoted their messaging overnight. Conference sessions about RPA rebranded to sessions about "intelligent automation" or "AI agents." The narrative was total and it moved fast.
I watched this from inside UiPath, which builds RPA and has been building agentic automation on top of it for several years. And the consensus narrative was wrong in a way that matters.
RPA is deterministic automation. It executes rules. Given input X, produce output Y. If condition A, take action B. It doesn't learn. It doesn't adapt. It doesn't handle the case that wasn't in the specification.
These are not limitations. They are design choices. Deterministic automation is exactly what you want for processes that must produce consistent, auditable, reproducible results. Compliance-critical workflows. Financial reconciliation. Regulated document processing. For these processes, the "limitation" of doing the same thing every time is the entire point.
The organisations that adopted RPA at scale — major banks, healthcare systems, insurance companies, government agencies — didn't adopt it because they misunderstood what it was. They adopted it because they needed exactly what it delivered: reliable, auditable, scalable execution of defined processes.
The AI wave created pressure to frame everything in terms of the new capabilities. Adaptability. Natural language understanding. Handling ambiguity. These are genuinely valuable capabilities for a significant class of business problems — the ones that weren't amenable to RPA precisely because they involved judgment, context, and cases that couldn't be specified in advance.
The narrative error was concluding that because AI could handle problems RPA couldn't, RPA was therefore obsolete. This is like concluding that because spreadsheets can handle analysis that a calculator can't, calculators are therefore obsolete. They solve different problems. The existence of a more capable tool for some problems doesn't eliminate the need for the right tool for others.
The enterprises that bought the "RPA is dead" narrative in 2023 and pivoted their automation strategy toward pure AI agents discovered this the hard way. AI agents are genuinely powerful for the ambiguous, judgment-intensive parts of a process. They are considerably less reliable than a deterministic rule engine for the parts of a process that must be executed consistently, must produce an audit trail, and must work every time without exception.
What the leading enterprises are building in 2025 is not a replacement for RPA. It's an orchestration layer that decides which parts of a process should be handled by deterministic automation and which should be handled by AI agents — and manages the handoffs between them.
A loan processing workflow might use AI to extract and classify information from unstructured documents, deterministic rules to apply regulatory requirements to the classified data, an AI agent to handle the edge cases that fall outside the standard rules, and deterministic automation to execute the final approval steps that must be auditable. The orchestration layer knows which mode to apply and when.
This is not a hybrid compromise between old and new technology. It's a more sophisticated architecture than either pure RPA or pure AI agents alone. The enterprises that are furthest ahead in deploying production-grade AI automation are almost universally the ones that never stopped valuing what RPA does well. They added AI judgment to an existing foundation of process discipline, rather than replacing the foundation.
"The companies that are winning in agentic AI kept their RPA discipline and added intelligence on top. The ones that abandoned process rigour for AI flexibility are the ones still stuck in pilot."
The "RPA is dead" take was intellectually lazy in a specific way: it confused the capability envelope of a technology with its utility. RPA's envelope is narrow and intentionally so. Within that envelope, it remains the right tool. Outside that envelope, it was never the right tool, and AI agents are filling that space correctly.
The interesting question for 2025 is not "RPA or AI?" It's "how do you build the orchestration layer that makes them work together?" That's the question the leading organisations are answering. The ones still debating which one to choose are asking the wrong question.
Kuber Sharma leads platform product marketing at UiPath. He writes Positioned, a newsletter on AI-era product marketing strategy for enterprise PMMs.