Observations

Robotic Process Automation (RPA)

Robotic process automation (RPA) refers to the automation of rule-based, well-defined, and structured business processes. RPA leverages software robots to mimic the actions a human user would take to complete a task, such as data entry, form filling, or triggering system actions. These bots operate based on predefined rules and scripts, making them highly efficient at repetitive, high-volume tasks.

The key advantages of RPA are its ease of deployment and ability to integrate with existing IT systems without requiring complex integration projects. RPA has seen widespread adoption across industries, with use cases ranging from automating administrative tasks to accelerating digital transformation initiatives3.

Intelligent Automation

In between RPA and the emerging paradigm of AI agents lies a category often referred to as intelligent automation. Intelligent automation blends techniques from RPA, API automation, and document processing through OCR to enable a higher degree of process automation. Intelligent automation solutions often mix low-code/no-code capabilities with some coding elements, requiring specialist skills to deploy effectively.

Intelligent automation signals a move beyond point-and-click automation towards more intelligent process automation. However, at its core, it still relies on rule-based, structured processes – it is an evolution of RPA, not a fundamentally new approach.

AI Agents

AI Agent Architecture

AI agents represent a radical departure from the RPA and intelligent automation paradigms. Powered by advancements in LLMs, AI agents can natively understand unstructured data and unstructured processes through their flexible reasoning capabilities. This allows them to engage in tasks that cannot be captured through simple rules, but are described in natural language runbooks and working guides.

Unlike RPA bots that rigidly follow predefined scripts, AI agents can correct themselves if they encounter errors, and even reach out to human collaborators for feedback. This makes them more resilient and adaptable to the unpredictable nature of real-world business processes.

The key differentiators of AI agents compared to RPA and intelligent automation are:

  1. Unstructured Process Automation: AI agents can automate tasks and workflows that are not easily captured in structured, rule-based scripts. They can understand and process unstructured data like emails, documents, and natural language instructions.
  2. Flexible Reasoning and Autonomy: AI agents can adapt their behaviour and decision-making based on context, rather than rigidly following predefined rules. They can self-correct and collaborate with humans when necessary.
  3. Expanding the Scope of Automation: by automating unstructured processes, AI agents can take enterprise automation beyond the limitations of RPA, which is primarily suited for well-defined, repetitive tasks.

RPA

AI Agent

Automate simple tasks in a specific context.

Autonomous with human-in-the-loop.

Trained on browser recordings.

Fine-tuned on work-specific natural language data.

Structured data or OCR on PDFs.

Both structured and unstructured.

Rule-based and fragile.

Execute with human-in-the-loop and recover edge cases.

No memory.

Long-term memory.

 
RPA vs. AI Agents: Complementary Roles

RPA vs. AI Agents: Complementary Roles

It’s important to note that the emergence of AI agents does not mark the end of RPA. Both technologies have distinct strengths and use cases within the enterprise automation landscape.

If a task involves repetitive, high-volume data entry or structured system migrations, RPA remains the optimal solution. RPA bots excel at executing routine, rule-based processes with speed and precision. Trying to fit an AI agent to solve such a task would be an unnecessary and inefficient approach.

However, AI agents expand the scope of enterprise automation by allowing organisations to tackle unstructured processes that were previously beyond the capabilities of traditional RPA. These include workflows that involve ambiguous, natural language-based instructions, complex decision-making, and the need to adapt to dynamic conditions.

The most effective automation strategies will leverage both RPA and AI agents, applying each technology to the use cases it is best suited for. A common approach is to use AI agents to grow the scope of automation beyond what’s possible with standard RPA – for example, by having agents handle the preparatory and follow-up tasks around RPA-driven core processes.

A Tiered Approach to Enterprise Automation

Tier 3

Tier 2

Tier 1

RPA

(and first AI agents)

Intelligent Automation

(No code / Invisible AI)

AI Agents

(Custom fine-tuned)

E.g. Fill forms, and data entry.

E.g. summary generation, meeting notes, and drafting quick email replies.

E.g. prioritise incoming emails and route them to appropriate departments or staff members.

Simple decisions, specific context.

Standard decisions, standard context.

Custom decisions, custom context.

Simple but specific work.

Standard work.

Business domain work.

A Tiered Approach to Enterprise Automation

We can visualise a 3 tiered approach to the different categories of enterprise automation use cases.

In tier 3, we have a large volume of tactical work that does not require complex decision-making but operates on highly specific, enterprise-unique data, documents, and systems. This is the domain of RPA, where software bots can be deployed to automate repetitive, rule-based tasks. First, AI agents can be deployed for such tasks and by iteratively increasing the scope of automation they can deliver beyond an RPA’s capabilities.

Moving up in tier 2, we have work that involves standard decisions operating in a standard context. This represents processes often well-captured by commercial platforms and systems of record like ServiceNow or Salesforce. Here, organisations will benefit most from buying AI and automation solutions directly from these vendors, who are best positioned to innovate on the data and processes they own.

Finally, at the top tier 1, we have the most strategic category of work that is central to how the enterprise operates. This work involves complex decisions operating on a custom context unique to the organisation. This is where built-for-purpose, enterprise AI agents will have the biggest impact, as they can understand, reason about, and automate these mission-critical workflows.

Next Steps

As an industry, we are still in the early stages of the AI agent journey. While the technology holds immense promise, it will take time to reach the full potential of automating the most valuable, strategic work. But now is the right time for senior technical leaders to start building the skills, platforms, and governance frameworks needed to unlock the power of AI agents.

  1. Establish robust governance and security frameworks to manage the risks.
  2. Define a roadmap and timeline for AI agent adoption, balancing short-term tactical wins with long-term strategic impact.
  3. Start with tactical use cases that expand the scope of existing RPA initiatives, such as automating the preparation and follow-up tasks around core RPA-driven processes.

Footnotes

  1. Embedding Robotic Process Automation into Your Low-Code Strategy’, IBRS, 2022.
  2. Trends for 2021–2026: No New Normal and Preparing for the Fourth-Wave of ICT’, IBRS, 2021.
  3. How Oncquest Laboratories Improved its Operations With Cloud’, IBRS, 2024.