RPA 2.0
RPA is so back, add to this graph (nodes & edges) / flow representations of processes…but…the creation of these flows are via Agentic Workflow Orchestration…
Building a chatbot or automating a business process involves painstakingly mapping out every step in a flowchart…as so many of us are acutely aware…
The graph-based approach that dominated traditional chatbot development environments and Robotic Process Automation (RPA), was all about boxes, arrows, and rigid workflows — handcrafted paths that dictated exactly how a bot should behave or a process should unfold.
Then came the AI revolution, and we were all ready to throw graphs out the window in favour of something shinier: dynamic, self-learning AI agents.
But here’s the twist — those structured flows are making a comeback, and for good reason. RPA/Graph/Flows, it seems, is so back.
The Rise and Fall and Rise of Graph
In chatbot development, developers sit down, sketch out conversation trees, and define every possible user input and bot response.
It was a bit like playing chess against yourself — anticipating moves and counter-moves.
RPA followed a similar playbook. Automating a mundane task, like data entry or invoice processing, meant creating a detailed sequence of steps for a software robot to follow.
Precision was important, and those rigid workflows delivered repeatable, predictable results.
But then AI Agents burst onto the scene, promising to liberate us from the tyranny of flows.
Why bother mapping every decision when a Large Language Model (LLM)-based framework could figure it out on the fly, right?
Chatbots started to morph into conversational engines, handling freeform inputs with ease, while RPA gave way to intelligent automation powered by machine learning.
The graph approach started to feel like a relic —
- too slow,
- too brittle,
- too human-dependent in a world where AI could adapt and improvise.
We traded structure for dynamism, and for a while, it felt like progress.
The Agent Era: Freedom with a Catch
AI Agents brought a new paradigm….
Instead of predefined paths, they relied on a Language Model as the backbone of an agentic framework, tools, and the ability to reason and act.
Need a chatbot to handle customer complaints? No need to script every scenario — just train an AI Agent.
Want to automate a complex process? Feed an AI the right inputs and watch it connect the dots.
It was liberating, and the POC results were often impressive.
But there was a catch…
As I explored in my post From Handcrafted Workflows to AI, this shift traded control for flexibility.
Without a clear map, it became harder to inspect what these agents were doing under the hood.
Were they taking the most efficient route?
Were they missing edge cases?
And when they failed, debugging was impossible without appropriate telemetry implemented. The black-box nature of AI left us marvelling at its outputs but scratching our heads over its reasoning…
And it’s not just about efficiency — it’s about creativity.
AI Agents & Real-Time Chain of Thought
Here’s where things get interesting…
As we started using and running AI Agents, we started to notice a pattern…
Their dynamic behaviour wasn’t entirely unpredictable. In fact, we could reverse-engineer their actions into…flows!
In my piece Transforming Documentation into Actionable Insights, I touched on how we’re now extracting structured knowledge from unstructured systems.
The same applies to AI Agents. By observing their chain of thought, we can map out the paths they take.
This isn’t just about understanding what’s happening — it’s about learning from AI Agents.
Take a customer service AI Agent, for instance…
By analysing its interactions, we might discover it’s great at handling refunds but stumbles on technical queries.
RPA Redux
This brings us full circle to RPA…
The graph-based approach we once abandoned is making a comeback, but with a difference.
Instead of humans painstakingly drawing every arrow, we’re letting AI Agents generate the first draft.
Think of it as RPA 2.0: a hybrid approach where dynamic AI Agents explore possibilities, and we refine their work into structured workflows.
As I wrote in The Future of AI Agents, the next frontier is about balancing autonomy with oversight.
Mapping agent flows gives us that balance — freedom to innovate, paired with the ability to inspect and optimise.
Imagine an RPA bot tasked with processing insurance claims. An AI Agent might first tackle the job freestyle, learning from data and adapting to anomalies.
We then trace its steps, refine the flow, and deploy a streamlined version that’s faster and more reliable.
It’s the best of both worlds: AI’s adaptability meets RPA’s precision.
And it’s not just about efficiency — it’s about creativity. These mapped flows spark ideas for new objectives, revealing paths we hadn’t considered.
Why RPA Is So Back
So, why is RPA staging this comeback?
Because structure never really went out of style…
It just needed a modern makeover. Graphs give us visibility in an AI-driven world, turning opaque agent behaviour into something we can study, tweak, and scale.
Sure, the days of manually plotting every step are behind us, but the spirit of RPA — systematic, repeatable automation — lives on, smarter than ever.
Chief Evangelist @ Kore.ai | I’m passionate about exploring the intersection of AI and language. From Language Models, AI Agents to Agentic Applications, Development Frameworks & Data-Centric Productivity Tools, I share insights and ideas on how these technologies are shaping the future.
