Automation Reinvented by Agentic AI: An Evolution, Not a Revolution

By
Sarah Legendre Bilodeau, 14 octobre 2025
IA agentique
Automatisation intelligente
Création de valeur

Let me set the tone right away: I might surprise a few people!
I’m going to talk about agentic AI — but not as a revolution. To me, it’s the result of an evolution that started years ago. Here’s why.

Automation has long been embedded in the business world. For years, we’ve used tools to simplify repetitive processes, such as RPA (Robotic Process Automation), which follows predefined rules to perform tasks. These systems are reliable, but also very rigid. If the process changes — even slightly — the bot must be reconfigured, which takes time and increases costs.

Today, agentic AI is emerging as the next step in this ongoing evolution.


What Is Agentic AI?

Imagine an assistant capable of carrying out complex tasks autonomously and making decisions without supervision. That’s the core idea behind agentic AI.

These agents don’t just follow rules — they interact in natural language, access databases, and adapt their actions based on context, much like a human collaborator would.

A concrete example: in the insurance sector, an agent can analyze a customer’s claim email, verify the case number, review company policies, and even process the claim if it meets the required conditions — all from start to finish, without human intervention.


A Gradual Evolution, Not a Sudden Revolution

Agentic AI didn’t come out of nowhere. It’s the culmination of several waves of innovation:

  • RPA, which paved the way by automating rule-based tasks.
  • Machine learning, which introduced prediction and anomaly detection.
  • Generative AI, which allowed systems to communicate and reason in natural language.

Agentic AI combines these approaches to create intelligent, flexible assistants capable of orchestrating end-to-end business processes.

A concrete example: Scotiabank. In its article “Agentic AI: the next big innovation in artificial intelligence”, the bank details how it gradually transitioned from traditional AI solutions to agentic AI.

  (L’IA agentique: la «prochaine grande innovation» en intelligence artificielle | perspectives).


The key steps in the journey:

  • Experimental phase (2018–2022): internal development of AIDox, a document analysis tool, first used for simple tasks (e.g., confirming insurance letters).
  • Industrialization phase (2023): deployment of AIDox in Commercial Banking, enhanced with business-specific data to improve accuracy and relevance.
  • Agentic phase (2025): integration of agentic AI capabilities to automatically process customer emails, route cases to the right teams, and create entries in internal systems.
    Results: processing times reduced from hours to minutes, 90% of 1,500 daily emails handled automatically, and 70% of staff redeployed to higher-value tasks.


From Convenience to Value Creation

In a previous article, I explained the difference between convenience AI and value AI. This distinction is key to understanding agentic AI.

  • Convenience AI: tools like Gemini, ChatGPT, or Copilot that make daily work easier (e.g., text generation, suggestions, drafting assistance).
  • Value AI: complete solutions that transform processes and create real competitive advantage (for example, an agent managing an entire business process end-to-end).

That’s where agentic AI truly shines: it shouldn’t be seen as a fancy tech gadget, but as a strategic lever to reinvent how companies create value.

Of course, this power comes with new challenges. Unlike RPA, which is deterministic, agentic AI operates on probabilistic models — meaning it can make mistakes. That’s why implementing guardrails and human-in-the-loop supervision is critical.


Agentic AI: A Progressive Integration

Deploying agentic AI within an organization is a gradual process — especially for companies that aren’t “digital natives.” You can’t jump overnight from simple automation to a complex multi-agent system.

Historically, generative AI first focused on creating content (text, images, etc.). Agentic AI represents the next stage: enabling language models to interact with tools and perform tangible actions. This incremental approach helps organizations adapt and manage the risks associated with the technology.

Agentic AI is more flexible than RPA — but also more complex. It can make errors, which makes robust control mechanisms essential. However, it brings a unique opportunity: reusability. Once developed, specialized agents can be adapted and shared across teams and functions, accelerating automation at scale.


In Summary

Agentic AI shouldn’t be seen as a radical break, but as the logical continuation of a journey already underway — from RPA to predictive models, then to generative AI.

It marks the shift from automation of convenience to automation of value, capable of deeply transforming how organizations operate and deliver impact.