The impact of AutoGen on Robotic Process Automation (RPA) workflows

All answers to this question.

We are seeing a massive shift from "Deterministic RPA" to "Agentic RPA." Traditional bots break the moment a button moves three pixels to the left. An AutoGen-based system uses vision models and semantic understanding to "understand" the page. If a layout changes, the agent can reason through it and find the correct field. This reduces maintenance costs by nearly 70% in some of our pilot programs. The future isn't just a bot that mimics a human's clicks; it’s a team of agents that understands the goal of the process and adapts to achieve it.

   Answered 2025-06-14 by Kimberly Foster

Do you find that these "agentic" bots are slower than traditional RPA because of the LLM inference time?

   Answered 2025-08-01 by Charles Reed

• Charles, they are definitely slower per step because of the token generation. However, they are "faster" in terms of end-to-end reliability. A traditional bot might run in 10 seconds but fail 20% of the time, requiring a human to fix it for 10 minutes. An agent might take 60 seconds but complete the task successfully 99% of the time. For enterprise scale, that reliability is much more valuable than raw execution speed.

     Commented 2025-08-10 by Gregory Adams

I think they will coexist. RPA will handle the high-volume, simple tasks, and AutoGen agents will handle the exceptions and complex decision-making.

   Answered 2025-09-05 by Pamela King

• That hybrid approach is likely what we'll see, Pamela. Gary, start thinking of AutoGen as the "brain" that tells the RPA "hands" what to do when things get complicated.

     Commented 2025-09-12 by Gary White