Class Overview
What will we do?
This is the first of two classes aimed at covering more advanced material in intelligent automation and RPA space. The goal of these two classes is to assist in identifying how to find solutions to more complex automation problems, and to cover some of the more advanced tools and techniques that can be used in these cases. We will also cover some of the governance and security issues around RPA and agentic workflows, which will be important for the final project.
Materials and Preparation
Class Materials
- Link: A background reading on the progress made and challenges faced on the use of agentic workflows in companies.
- Link: EYARC Experience Platform Ethics and Governance materials log in materials.
- Link: Cybersecurity materials relating to Lakera's Gandalf Challenge. Can you hack the LLM?
- Slides: PowerPoint or PDF
-
Suggested Pre-Class Preparation
- You should have completed the background reading before class, and be prepared to discuss the challenges companies are facing with agentic implementations as covered in the MIT study.
- You should also have signed up on the EYARC Experience platform and started the ethics and governance materials.
-
Class Plan
- We will start class with a check-in regarding the current state of agentic automation in companies based on the MIT study.
- The topic for today's class will include discussion of the use cases for multiple agents within a workflow. We will consider how this allows us to build more complex automation solutions.
- A starter workflow will be set-up using openrouter.ai to connect to LLMs, make sure you can access your api key and/or can create a new one for this workflow.
- We will start to discuss issues around context management and prompting for agentic workflows to solve the issues relating to memory of the workflow and the design of prompts to provide the agent with the right information to complete its task.