Agentic Workflows: Design and Implementation

While ChatGPT was disruptive, 'autonomous AI agents' promise to be potentially even more revolutionary.

Essentially, autonomous AI agents allow you to instruct AI to perform tasks at a higher level of abstraction. For example, instead of providing detailed instructions, you can simply ask the AI to 'book a holiday to Greece' - the AI would then create and prioritise subtasks, and semi-autonomously execute them to achieve your goal.

This brand new course Agentic Workflows: Design and Implementation covers the development of systems concerned with autonomous AI agents (agentic workflows).

Autonomous AI agents are large language models (LLM) based systems that are distinguished by ‘agency’. These systems can act semi-autonomously to interact with other systems, make decisions and perform a complex goal. In contrast, traditional systems are concerned with a transactional interaction i.e. a one pass engagement.

Components of agentic workflows include (as proposed by technology entrepeneur Andrew Ng):

• Reflection: the ability to examine and improve its own work

• Tool use: actuate tasks by invoking tools 

• Planning and reasoning: develop and execute multi-step plans to achieve the goal (problem solving abilities)

• Multi-agent collaboration: the ability for multiple AI agents to work together to communicate and coordinate to solve a larger task