This issue of The Batch commented on AI agent workflows as driving AI progress, perhaps more so than next-gen foundation models.
The newsletter mentioned some AI agent workflow patterns, improving upon zero-shot (non-)workflows when dealing with llms:
Reflection: The LLM examines its own work to come up with ways to improve it.
Tool use: The LLM is given tools such as web search, code execution, or any other function to help it gather information, take action, or process data.
Planning: The LLM comes up with, and executes, a multistep plan to achieve a goal (for example, writing an outline for an essay, then doing online research, then writing a draft, and so on).
Multi-agent collaboration: More than one AI agent work together, splitting up tasks and discussing and debating ideas, to come up with better solutions than a single agent would.
My observation is that these AI agent patterns often originated from human agent workflows and patterns. I’m thinking of a spectrum of specificity:
Heuristics … Patterns … Protocols … Workflows … Optimizations (and more local stuff)
Also thinking about where will devs find out about all these stuff:
Regulations … Standards … Platforms … Conferences … Chats (and more private stuff)
How might protocols inform upon the creation and proliferation of AI agent workflows?
