Owen Zanzal, who happens to be one of the organizers of the Charlottesville AI Explorers meetup that I regularly attend, wrote a piece on Fast Agent, a Python-based framework that swims in the same lane as LangChain, and it strikes me as an elegant and intuitive way to architect multi-step LLM processes into legible workflows that are easy to think about.
The syntax reminds me of the Celery “canvas” decorators that
make thinking about asynchronous tasks easier. I love a good abstraction and the @fast.agent() decorator seems to be
a particularly well-conceived way to fit together the pieces that matter into something that’s reasonably easy to
follow.
Beyond being a handy abstraction for agent interaction, Fast Agent appears to treat MCP as a foundational layer, able to both consume MCP services and expose MCP functionality for any workflow defined with it by simply adding a command-line option to the calling command.
I’m trying to branch out a little from my comfy Python world, but I’ll have to try it.