I love the series. Is it really an either/or on Workflows versus Agents? (The execs I talk to all speak in terms of workflows, even when thinking through how to adopt agents)
Greate content, thanks a lot for sharing. I have one side comment, afaik what you refer as workflow is what I it is known as a static workflows (a workflow whose steps are hardcoded). This comment comes from my experience with workflow orchestrators, because an AI agent can be underestood as a dynamic workflow where LLMs are called on tasks and the tasks to be executed are defined on runtime based on the output of LLMs tasks, for example, this video explains how to implement an agent loop with temporal (https://youtu.be/s58zhw9XpEQ?si=cwYL8OwHThgPtun9).
I may be wrong, but any setup of AI agents can be reproduced with a workflow orchestration engine if it supports the following features:
It's a simple yet informative article. Thanks.
Excited to hear that! More incoming
Thanks Paul
Here to help ✌️
Simple and well explained content. really like this Series and looking forward to read more.
Excited to hear that, man! This week I just wrapped up all the lessons: from workflows to multimodal AI Agents
I love the series. Is it really an either/or on Workflows versus Agents? (The execs I talk to all speak in terms of workflows, even when thinking through how to adopt agents)
I write about it some more here: https://open.substack.com/pub/therosen/p/the-importance-of-workflows-in-ai
Thanks man! No, I wouldn’t say it’s an either/or but a slider between the two which controls autonomy
Greate content, thanks a lot for sharing. I have one side comment, afaik what you refer as workflow is what I it is known as a static workflows (a workflow whose steps are hardcoded). This comment comes from my experience with workflow orchestrators, because an AI agent can be underestood as a dynamic workflow where LLMs are called on tasks and the tasks to be executed are defined on runtime based on the output of LLMs tasks, for example, this video explains how to implement an agent loop with temporal (https://youtu.be/s58zhw9XpEQ?si=cwYL8OwHThgPtun9).
I may be wrong, but any setup of AI agents can be reproduced with a workflow orchestration engine if it supports the following features:
- workflows are stateful processes
- workflows can interact with other workflows
- external services can interact with workflows
The line between them is very blurry. That's why I concluded the article with the idea of the autonomy slider, and not what a workflow or an agent is.
For example, based on what you said, both workflows and agents are:
- stateful
- interact with each other
- with external services
In future articles from this series, I will dig more into what you said above.
I am looking forward to those articles, really good content :).
Thanks for the elaborate walkthrough for the good 😊
Always 🙏
Informative article! But first few paragraphs repeated twice haha
Thanks! Substack duplicated them when adding the images. Fixed
Very helpful , Thanks .
I agree, Thanks for the article.
Glad you liked it!