I think the sentiment in this post is shared with most folks: We don't need yet another workflow tool. That said, I don't think no and low (or even all) code are mutually exclusive. Today these tools definitely speak to an audience that's technical enough to know how to get around in these tools but not so technical that they know how to deploy, monitor and maintain these workflows. What will happen with time is that any complexity that does exist in these low-code tools will be abstracted away with AI but you can dig into deeper layer if you are technical. This might be that very last point you're making?

> How can we make code generation models better at writing LLM powered workflows/agents

For context, the approach we've taken at Fabi is that you can manually build the workflows or you can ask the AI to entirely build it out for you and you can pick up where the AI left off at any point and edit the code yourself -- we're not a generic workflow solution like n8n or AgentKit, we're 100% focused on data analysis, but the same should be easily applicable to these solutions.