💡 Experimental action: Ask ChatGPT

21

TL;DR: on reflection, none.

Thanks for the question. I've given it some time to reflect on it. As I went through use cases that interspersed AI capability with easymorph capability, I realised that in most instances I was thinking of adding specific easymorph flows to the AI's toolbox, not generic easymorph functionality. That's not something your can provide. We'd need to tell the AI what each flow does (and while we're at it, we can tell it how to use the CLI to run it).

There's one instance in which I was thinking of having an AI agent actually using easymorph: we get heterogeneous data from our clients and need it to be transformed into a standard format. To determine which columns are which, whether some rows need to be excluded, etc. we use a combination of SQL and easymorph. An AI agent may well be able to do 90% of that work, and I was naturally thinking of giving it easymorph as a tool because I find it so helpful. It's a great hammer.

But not for this nail. An AI agent would need to be taught how to use easymorph. It already understands Python pandas and can use it to draw conclusions. LLMs aren't great at data reasoning, but giving it a new, unknown (to it) data crunching tool is not going to solve that.

22

EasyMorph workflows can already be invoked by AI agents via the command line. The command line interface (CLI) to run workflows is not available in EasyMorph Desktop, however it's available in EasyMorph Hub (aka Server) and Command-Line Worker.

In case of EasyMorph Hub, you can trigger pre-configured tasks from the cross-platform CLI (called ems-cmd).

In terms of discovery, you can do it semi-automatically:

  • Put an LLM-friendly task description in the task note
  • Use the "EasyMorph Server command" action to list tasks and export the result into a text file available to the AI agent

In case of CLW, you would need to keep a similar file manually, or, alternatively, put the LLM-friendly workflow description in the project annotation, and then generate a text file with all workflows and all corresponding annotations using the "Project metadata" action. It also helps that workflow parameters can be annotated as well.

23

Yeah exactly! Thanks for fleshing that out: easymorph can be used by AI bots, but it's our job as workflow creators to make sure the bots understand what a flow is for.