Discussion: Automator Basic mode limits field types

I'm not going to say its on purpose or some great idea behind it, the history of it (and the AI Interpolator that became Automators) is roughly:

We needed to create a whitelable solution for social media teams to be able to see in realtime what people are saying about different brands for crisis communications. Different brands requires different data, so it should be dynamical on all types of services/fields.

• We set it up in ECA - the social media team tells us they don't understand and will not use it.
• We set it up in Fields API with tokens - the social media teams understands fields, but gets frustrated by how tokens works.
• Ok fine - we create a "editorialized" token system of what you most likely need - the social media team is happy.

Then we realized that we needed multiple tokens anyway, so advanced mode was created, but only the developers/tech people set that up.

I think the same could however be solved via custom tokens and some kind of token browser that shows the "important" tokens and hides the rest behind some accordion or something.

The image example for vision LLM is actually working, that is its own drop down system currently, so tokens or not shoudn't really matter there - but other field types as you mention would not work.

You can't select a media field as the base field in basic mode: working on a Drupal CMS set up with the "Blog" recipe, the only things I can use as context for a text field are other text fields.

This is on purpose though, to make it as easy to use as possible and only allow one prepared context to move to another. This means that for isntance when the Automator Type is a taxonomy, you have an easy placeholder for getting the list of taxonomies with id and label in a format that AI can read easily.

Sidenote - you can't feed an image as a main text context even with tokens currently, unless you want to give the filename as context in the prompt.

That has to happen in the related image fields, since that is a side payload, next to the text prompt.

Now you can't still load Media, but this issue tries to solve this: https://www.drupal.org/project/ai/issues/3503018 ✨ Use image fields from media entity reference Active

But your general issue is true though, for instance a taxonomy term list is not possible to give to a text field using the simple mode.

I didn't say it wasn't on purpose - I just said that you couldn't do it :) If you have a text field, you can't fill it with the output of telling a chat operation type to describe the image - the best you currently can do is set the alt text on the image in an image field, unless I'm misunderstanding the current set up. This feels like something someone would want to do (the media entity for example is often set up with a separate "caption" field which is different to alt text) but currently can't.