Low-AI-Code Development

  In my previous post , I showcased adhoc sub processes with my demo app "IT-Support". What I didn't mention is that this app also uses another feature of 26.1: Async processing*. The task "Search Knowledge Base" calls a LLM that returns a decision used by the exclusive gateway. With "async before" turned off, the end user would have to wait until the LLM has returned with a response. This might take longer then 3 seconds, after which the user starts to suspect the app isn't working correctly. After submitting a ticket to the IT help desk, we need to immediately return control to the user again. This is exactly what happens when "async before" is turned on. Technically, the LLM call is now handled by a database job. Turning this feature on might require you to handle a new state the application is in. Before async processing, there was "before the LLM gets called" and "after the LLM returned a response", now you have ...

A marquee feature of Flows for APEX 26.1 are adhoc sub processes*. It allows you to combine deterministic with non-deterministic process steps. Let us take a look at the following diagram that could be used by an IT support help desk to better understand this feature. This process has deterministic steps and events like creating a ticket, search the knowledgebase and notify the customer. However, to come up with a solution an engineer might request logs, run a remote session, consult a LLM or might escalate to a level 2 support engineer. The order in which they occur is not known up front and some of these activities might even occur multiple times. To handle such situations, an ad-hoc sub process is what you will need (the box in the middle containing all activities).  For choosing the next activity, there are several modes in which an ad-hoc sub process can be modeled: - manually  - with or without a recommendation from a large language model - hybrid  - AI decides the ...