Appian Composer - Cannot build AI Skills - Does not understand subprocess logic
I wanted to share some feedback on Appian Composer after reviewing an app that was generated with it. My manager used Composer to build out an app idea, and when I went in to verify functionality, I ran into several issues worth flagging.
1. Unnecessary API for existing data
Composer built an API to pull data that was already available through JDBC. This added complexity with no real benefit.
2. Broken PDF generation subprocess
The subprocess responsible for generating a PDF was using an incorrectly formatted template PDF, so document generation didn’t work as expected.
3. Misunderstanding of subprocess to parent data flow
Composer seemed to have an incorrect understanding of how data is passed from a subprocess back to the parent process, which caused the process model to keep failing. When I instructed it to stop passing the data back from the subprocess and instead add a script task in the parent to fetch the document ID, it was not able to implement that change.
4. Activity chaining removed on every edit
Each time Composer edited the process model, it removed activity chaining. As a result, users could not complete the process in one continuous session. I had to go into the process model and re-enable activity chaining manually.
5. Outdated functions and version hallucinations
Composer doesn’t appear to have an accurate understanding of its own platform. It frequently referenced incorrect Appian functions and versions. For example, it used the 20.4 version of a!queryRecordType instead of the current version for our 26.9 environment.
Summary
Composer was useful for getting an idea into a rough working shape quickly, but the output needed significant developer review and rework before it was functional. The biggest concerns for me were the process model behaviour (subprocess data flow and activity chaining) and the use of outdated function versions.
Has anyone else run into similar issues? I’d be interested to hear if there are any workarounds or prompting approaches that have worked better for you.