I wonder, if those of us who aren’t coders could create Pieces using a “no code” interface, if your contributions would explode. Looking in your Roadmap I see there are over 30 Pieces waiting for development. Perhaps using such an interface could automatically contribute your Piece to the Repository.
I went through your developers documentation to see how difficult it might be to create my own Piece. I suspect an experienced coder would find it trivial but for my limited skills each click to the next page convinced me this is not something I can take on.
I’ve looked at a number of different automation/connection services and none of them have addressed the difficulty of creating connectors/Pieces. Would this not be an excellent use of AI?
Zapier is the only one I could find that includes the two Pieces I want (Everwebinar to Mailjet) but it can get expensive in a hurry.
Just my 2 cents as I wander the Internet looking for solutions.
It’s an interesting idea, but very difficult to implement well. The amount of resources that’d have to be invested in this might be better spent on writing more integrations by hand (or polishing, shameless plug).
Someone like @kishanprmr has worked on a lot more integrations than me though, I’m curious what their thoughts are. Maybe there are more repetitive easily abstract-able patterns than I think.
Sort of, but depending on the integration it can be trickier than even we expect. See for example Google Sheet Really Crazy Bug and Flawled Google Sheet integration.
Stitching together a couple API requests is fairly easy, but integrating them in a way that works well for non-technical users in a no-code/no-supervision setting is a lot more involved. Especially something like error handling can be a ginormous pain in the behind.
Secondly, even in a no-code setting the integrator needs to understand the API of their service. That involves reading technical documentation intended for programmers. The venn diagram intersection of people who can do that but not code it by hand seems very slim to me.
Standardized documentation like OpenAPI could help scaffold the basics, but this only describes structure, not semantics.
The current generation of AI is not good at dealing with very niche and nuanced contexts, such as the programming interfaces of Activepieces and services that don’t already have an integration. Even if it produces something working-ish, they’re notoriously prone to subtle yet serious security and/or reliability problems.
Having said all that, I really admire your initiative in exploring this process! Even if this wasn’t for you, that willingness to take action and put in the effort is something I sometimes wish more people had.
A colleague of mine at SwanBlocks tailored a ChatGPT prompt to analyze OpenAPI specifications and API documentation websites, enabling the generation of piece code with minimal error for seamless integration using standard input types and webhooks, you can find it here:
