We tried to use AI for custom Fibery integrations development with some success. We’ve successfully created Linear, Stripe and Monday integrations and they often work from the first try.
In the best case scenario you can get some working integration in 30 minutes (initial tools setup may take some time though). We’ve prepared repository template and prompts that you will use without any modification to get it done.
Super cool!! Love that this is getting some love. I found some bugs when sharing templates that contain custom integrations. Would make it easier to share custom made integrations together with a template to show how it can be set up.
Oh wow, this is exciting indeed! And just as I’m weighing whether to subscribe to Claude Max (yes, probably I will, and now need to get the most out of my increased limits ).
That’s great! I have other reasons to want Max, but it’s good to know the latest, most expensive models may not be needed. I suspect that may be partly because of good documentation of APIs and of Fibery’s integration approach.
It works amazingly well. I would say I’m a “technical user” but not an engineer so a full custom integration is way beyond my skill level.
My case was to bring in a Postgres db which is a core dataset for our business.
It doesn’t have documentation like an API so the initial research was a little tricky. But I did already have schema in a doc, a SQL query for the data I wanted, and a decent understanding of the data so I knew what it needed to look like.
Overall it took me probably 3 hours to get it to work. A lot longer in the initial research phase to make sure it covered everything (to make up for lack of proper API documentation) and a whole bunch of time to install and work out the other stuff like pnpm, ngrok, and vercel. So basically most of the time was me learning how to do stuff not actually prompting/building.
For an engineer who knows that stuff it would honestly be as easy as quick as the demo video.
The first “1-shot” build actually mostly worked. I didn’t do any testing so its expected. A few minor tweaks and I got it to work.
I’m using Claude Max 5x for what its worth…honestly can’t live without it. Pro actually does go a very long way. But I also use it for a lot of non-coding tasks and I’m paranoid about running out while I’m “in the flow” so its a necessary business expense
Some tips
I updated the research prompt slightly so it would output an .md file. Then I can freely edit and reprompt for follow up edits. Then when I’m happy with it, kick of the build by referencing the .md file.
I put the prompts behind a slash command i.e. “/research-api” and “/build-custom-integration” so I felt like a pro
Follow up question, if its that easy for me to do it, why doesn’t Fibery team add a whole bunch of native integrations? Its a major USP for Fibery to have all data in one place but its still very difficult for me to get data in.
).
