Hey team,
I’ve just been testing the AI integration agent to create an integration with our inventory management software qoblex. The purpose of this was to compare the capabilities of the agent with my experience of using this guide and hosting it externally:
Fibery, The. on Fibery Creating integrations using AI
Though it took me a couple of hours to get up and running, I found the guide in conjuction with codex managed to get a very functional integration hosted by a hetzner machine. My experience of that was positive and I would be happy to repeat that in the future.
I was very excited to try the integration agent and spent some time with it today. The main 2 things I was excited about are:
-
You host it!
-
Your agent gets up and running very quickly compared to the externally hosted procedure
While working with the agent, I found:
-
It very quickly established a codebase which was close to what it needed to be (but often not close enough)
-
Often the agent would replace working code with non-working code
-
It was quite difficult to troubleshoot issues as there was no user visible logs
-
As a result of the above, it was quite easy to chew through my AI allowance without making any progress
Things that I found improved the process:
-
Exporting all code files to an external large language model and providing it context
-
Version controlling the code externally and manually exporting files to see what was changed by the agent
Something that would make life a lot easier at my end would be the ability to export all of the connector files with a single click. That way I could look at the changes externally and work with Codex to solve problems I’m not solving easily with the integration agent.