This topic was automatically generated from Slack. You can find the original thread here.
I am really struggling at the moment. I saw a YT video showing how great it is to create a workflow that send a daily slack message with posthog data. I have a background in AI Engineering and generalized web app dev. I provided the perfect prompt and obtained the needed data for integration however the agent kept getting everything wrong. I spent 6.6 million tokens so far and only finally got it to extract the data. Though it is still giving objectObject for date range. At this point I don’t know if I should continue trying to get it right and risk it breaking yet again or what. Does the company ever return credits to the user due to the agent messing everything up?
I am now at 17 million tokens used and it is still not working. I am seeing that it will continue to revise the working code when trying to fix the broken areas causing a loop of non stop issues.
I have even been using Posthogs’ internal AI “Max” as well as GPT5 thinking to try and overcome this. I do wish we could edit the Pipedream AI code blocks to manually fix small issues instead of wasting tons of tokens and errors.
Could you share your Pipedream string chat url here so we can take a closer look? (no worries only Pipedream internal team can access it for debugging purpose)
In general, my way of working with LLMs is treating them as an assistants instead of an independent workers, meaning we still need to do the majority of the work with LLMs help on researching/synthesising data. LLMs currently do not smart enough to do everything on their own.
Hence, my recommendation is to have Pipedream String to scaffold your workflow, then manually test and change it on Pipedream Builder UI, with the help of Pipedream Pi assistance (available on Pipedream builder UI as well)
I ended up using almost all my credits before I discovered the pipedream editor. I then used Chatgpt & Gemini as well as Max AI within Posthog to help fix everything. I am happy to share the string however I would prefer to do it in a dm if that is ok with you.