Here’s the full context:
What’s happening:
A previously stable Claygent prompt now fails 100% of the time
It either hangs indefinitely at “molding/sculpting/calibrating”
Or returns fallback errors like: “Unable to retrieve current employment information from LinkedIn”
This started suddenly — the prompt worked fine for 7 months before this
Critical context:
The exact same LinkedIn URLs succeed in other Claygent prompts
All prompts are in the same Airtable table, using the same API key and GPT-4o Mini model
This is isolated to one prompt only — every other prompt with the same setup is working
What we’ve already done (repeatedly):
Fully rebuilt the prompt from scratch (not duplicated)
Removed all conditional run logic
Switched output format from Fields → JSON Schema
Simplified the prompt structure and fallback logic
Cleaned and standardized all output field names
Tested with known-good profiles that work elsewhere
Tested from multiple devices, users, IPs, and regions
Why this needs escalation:
We’ve been told we’d get an update — we never received one
This issue has now gone unresolved for 2+ weeks
A client has already paid for 2 weeks of GTM services we couldn’t deliver because of this
Claygent appears to be glitching and re-running the prompt on its own, which has doubled our API costs (~$500/week)
Our ask:
Escalate this to engineering immediately
Inspect backend execution logs for the failing prompt
Confirm where the job is failing (dispatch, scrape, parse, etc.)
Reset or recompile the prompt object if needed
We can share prompt IDs, test data, and screenshots privately — but this can’t continue. We’re completely blocked.
cc Sayudh M. John D.
